We bought a 2TB Sandisk SSD back in 2024 for around $95 from best buy. Today 1TB ssd by sandisk is $166(the cheapest that i found, and it goes for $199 in walmart). Market is forcing people to become renters than buyers and there is no force countering that idea. Its a market failure that people will study 10-15 years from now.
It's not a market failure, it's just supply and demand. There are many computer components competing for the same resources (fabs, wafers). Demand for GPUs, RAM etc. has increased a lot due to AI, but supply is still the same due to new fabs being huge investments that take years to build. Of course the price goes up.
> There are many computer components competing for the same resources (fabs, wafers).
That's a bit off the mark. Processors, RAM, and flash memory each require their own specialized fabs. TSMC makes processors but not DRAM or NAND flash. Kioxia makes flash only. Samsung has all three types of fabs. Micron does DRAM and NAND, in different fabs.
The increase in SSD prices is not because there are many components competing for the same constrained resources, it's that they are complementary goods being subject to the same dynamics in parallel because the servers that are causing this demand spike need all of those components. Where we do see competition for the same fab capacity is in the mix of DRAM types, where GPUs want HBM, server CPUs want DDR5, laptops and phones (and the occasional server CPU) want LPDDR. And competition among different types of processors for TSMC's fab capacity.
I mostly agree, but if you go further back in the supply chain there are a number of common inputs/tools. For example, as 2025 the new LPDDR lines are using EUV systems, which means for new fab production both DRAM and logic producers are competing for machines from ASML.
It doesn't change your point that these lines are different and the immediate price spike is not about them competing for capacity in the same facilities, but the fact that manufacturers are committing to large enough future purchases to drive new fab construction means that future pricing outlook (which does impact the current prices to extent) does involve some amount of competition between different types of semiconductor products.
An industry doesn't need to increase supply when demand increases, they can absorb the demand as profits. More so in an industry that is hard to enter into. Consumers hate this and call it all sorts of things like market failure, gouging, etc... In this case, the suppliers are slow walking increases in production and enjoying a run up in profits. They raise concerns about oversupply, as if this deep learning thing were some passing fad and the market will have no use for the new supply if it passes. It's a dubious notion though, the demand is here to stay and new supply needs to come online to meet demand. We simply need more silicon in the market. High prices should eventually bring new supply online, but I'm a little disappointed by the rate of the ramp up.
>It's a dubious notion though, the demand is here to stay and new supply needs to come online to meet demand.
This is a big bet. Look at what happened in 2001 with the dot-com boom. We're still trading on their dark fiber over-build today. Meanwhile any overcapacity built in fabs will quickly be made obsolete by newer and better fab technology (or at least, that's been the pattern for the past 30 years).
I think you're missing the fact that building new supply takes time and sustained commitment, and there's simply nobody in a good position to make that commitment without losing big if your thesis turns out to be wrong.
If the demand for new AI builds is eventually satisfied, or worse, craters overnight, then who will be left holding the bag? It sure won't be Google or Apple, or even NVIDIA - it will be TSMC and Samsung.
But as you mention, this is all temporary. Either you're right, and demand will remain sustained long enough for some of the providers to decide to take that risk, or demand will crater and prices will fall.
The fact that the S&P 500 is near record highs at the same time as consumer confidence is at a 70 year low is not encouraging for continued all steam ahead in my mind... but then it's easy to predict a general future recession, and much harder to predict it to the day.
The idea that we're still trading dark fiber from 2001 is an old narrative right? I guess it is still floating around. But, we're in a second big fiber build out for not only residential, but also to connect all these new data centers. Could there be some demand oscillation for silicon? Sure, but overall deep learning is not some passing fad even if someone gets too far out over their skis and over buys. So far Big Tech has increased spend 3 years in a row and next year they'll spend more than this year. Demand has evidently not let up for AI services, we still need more silicon.
I doubt consumers will regret these compute purchases in 3 years, my 3 year old GPU is still holding value, actually increased in value, and I use it more now than ever.
I did predict or expect that if oil went to $150 we'd be in recession territory, currently hovering below that level and folks are feeling the squeeze and aren't happy. Things could get worse or better, tough to tell, but I think silicon demand is more or less secular and will do relatively well in a variety of macro conditions.
This seems like it would be easy to structurally solve - big tech could make strategic partnerships with 5 to 10 year horizons with the very fabs in question. If they promised to spend a flat or growing $X on silicon (without easy contract cancellation) over the next decade, then the risk would be entirely on the tech companies and not on the fab companies. Of course, there's always bankruptcy to worry about, but that's less of a threat for Google and Apple than it is for OpenAI or NVIDIA.
The fact that we _haven't_ seen such deals be made, and that ~50% of new datacenter builds have been quietly cancelled[1], suggests to me that we're dealing with paper demand more than true sustained demand.
The dark fiber from the dotcom era is approaching the end of its life expectancy. Contracts for leasing dark strands reflected that. Most of it likely still has useful life left that the owners will want to monetize but it might change how it is used.
Bad metaphor. Not being forward enough is a common mistake. When learning, you may want to make yourself _feel_ like you're too far forward. We can commonly think we're too far forward when actually overall we're still too far back. Teaching it is very hard.
Although it is dependent on your style of skis (e.g. carving) and style of skiing (powder, racing, cruising).
For me, it is very rare that I go too far forward (I began with a leant back style and I haven't rectified that after many years of skiing). I try to prioritise fun over ability.
> the suppliers are slow walking increases in production and enjoying a run up in profits
The major companies have been trying to build >$100B of new production capacity in the US for years now. All of these manufacturing facilities have been significantly delayed by NIMBYs using the same "environmental and community concerns" advocacy slop that hinders almost all productive industry in the US.
Blaming the suppliers is a lazy take. They aren't responsible for degrowth activists being able to dictate what we are allowed to build.
Or their continued collective inability to predict or manage supply and demand, it happens repeatedly and previous events were very minor in comparison... The other obvious reason is profiteering but pretty much impossible to prove.
They are public companies, their bottom line isn't exactly secret. The only company that is really squeezing the market in term of prices is nvidia and they aren't the production bottleneck.
When supply is constrained, widgets to get sold to whoever is willing to pay the most. It’s capitalism, baby. Arguably this is a good thing, because the limited resources are allocated to the people who can create the most value from them. And high prices provide an economic incentive to bring more supply online.
Nvidia is making big profits right now because they don’t have a lot of competition for their fancy AI chips. The prices will come down when they have more competition.
That only works if the market is not controlled by basically 3 vendors and if the market isn't also pretty much impossible to break into. The value proposition is tenuous at best, the real value is being produced in other arenas, not this one.
We saw this during the pandemic when everyone wanted to purchase bicycles. Shimano (they make bike transmission parts and brakes) saw a huge increase in demand. Shimano didn't want to invest into new factories as they assumed the demand was a temporary spike. I'm pretty sure this is what is happening currently with RAM, SSD's and processors. New fabs are coming on line, and Apple is looking at both Intel and Samsung to bring additional capacity online. If the AI boom dramatically slows, it's going to be interesting to see how the industry responds.
There's definitely a world where we need newer products, but not as many or as fast in iteration. My gaming PC has 128GB of RAM... And I built it years ago when RAM was laughably cheap. I still never touch the sides on it.
I think I'm agreeing with you but its also not something easily dismissed. The DRAM Cartel has been found to be distorting the market on numerous occasions by various regulatory bodies. There is a boom-bust cycle that occurs with DRAM and Flash memory. The Cartel claims they always lose despite the fact that demand seems to always steadily rise.
The pandemic caused once such boom-bust that resulted in a rather large downturn in demand in 2022-2023 referred to as the pandemic hangover. During that time demand dropped following overspend during the pandemic and members of the cartel drastically cut production at times to keep prices above cost. Even after the demand recovery began in 2023, the cartel members were slow to increase production and made little to no investment in production capacity in 2024-2025. Creating a shortage.
The AI hype cycle has exacerbated the shortage by creating speculative purchases and then panic buying. Remember the shoe company that pivoted to AI?
So Cartel market manipulation is partially to blame for the over 100% increase in prices and the shortages.
It is a market failure as the markets fail to cater for the demand due to structural factors. It’s basically designing database structure that causes huge latency due to use pattern changes and you are not able to alter the structure because that will demand even more painful downtime and latency and the pattern may eventually change and make the alterations redundant.
> It is a market failure as the markets fail to cater for the demand due to structural factors.
The Market™ is perhaps going after higher margins. If you can charge $200 instead of $100 for a widget, why wouldn't you?
Ideally at some point someone will see the margins and be motivated to go after some of it and offer things for $190, but the ROI on upfront fab costs given market risks may not be high enough for the business uncertainly that needs to be taken (boom-bust cycles).
No participant in The Market is obligated to "cater" to demand if they do not think the juice is worth the squeeze.
It’s market consolidation unchecked. How many nand suppliers are there in the US? How many globally? Now add the sanctions on China, and tell us how that’s a competitive market.
The reality is all the things we learned from the Great Depression have been forgotten, and half the voting public seems to champion monopolies while simultaneously acting dumbfounded that the end result will always be an increase in costs.
The demand side is the world's "investors" buying up a product without producing profit. This results in locking out companies operating without such "investors" from accessing the product and using it to produce profit. This is a failure.
And it's exacerbated by the companies that have to invest their own Real Money to produce factories to make the chips. Same goes with the power utilities. Everyone knows it's a bubble and they don't want to be the one left holding the hundred billion dollar bag like back when Enron went belly up.
But they create a new problem for themselves of converting consumers into non-brand loyalty pure price hunters. I don't think I'm speaking for myself when I say I will use the other guy if it saves me a penny for the rest of my life.
The counter-example I hold up is Harley-Davidson. They could hardly pay a bill in the late '70 through the '80s. They'd even get free samples from suppliers and put them into production. They survived with very strong brand loyalty from customers and after-market companies.
Samsung sells directly to customers of all sizes, so they have a very good idea as to where products end up. Cutting the legs off of innumerable small businesses is despicable behavior.
I don't care if Samsung is literally starving to death, where I was previously at least a marginally loyal customer not even looking at other brands because prices were fine.
Supply, to a great extent, is a policy decision. Demand, to a great extent, a policy decision. This is not the silly metaphor of individual consumers and producers acting independently of each other.
This was avoidable. Over $100B of new production capacity in the US alone has been in the works for years. The usual "environmental and community concerns" lawsuits and activism that plagues building in the US has delayed almost every single one of these projects by multiple years.
NIMBYs and degrowth activists own this particular "market failure". The companies have been trying to invest in capacity ahead of demand but have been prevented from doing so.
I think it's really easy to blame one or another specific group for the "market failure" but the reality is that if there were 100B worth of capital looking for 1T worth of gross over 10 years there would be a lot of places that would welcome that kind of spending. I think it's probable that in specific places there's been push back, and those places are probably where the work was historically concentrated. But if that's the case, you should really be asking "why is there push back now when there wasn't before?" And I don't think the answer is because the entire community has adopted a "degrowth" agenda. Rather, I think it's going to be more about chemical dumping in the water supply or use of scarce resources that the community would rather not spend for the profit of a few people who don't live there.
I think a lot of people will accept industry when industry tries to be a good steward of the community's environment.
It is happening to production facilities across the US. Both new production facilities and expansion of existing facilities. The locale doesn't seem to make a difference.
As with the sudden inexplicable "concern" about data center construction, I don't think this activism is organic.
As with the sudden inexplicable "concern" about data center construction, I don't think this activism is organic.
I don't think it is sudden, there was much of the same concern about Amazon warehouses and anything else that cuts down trees, disrupting wildlife, and human neighbors. The reason it is more of a concern with modern datacenters is because they don't come with a significant amount of new jobs for the communities they are affecting. People will put up with a lot if means their unemployed loved ones can get on their feet.
With AI being a looming threat to everyone's livelihood it is compounding the reaction, not only do these datacenters not provide new jobs, they are actively contributing to people losing their current jobs.
If these new datacenters were going into empty buildings in cities that already had adequate infrastructure, then no one would care. They would still be against ai in general, but the datacenters wouldn't be a problem.
edit:
They could probably get a ton of support if they went into one of the many cities with crumbling infrastructure, took over abandoned buildings, and paid to fix up the roads and stuff. Like if they moved into an abandoned factory in flint, and fixed the water supply they could be heros.
You find populism in the US inexplicable in 2026? Where have you been for the past 18 years? You get instant political capital by framing anything as elites vs the common man, and the tech company billionaires cannot escape that "coastal elite" label, or hide that they want to build more data centers.
Agreed on the insignificance of the degrowth moevement on seriousness. However there is a serious NIMBY movement, and the degrowth movement provides the moral cover for a lot of very selfish people that are not serious about degrowth but wish to accomplish the same goals for individual projects near them.
You were not are renter in 2021 when NVMe were same price as now by TB,
stuff is just becoming more expensive on market shortages.
Last half year ate up three to four years of earlier price regression, that's about it.
As long as this plateaus here, as prices did for last 4 months, that's just the new equilibrium where it has the chance to get better again, would no be all doom and gloom about personal computing yet.
The dollar is losing value, we cant hit equilibrium, it will always cost more dollars. And that is assuming the mfg are not price maximizing (they are).
> Market is forcing people to become renters than buyers and there is no force countering that idea.
The doomerism surrounding this topic is wild.
Gas prices have also gone up recently, but I don't see the same claims about how it's the end of personal vehicles, that the prices are never coming down, or that we're all becoming renters instead of buyers.
Saying that there isn't a force countering this is peak doomerism. New factories take a long time to build. These companies didn't have spare production lines waiting for demand to go up.
If you don't see people talking about the "end of personal vehicles", it could just be that you haven't looked very hard.
It's intuitively obvious to a lot of people that the era of personal, wholly owned transportation is waning. A lot of people seem to miss the second clause of that old "you'll own nothing" phase, the part where most people are happy about it!
When vehicles drive themselves, and there's a large enough pool that one can show up pretty reliably within a few minutes of your needing one, how many people are going to choose to own when renting is cheaper and easier?
Because renting will likely never be cheaper. Maybe the occasional city dweller may see a profit, but those of us who spend hours in our cars each day will not.
Probably everyone that doesn't live in a dense metropolitan area, and anyone that can do the math that "Owning my vehicle for 100,000 miles costs an average of $0.50/mile compared to the $1+/mile getting a lift costs, and once it's done I still have whatever value remains in my vehicle"
> If you don't see people talking about the "end of personal vehicles", it could just be that you haven't looked very hard.
Or it could be that I mostly talk to people in the real world, and less so follow the echo chambers online that think "you'll own nothing" is a foregone conclusion about the future and fit their worldview to match.
When I upgraded my computer a little over a year ago, I got 128GB of DDR5 RAM instead of just 64GB. Glad I did.
I also thought about upgrading my gen4 SSD at the same time. It was still meeting all my needs, so I thought if I waited, I could get a bigger, faster one for cheap later. Mistake!
Are things too expensive now, or was that miraculously cheap because of a glut of supply relative to demand at the time? I think my first SSD was something like $600 for 160 gigs, one of those intel drives that came with a speed demon sticker, so let's appreciate what we have for a sec. Or we could go back to 5 megs of spinning rust for $thousands.
Things will normalize, semis are a boom and bust industry, and it takes a massive amount of capital investment to increase supply. Because of that boom and bust, the producers are wary of growing their supply to keep up in lockstep with demand, because they might end up with a glut that they have a hard time selling, and a large pile of debt from building out that glut. Semiconductors at the leading edge take very large, very expensive, very rube goldbergian machines to make them, and a very skilled workforce to make those work.
If this stuff stays this expensive/profitable, industry will grow capacity to boost earnings until they overshoot, there's not some grand conspiracy to make you rent everything. We're just in a massive dislocation right now because we've discovered this awesome new tool that's boosting demand by a ridiculous degree. Last year we were at ~$300B of capital investment in DCs in the US, this year it's looking like ~$750B.
> If this stuff stays this expensive/profitable, industry will grow capacity to boost earnings until they overshoot, there's not some grand conspiracy to make you rent everything
Production cartels exist to prevent individual greed from "ruining" it for the rest of the group. OPEC does it in the overtly, memory manufactures do it covertly, and have been convicted of conspiring in multiple jurisdictions, on multiple occasions.
It's not a goal, but a secondary effect. Companies that build long-lasting products for sale to customers will have lower profits than those that rent-out cheaply made goods. Over a multi-decade window, with some bankruptcies, consolidation roll-ups and astute leadership, you can guess which model will win, based on nothing but market forces and imperfect information. The collusion is based purely on P&L performance.
I remember buying a 80GB Intel SATA SSD (which was $450 USD at the time) and putting it into the DVD drive bay in a 2009 17" Macbook Pro. At the time there was no official SSD support, but there was an aluminum holder you could buy to remove the slimline slot loading DVD drive, and put a 2.5" form factor (7.5 or 9.5mm?) height SSD or HDD in its place. Was a huge performance increase for the OS/boot drive.
I just bought a few more Intel 310 Series 40GB mSATA ssd’s on eBay for $10 each. They are ancient and comparatively slow, but reliable and extremely low power, so much so they can be perfectly powered by a USB 2.0 port in an enclosure if needed, and don’t need a heat sink. My original 80GB version ($190) that I put in my W520 thinkpad in 2011 is still going strong.
Basically I like to use them as a cheap “better” thumb drive, putting ventoy on them with no concern for space, along with a fat32 partition for bios updates.
These along with a nanoKVM box my siblings can attach to the computers I built for them makes remote troubleshooting relatively painless when rustdesk isn’t sufficient.
When people talk about AI investments of trillions of dollars, its never mentioned how high a price we are paying at a planetary scale in externalities. Billions of people are now paying a lot more for these components, memory, ssd, cpu; all these costs must be factored into the total cost of AI. AI companies are being subsidized at the expense of consumers. The current supply chains were built for consumer demand, which is hijacked by the AI grifters.
Same a consumers (car drivers) subsidizing businesses (trucks). Road damage is almost entirely caused by trucks, 4th power of weight, but the cost is borne by us.
True, I've been searching for 2 weeks now for components to get a PC back online. I suspect the VRM blew in my motherboard, and trying to find x570 or b550 old Gen tech is drying up because that's the only thing affordable. Meanwhile decent 3D cache CPU + motherboard combos afe going cheap because they can't move stock. The price looks nice but the second you add DDR5 RAM it is just a gut punch. Scalpers are taking advantage of this as well, as well as Chinese "Brand New" x570 motherboards listed as gaming consoles on Amazon.
This is the first time I've had a computer have a complete failure and no way to fix it as I look around at the used market and the latest generation market. I'm kind of just disillusioned.
So you pay for parts that are all very expensive due to AI in order to get the machine online so that you can pay for AI subscriptions to do work that is disappearing due to job cuts due to AI.
the interesting thing is that scaling laws (at least Moore's and the like) are complete lies. (Moore is fine if you treat it as a weak observation that shrinking features to half gives you 4x as many devices in the same area.)
what happens is that the industry goes through certain discrete technical upgrades. for instance, EUV in fabs, or GMR disk heads. none of these are really planned, none of them are exponential. and they usually interact with other phenomena (such as Dennard scaling).
in a sense, the phenomenon is more like "expectations are exponential, and this motivates manufacturers to schedule updates".
hard disks are still improving, arguably similar to how they have in the past, but there are limits to demand. the consumer market has mostly dropped out, for instance due to flash.
even in flash, there is no exponential scaling in devices. people got excited in the initial startup, when for instance, mature TLC is so much better than early SLC. but all that's over: it's both mature and we'll probably never see PLC. even QLC is interesting in that it illustrates that most of our storage is very cold.
Meanwhile I'm still dreaming about any consumer and affordable 32TB or even 16TB portable SSD. Innovation and market for consumers are going backwards.
Funny thing is that one of the best you can get is the Crucial (Micron) 8TB one but even that one gets more expensive. I have the feeling it will be gone completely soon.
There was once a 2.5" SSD Mushkin Source 16TB SATA drive. At its cheapest it was ~1700 USD (or 1500 EUR). That was mid 2023 (like 3 years ago!).
Nowadays it feels like that this time and price region is like decades away in the future. I was hoping I can store more data in future on modern tech like SSDs and not less.
Yeah it sucks :( Almost exactly a year ago, I got a brand new 15.36TB Kioxia CD-6R (u.3 pcie4x4 drive) for $1450+tax from serverpartdeals.com - that same drive is now listed for ~$4600 (and it’s also out of stock there)
Nobody is stopping anyone from buying a USB-C powered and connected very portable 2 or 4 slot external NVMe enclosures.
The old SATA SSD form factor is dead and wont come back.
OWC ThunderBlades exist, but 32TB will set you back 9 grand.
You should be able to assemble something with USB-C for under 5k.
That's not a mass consumer market thing, but perfectly doable, if your use case warrants it.
We are stuck with 2021 pricing, but now with options of 8TB per NVMe drive at way higher speed.
36TB+ HHD external WD drive combos were always around EUR 1000 over last 5 years. With a short low end around EUR 600 in 2023
>Nobody is stopping anyone from buying a USB-C powered and connected very portable 2 or 4 slot external NVMe enclosures.
That's more expensive than buying a single large capacity drive. It's also a terrible idea. I would never trust a low cost chinese controller with terabytes of my data.
>The old SATA SSD form factor is dead and wont come back.
I will not miss the awful, half-duplex protocol that never should have won over SAS. I just wish that PCIe switches and cheaper eMMC/UFS flash on M.2 were available for more flexible and cost effective storage options.
Yeah, the problem is most consumer motherboards don't come with that many PCIe lanes. Building storage servers for home use with M.2 SSDs is just not feasible.
PCIe is a switched protocol. We ought to have had better options to make better use of the limited lanes by now. There are a few cards with switches that can give you multiple M.2 slots on x4, but that's only with gen3, which the latest GPUs won't init on, and trying to cable that is hell. Better PCIe switch options for external enclosures exist, but aren't available to mortals that I've seen.
I find it very odd that there is so much faith in "innovation" (and probably "economies of scale").
there is no sign of any impending breakthroughs that would change flash economics much.
slc-mlc-tlc-qlc was very nice but plc will not happen. layer-based flash was also nice but it is ultimately linear (more layers, more cost, lower yield). dimensional shrinks are already stalled because of a tragic electron shortage (per cell).
I guess there's no harm in pining for some other NVRAM technology (spins, etc).
The prices aren't going down for large consumer drives because the market is so small, and because the AI DC market is swallowing up everything. There's little demand from your average consumer to have 30TB of storage, let alone specifically SSDs. The average user doesn't have that much data, and if they do a HDD is fine for any practical purpose.
Despite the recent AI bubble you can still buy HDDs in the tens of TBs for a few hundred EUR/USD and you still don't see them in every computer. How high could the 30TB SSD demand be to justify the kind of volumes that drive price down?
In the DC it's the opposite, large and efficient drives are a must to save support all those fancy workloads while driving down space, power, cooling needs.
A few months ago I finished building a new media server based on UnRaid. I populated it with WD 26TB drives. At the time they were about $400 (steep, but a decent capacity/dollar buy). Now they are nearly $1000 on Amazon, a 250% increase. I just hope I don't have a drive failure.
With regards to the new Micron SSD - I wonder how they keep it cool? I don't see coolant ports on it so they must strap a heatsink on.
The product brief says maximum 30W and it looks like the whole enclosure is a heatsink, even has ribs on the back. The expected operating temp is 50C but it's probably rated to operate at higher than that.
P.S. I had to shuck 20TB WD drives that cost 350EUR on sale (now at 400EUR). 26TB drives are now ~700EUR. These external drives were the cheap option. Standalone drives usually cost more.
The average user consumes that much quite regularly. They've been taught to stream it off of someone else's computer, mostly so that the next time they stream it they can be compelled to pay for it again. It's fun going back to dumb terminals.
Consuming and needing to store are very different things. Most media is disposable, one-time consumption. How many people stored the newspapers they read?
Why would you want to store every movie or series you watch? 30TB of data is something like 1 year of uninterrupted streaming at average Netflix 4K bitrates. Even more at HD bit rates. How many people would ever store years worth of movies on SSDs no less? Enough for it to drive huge sales in the market?
It isn't. The number of books I've read again, that I've wanted to have my children read. Music. Shows. The old VHS tapes were only disposable because the format was so deteriorating. It was inconvenient because of the physical size.
But, if people realize that media could be non-disposable, how could the recording industry get rich selling you the same albums you'd already bought a half dozen times before? I've been storing everything as FLAC... it'll be good for centuries. Books already have a "good for centuries" file format. I suspect very strongly that 4K is the ultimate res for all media made up until today (no way to do 8k remasters). It doesn't need to be disposable, it just is because there are people who want to get rich off of you renting your own culture back from them.
>How many people would ever store years worth of movies on SSDs no less?
Send me a boxfull of these 245TB SSDs, I'll show you. My film count surpassed Netflix's about 12 years ago, and I haven't slowed down once.
That is pretty awful write performance. Does anyone know more about this? I assume all of these hyperdense SSDs suffer from the same drawback. Also, I heard that the E3.L interface can support up to 16x lanes, but there are no practical commerical products at this point.
Consumer and data center drives play by different rules. The super high write speeds you see for consumer SSDs are usually achieved through tricks like using sections of the drive as a high-speed buffer and then using a background process to rewrite the data into the drive's high-density NAND storage during downtime. They can also use caching techniques that aren't resilient to power loss. They might allow burst performance that heats the drive up until it throttles.
This is all fine and even desirable for a consumer who will only be writing for at most a minute or two, but with a 245TB server drive you need to assume the performance will be needed constantly. They target sustainable and predictable performance.
A more convenient (and dare I say, faster) tape drive replacement for backups? They do make a good point, it would take 10*24TB drives working in the worst raid configuration to even come close to these speeds.
Note how that is still well in excess of what e.g. AWS EBS GP3 volumes offer (or at least used to, though even now their "80K IOPS" is measured with 64 KiB random transfers, whereas Micron measured that 42K IOPS with 4 KiB random transfers), which is what the person above is gesturing towards.
The same EBS GP3 used to be specified with 16K max IOPS at 16 KiB random transfers until pretty recently.
What’s the intended block size of these things? I thought 4KB was normal, but that doesn’t make sense at 40K IOPS, and doesn’t align with the benchmarks I’ve seen.
Also: price is expected to be $80k. I suppose density is the selling point here, not speed.
The u.2 form factor is slightly larger than a 2.5" drive. I can imagine the entire space in it taken by Flash chips. I can't imagine what cooling scheme do they employ for the chips in the middle.
The U.2 form factor is a 2.5" drive, not larger than it.
"U.2" does not change anything in the mechanical characteristics of a 2.5" drive, it just replaces the SATA or SAS electrical interface with a NVMe electrical interface.
You can mount a U.2 drive in any location intended for 2.5" drives, as long as its height can fit there.
However, 2.5" drives come in various heights. Many laptops and mini-PCs that accept 2.5" drives accept only some of the smaller heights and they do not accept the greater heights, like 15 mm, which are typical for enterprise SSDs and HDDs, regardless whether they have a NVMe, i.e. U.2, or a SAS interface or a SATA interface.
This new high-capacity U.2 SSD has the standard 15 mm height of the 2.5" form factor.
Apparently TDP is 30 watts¹, according to the product brief. I would imagine it's a single PCB with flash chips on both sides then thermally bonded to the aluminum chassis. That should keep all chips at approximately the same temperature. On its own it could be easily air cooled, but with 24 in a 2U chassis you'll be having some decently hefty forced air over the drives.
1. For comparison, an HDD usually comes in around ~10 watts
I tempted to say that blood is better one. Among other things blood has iron, while tears just salt. Last, but not least it's for thermoregulation of the body.
If we're evaluating blood and tears for cooling, I'd argue that sweat is significantly better as a renewable resource, and also specifically adapted towards evaporative cooling.
The transfer rates limit how much each chip can be active at any given time, so a heat-aware writing allocator can pick the least active blocks for the next writes and distribute the heat accordingly. Even if it’s not heat-aware, the tendency will be that the writes will be distributed over as many chips as there are, and so will be the heat generated.
Now, I would LOVE to see this much SLC flash on a direct to bus attachment setting.
the u.2 form factor indeed evolved from chassis designs that were originally 2.5" drives. It's now kind of becoming obsolete with new designs using things like E1S, E1L (exactly the correct height to be slotted into a 1U server, it's like a slightly wider M.2, but meant to be insertable and removable), and E3S and E3L.
Note that the 245TB is an E3L, the half size version of it come in smaller size.
Over the past few years the main improvement in SSD capacity has been due to them stacking an ever-increasing number of NAND layers in a single chip, with state-of-the-art SSDs already having over 300 layers.
No need to worry about cooling when each layer in the sandwich is only a fraction of a micrometer thick!
> What accounts for the premium price/TB of these extremely high capacity enterprise-targeted drives?
Spare capacity, mostly. That’s why they have higher endurance. If you want to double the endurance of a given drive, tell the controller to allocate twice as many spare blocks and report less capacity than you would otherwise.
In this case, you are also paying a premium for the PCIe attachment instead of SAS, and a lot for price elasticity. You see, with drives like these you slash space and energy consumption in relation to HDDs by a large number, and that allows you to pay a premium for the device, because, at the end of its lifetime, it’ll have more than covered the cost difference in saved space and energy.
as in that it reflects market prices in any way. I feel like anybody who works in this field knows that Dell etc. rip off naive customers this way and you can either negotiate it back to reality or just buy 1 small drive and order your own separately
3.84TB SSD SAS ISE, Read Intensive, up to 24Gbps 512e 2.5in with 3.5in HYB CARR, AG Drive
Dell Price $8,825.13 /ea.
3.84TB SSD SATA Read Intensive 6Gbps 512e 2.5in Hot-plug AG Drive,3.5in HYB CARR, 1 DWPD
Dell Price $7,893.91 /ea.
3.2TB Enterprise NVMe Mixed Use AG Drive U.2 with carrier
Dell Price $6,596.39 /ea.
I don't see a 'write intensive' option (I only looked around for a few minutes), but I can imagine them being 2-4x those prices.
4-5x times what it would have been if not for the demand from AI. According to my rough calculation 4-8tb ssd drives were going to reach parity with hdd this year
With a modern CPUs hitting 400W it's already a problem to fill a rack top down with servers like you could do before: too many heat to dissipate and transfer, too much power to provide in the first place.
Just imagine something like 2S 9565 in at least 2U machines: with 10 server x 2U x 2 CPU you would have 8kW in the processors alone and you didn't even fill half of the 42U rack.
Getting rid of 30 watts of heat is trivial compared to say, 300 (I don't quite know how to read that ratio since a 2.5kW SSD seems a little high to me).
A big consideration for efficiency and TCO calculations is the number of servers required to house the drives. NVMe drives tend not to be in external JBOF enclosures.
Fewer servers means fewer cpus, less RAM, fewer fans, and maybe fewer switches.
(Im)patiently waiting for this AI-generated memory crisis to pass (or the bubble to pop) so SSD prices can crash back down again. Been dreaming of replacing my RAID6 HDD setup with a RAID1 of SSDs and a hot spare.
Can someone who knows explain what is the benefit of having all that data in one ssd instead of splitting it up into hundreds of individual drives? Does the single ssd benefit is more performance or does it really tuen out to be cheaper than hundreds of individual drives?
It’s about density in a datacenter. With this you have 1PB in 4 drives, fitting in a 1u rack, which is just incredible. Also these drives don’t use regular SATA or SAS, they use PCIe, so these drives are also quite fast in comparison. Density has a power efficiency aspect as well both in just having fewer drives and requiring fewer servers to put drives into.
A 42U rack filled with 1u servers with 8 drives each, will have 84PB of data. It feels like it was a few month ago where you could buy a rack with 1PB of storage, and that was awesome. Not anymore.
You’re actually right, it’s just that datacenters like density and will gladly split your data onto hundreds of these little amazing magical bits of technology rather than hundreds of less magical ones in the same physical volume.
DENSITY. Hyperscalers want to store as much data per rack and per data center as possible. They will eventually have hundreds of thousands of these drives.
God damn. I know somebody that became a multi-millionaire from web hosting in the 2000s and his entire data center back then could have been replaced with just one of these SSDs.
Want, but then need two for reduncancy... then a spare for recovery... why not 3 raid or zfs... imagine the resilver time on this. It's hit the limit of data surety surely.
The datasheet shows 3GB/s sequential write, which for 245.76TB means writing the whole drive takes around 22h45m. Odd that the endurance is specified as "1.0 SDWPD", which is almost meaningless since the drive takes roughly that long to write at full speed.
At scale, 1.9 times more energy is required for an HDD deployment
...but those HDDs are going to hold data for far more than twice as long. It's especially infuriating to see such secrecy and vagueness around the real endurance/retention characteristics for SSDs as expensive as these.
On the other hand, 60TB of SLC for the same price would probably be a great deal.
Perhaps their usual buyers just care less about retention?
Those drives aren't going to be used for cold storage, and it is basically a guarantee that there will be checksums and some form of redundancy. Who cares whether the data is retained for 10 or for 15 years after writing when you can do a low-priority background scrub of the entire drive once a month, and when there are already mechanisms in place to account for full-drive failure?
QLC retention reported to be around 1 year in unpowered state. I would assume, that drive does background refresh, though. No idea what effect it has on total drive lifetime. It is still mean that if you use it for cold storage it has to be powered.
A drive's write endurance rating is derived at least in part from the JEDEC standard data retention requirements: 1 year at 30C for consumer drives, 3 months at 40C for enterprise drives, IIRC. Thus, a drive that has reached the end of its rated write endurance can be expected to have those retention characteristics. A drive that hasn't been subjected to that much wear will have significantly longer retention.
Why is it mean? Why would you want to use a technology that is unsuitable for cold storage for cold storage? You won't even get the power / IOPS benefit if all it does is an infrequent replication of data and is then switched off.
I believe it has read speeds of 13GB/s, not 3 (unless you are referring to an equivalent array of 10 HDD). It will almost certainly be used to store training datasets and model weights. Which I assume are good use cases for fast sequential reads.
This is an ad, not a spec sheet. The vast majority of the people buying this understand the endurance and retention characteristics of this type of device. It isn't going on Amazon or Best Buy, and the target market knows how to ask the right questions.
In the fall, I bought a 4TB OWC Thunderbolt 5 external SSD for my Mac Studio 512 GB for $500. It was cheaper and faster than any other solution. Now the price of both have at least doubled. I've seen Mac Studio M3 Ultra 512 GB go for $20k+ but I don't know if those are real prices or scams. We are going on 6 years of global supply crunches, I'm really getting sick of it!
We bought a 2TB Sandisk SSD back in 2024 for around $95 from best buy. Today 1TB ssd by sandisk is $166(the cheapest that i found, and it goes for $199 in walmart). Market is forcing people to become renters than buyers and there is no force countering that idea. Its a market failure that people will study 10-15 years from now.
EDIT- The same 2TB ssd is now $329.99 at bestbuy.
It's not a market failure, it's just supply and demand. There are many computer components competing for the same resources (fabs, wafers). Demand for GPUs, RAM etc. has increased a lot due to AI, but supply is still the same due to new fabs being huge investments that take years to build. Of course the price goes up.
> There are many computer components competing for the same resources (fabs, wafers).
That's a bit off the mark. Processors, RAM, and flash memory each require their own specialized fabs. TSMC makes processors but not DRAM or NAND flash. Kioxia makes flash only. Samsung has all three types of fabs. Micron does DRAM and NAND, in different fabs.
The increase in SSD prices is not because there are many components competing for the same constrained resources, it's that they are complementary goods being subject to the same dynamics in parallel because the servers that are causing this demand spike need all of those components. Where we do see competition for the same fab capacity is in the mix of DRAM types, where GPUs want HBM, server CPUs want DDR5, laptops and phones (and the occasional server CPU) want LPDDR. And competition among different types of processors for TSMC's fab capacity.
I mostly agree, but if you go further back in the supply chain there are a number of common inputs/tools. For example, as 2025 the new LPDDR lines are using EUV systems, which means for new fab production both DRAM and logic producers are competing for machines from ASML.
It doesn't change your point that these lines are different and the immediate price spike is not about them competing for capacity in the same facilities, but the fact that manufacturers are committing to large enough future purchases to drive new fab construction means that future pricing outlook (which does impact the current prices to extent) does involve some amount of competition between different types of semiconductor products.
Shared skilled labor is another thing to consider.
That's only going to change on similar timescales to the fab capacity, which is to say multiple years.
Thanks for the correction. You're absolutely right.
An industry doesn't need to increase supply when demand increases, they can absorb the demand as profits. More so in an industry that is hard to enter into. Consumers hate this and call it all sorts of things like market failure, gouging, etc... In this case, the suppliers are slow walking increases in production and enjoying a run up in profits. They raise concerns about oversupply, as if this deep learning thing were some passing fad and the market will have no use for the new supply if it passes. It's a dubious notion though, the demand is here to stay and new supply needs to come online to meet demand. We simply need more silicon in the market. High prices should eventually bring new supply online, but I'm a little disappointed by the rate of the ramp up.
>It's a dubious notion though, the demand is here to stay and new supply needs to come online to meet demand.
This is a big bet. Look at what happened in 2001 with the dot-com boom. We're still trading on their dark fiber over-build today. Meanwhile any overcapacity built in fabs will quickly be made obsolete by newer and better fab technology (or at least, that's been the pattern for the past 30 years).
I think you're missing the fact that building new supply takes time and sustained commitment, and there's simply nobody in a good position to make that commitment without losing big if your thesis turns out to be wrong.
If the demand for new AI builds is eventually satisfied, or worse, craters overnight, then who will be left holding the bag? It sure won't be Google or Apple, or even NVIDIA - it will be TSMC and Samsung.
But as you mention, this is all temporary. Either you're right, and demand will remain sustained long enough for some of the providers to decide to take that risk, or demand will crater and prices will fall.
The fact that the S&P 500 is near record highs at the same time as consumer confidence is at a 70 year low is not encouraging for continued all steam ahead in my mind... but then it's easy to predict a general future recession, and much harder to predict it to the day.
The idea that we're still trading dark fiber from 2001 is an old narrative right? I guess it is still floating around. But, we're in a second big fiber build out for not only residential, but also to connect all these new data centers. Could there be some demand oscillation for silicon? Sure, but overall deep learning is not some passing fad even if someone gets too far out over their skis and over buys. So far Big Tech has increased spend 3 years in a row and next year they'll spend more than this year. Demand has evidently not let up for AI services, we still need more silicon.
I doubt consumers will regret these compute purchases in 3 years, my 3 year old GPU is still holding value, actually increased in value, and I use it more now than ever.
I did predict or expect that if oil went to $150 we'd be in recession territory, currently hovering below that level and folks are feeling the squeeze and aren't happy. Things could get worse or better, tough to tell, but I think silicon demand is more or less secular and will do relatively well in a variety of macro conditions.
This seems like it would be easy to structurally solve - big tech could make strategic partnerships with 5 to 10 year horizons with the very fabs in question. If they promised to spend a flat or growing $X on silicon (without easy contract cancellation) over the next decade, then the risk would be entirely on the tech companies and not on the fab companies. Of course, there's always bankruptcy to worry about, but that's less of a threat for Google and Apple than it is for OpenAI or NVIDIA.
The fact that we _haven't_ seen such deals be made, and that ~50% of new datacenter builds have been quietly cancelled[1], suggests to me that we're dealing with paper demand more than true sustained demand.
1: https://www.tomshardware.com/tech-industry/artificial-intell...
The dark fiber from the dotcom era is approaching the end of its life expectancy. Contracts for leasing dark strands reflected that. Most of it likely still has useful life left that the owners will want to monetize but it might change how it is used.
> too far out over their skis
Bad metaphor. Not being forward enough is a common mistake. When learning, you may want to make yourself _feel_ like you're too far forward. We can commonly think we're too far forward when actually overall we're still too far back. Teaching it is very hard.
Although it is dependent on your style of skis (e.g. carving) and style of skiing (powder, racing, cruising).
For me, it is very rare that I go too far forward (I began with a leant back style and I haven't rectified that after many years of skiing). I try to prioritise fun over ability.
> the suppliers are slow walking increases in production and enjoying a run up in profits
The major companies have been trying to build >$100B of new production capacity in the US for years now. All of these manufacturing facilities have been significantly delayed by NIMBYs using the same "environmental and community concerns" advocacy slop that hinders almost all productive industry in the US.
Blaming the suppliers is a lazy take. They aren't responsible for degrowth activists being able to dictate what we are allowed to build.
Also my understanding is that a lot of the suppliers remain skeptical / prudent about the long term demand from AI. Not their first boom/bust cycle.
Or their continued collective inability to predict or manage supply and demand, it happens repeatedly and previous events were very minor in comparison... The other obvious reason is profiteering but pretty much impossible to prove.
They are public companies, their bottom line isn't exactly secret. The only company that is really squeezing the market in term of prices is nvidia and they aren't the production bottleneck.
By the way, interesting podcast on that topic: https://www.youtube.com/watch?v=mDG_Hx3BSUE
> The other obvious reason is profiteering
When supply is constrained, widgets to get sold to whoever is willing to pay the most. It’s capitalism, baby. Arguably this is a good thing, because the limited resources are allocated to the people who can create the most value from them. And high prices provide an economic incentive to bring more supply online.
Nvidia is making big profits right now because they don’t have a lot of competition for their fancy AI chips. The prices will come down when they have more competition.
That only works if the market is not controlled by basically 3 vendors and if the market isn't also pretty much impossible to break into. The value proposition is tenuous at best, the real value is being produced in other arenas, not this one.
We saw this during the pandemic when everyone wanted to purchase bicycles. Shimano (they make bike transmission parts and brakes) saw a huge increase in demand. Shimano didn't want to invest into new factories as they assumed the demand was a temporary spike. I'm pretty sure this is what is happening currently with RAM, SSD's and processors. New fabs are coming on line, and Apple is looking at both Intel and Samsung to bring additional capacity online. If the AI boom dramatically slows, it's going to be interesting to see how the industry responds.
I don't get this, is there a world where we need fewer cpu/ram/ssds in the future?
Like, there's so many things that could benefit from a cheap processor involved in their operation, the growth seems effectively unlimited.
There's definitely a world where we need newer products, but not as many or as fast in iteration. My gaming PC has 128GB of RAM... And I built it years ago when RAM was laughably cheap. I still never touch the sides on it.
Ok sure, assuning your gaming machine lasts 5-10 years, what about everyone else who is born (or turns 18 or whatever) during that timespan?
I think I'm agreeing with you but its also not something easily dismissed. The DRAM Cartel has been found to be distorting the market on numerous occasions by various regulatory bodies. There is a boom-bust cycle that occurs with DRAM and Flash memory. The Cartel claims they always lose despite the fact that demand seems to always steadily rise.
The pandemic caused once such boom-bust that resulted in a rather large downturn in demand in 2022-2023 referred to as the pandemic hangover. During that time demand dropped following overspend during the pandemic and members of the cartel drastically cut production at times to keep prices above cost. Even after the demand recovery began in 2023, the cartel members were slow to increase production and made little to no investment in production capacity in 2024-2025. Creating a shortage.
The AI hype cycle has exacerbated the shortage by creating speculative purchases and then panic buying. Remember the shoe company that pivoted to AI?
So Cartel market manipulation is partially to blame for the over 100% increase in prices and the shortages.
It is a market failure as the markets fail to cater for the demand due to structural factors. It’s basically designing database structure that causes huge latency due to use pattern changes and you are not able to alter the structure because that will demand even more painful downtime and latency and the pattern may eventually change and make the alterations redundant.
> It is a market failure as the markets fail to cater for the demand due to structural factors.
The Market™ is perhaps going after higher margins. If you can charge $200 instead of $100 for a widget, why wouldn't you?
Ideally at some point someone will see the margins and be motivated to go after some of it and offer things for $190, but the ROI on upfront fab costs given market risks may not be high enough for the business uncertainly that needs to be taken (boom-bust cycles).
No participant in The Market is obligated to "cater" to demand if they do not think the juice is worth the squeeze.
That’s why it’s market failure, not some specific company failure
It’s market consolidation unchecked. How many nand suppliers are there in the US? How many globally? Now add the sanctions on China, and tell us how that’s a competitive market.
The reality is all the things we learned from the Great Depression have been forgotten, and half the voting public seems to champion monopolies while simultaneously acting dumbfounded that the end result will always be an increase in costs.
The demand side is the world's "investors" buying up a product without producing profit. This results in locking out companies operating without such "investors" from accessing the product and using it to produce profit. This is a failure.
And it's exacerbated by the companies that have to invest their own Real Money to produce factories to make the chips. Same goes with the power utilities. Everyone knows it's a bubble and they don't want to be the one left holding the hundred billion dollar bag like back when Enron went belly up.
But they create a new problem for themselves of converting consumers into non-brand loyalty pure price hunters. I don't think I'm speaking for myself when I say I will use the other guy if it saves me a penny for the rest of my life.
The counter-example I hold up is Harley-Davidson. They could hardly pay a bill in the late '70 through the '80s. They'd even get free samples from suppliers and put them into production. They survived with very strong brand loyalty from customers and after-market companies.
Samsung sells directly to customers of all sizes, so they have a very good idea as to where products end up. Cutting the legs off of innumerable small businesses is despicable behavior.
I don't care if Samsung is literally starving to death, where I was previously at least a marginally loyal customer not even looking at other brands because prices were fine.
Supply, to a great extent, is a policy decision. Demand, to a great extent, a policy decision. This is not the silly metaphor of individual consumers and producers acting independently of each other.
It is market failure when normal people can't afford technology that they used to be able to afford.
This was avoidable. Over $100B of new production capacity in the US alone has been in the works for years. The usual "environmental and community concerns" lawsuits and activism that plagues building in the US has delayed almost every single one of these projects by multiple years.
NIMBYs and degrowth activists own this particular "market failure". The companies have been trying to invest in capacity ahead of demand but have been prevented from doing so.
I think it's really easy to blame one or another specific group for the "market failure" but the reality is that if there were 100B worth of capital looking for 1T worth of gross over 10 years there would be a lot of places that would welcome that kind of spending. I think it's probable that in specific places there's been push back, and those places are probably where the work was historically concentrated. But if that's the case, you should really be asking "why is there push back now when there wasn't before?" And I don't think the answer is because the entire community has adopted a "degrowth" agenda. Rather, I think it's going to be more about chemical dumping in the water supply or use of scarce resources that the community would rather not spend for the profit of a few people who don't live there.
I think a lot of people will accept industry when industry tries to be a good steward of the community's environment.
It is happening to production facilities across the US. Both new production facilities and expansion of existing facilities. The locale doesn't seem to make a difference.
As with the sudden inexplicable "concern" about data center construction, I don't think this activism is organic.
As with the sudden inexplicable "concern" about data center construction, I don't think this activism is organic.
I don't think it is sudden, there was much of the same concern about Amazon warehouses and anything else that cuts down trees, disrupting wildlife, and human neighbors. The reason it is more of a concern with modern datacenters is because they don't come with a significant amount of new jobs for the communities they are affecting. People will put up with a lot if means their unemployed loved ones can get on their feet.
With AI being a looming threat to everyone's livelihood it is compounding the reaction, not only do these datacenters not provide new jobs, they are actively contributing to people losing their current jobs.
If these new datacenters were going into empty buildings in cities that already had adequate infrastructure, then no one would care. They would still be against ai in general, but the datacenters wouldn't be a problem.
edit:
They could probably get a ton of support if they went into one of the many cities with crumbling infrastructure, took over abandoned buildings, and paid to fix up the roads and stuff. Like if they moved into an abandoned factory in flint, and fixed the water supply they could be heros.
You find populism in the US inexplicable in 2026? Where have you been for the past 18 years? You get instant political capital by framing anything as elites vs the common man, and the tech company billionaires cannot escape that "coastal elite" label, or hide that they want to build more data centers.
Are the NIMBYs in South Korea blocking Samsung too? If not, then the root cause can't be that.
there is no serious degrowth movement in the US. absolutely delusional to pin anything, positive or negative, on them.
Agreed on the insignificance of the degrowth moevement on seriousness. However there is a serious NIMBY movement, and the degrowth movement provides the moral cover for a lot of very selfish people that are not serious about degrowth but wish to accomplish the same goals for individual projects near them.
You were not are renter in 2021 when NVMe were same price as now by TB, stuff is just becoming more expensive on market shortages.
Last half year ate up three to four years of earlier price regression, that's about it.
As long as this plateaus here, as prices did for last 4 months, that's just the new equilibrium where it has the chance to get better again, would no be all doom and gloom about personal computing yet.
The dollar is losing value, we cant hit equilibrium, it will always cost more dollars. And that is assuming the mfg are not price maximizing (they are).
> Market is forcing people to become renters than buyers and there is no force countering that idea.
The doomerism surrounding this topic is wild.
Gas prices have also gone up recently, but I don't see the same claims about how it's the end of personal vehicles, that the prices are never coming down, or that we're all becoming renters instead of buyers.
Saying that there isn't a force countering this is peak doomerism. New factories take a long time to build. These companies didn't have spare production lines waiting for demand to go up.
If you don't see people talking about the "end of personal vehicles", it could just be that you haven't looked very hard.
It's intuitively obvious to a lot of people that the era of personal, wholly owned transportation is waning. A lot of people seem to miss the second clause of that old "you'll own nothing" phase, the part where most people are happy about it!
When vehicles drive themselves, and there's a large enough pool that one can show up pretty reliably within a few minutes of your needing one, how many people are going to choose to own when renting is cheaper and easier?
Because renting will likely never be cheaper. Maybe the occasional city dweller may see a profit, but those of us who spend hours in our cars each day will not.
If this was really an economically viable plan, we’d all just take taxis everywhere.
And some do. Most don’t, for the same reasons self driving doesn’t matter.
It wasn't economically viable 20 years ago.
What about 20 years in the future?
It’ll be outcompeted by flying cars of course.
Probably everyone that doesn't live in a dense metropolitan area, and anyone that can do the math that "Owning my vehicle for 100,000 miles costs an average of $0.50/mile compared to the $1+/mile getting a lift costs, and once it's done I still have whatever value remains in my vehicle"
> If you don't see people talking about the "end of personal vehicles", it could just be that you haven't looked very hard.
Or it could be that I mostly talk to people in the real world, and less so follow the echo chambers online that think "you'll own nothing" is a foregone conclusion about the future and fit their worldview to match.
Ug. I picked up a 16TB hard disk last year for about $250. Went looking for another one last month and the same disk is about $500. Painful.
At least the fun part with these situations is that the prices of old crap goes up and I can sell that old crap I have for actual $.
I remember selling some ancient GPUs that I was going to throw away during one of the previous booms.
Indeed. Now...where did I put that pile of PC3 memory...
When I upgraded my computer a little over a year ago, I got 128GB of DDR5 RAM instead of just 64GB. Glad I did.
I also thought about upgrading my gen4 SSD at the same time. It was still meeting all my needs, so I thought if I waited, I could get a bigger, faster one for cheap later. Mistake!
Are things too expensive now, or was that miraculously cheap because of a glut of supply relative to demand at the time? I think my first SSD was something like $600 for 160 gigs, one of those intel drives that came with a speed demon sticker, so let's appreciate what we have for a sec. Or we could go back to 5 megs of spinning rust for $thousands.
Things will normalize, semis are a boom and bust industry, and it takes a massive amount of capital investment to increase supply. Because of that boom and bust, the producers are wary of growing their supply to keep up in lockstep with demand, because they might end up with a glut that they have a hard time selling, and a large pile of debt from building out that glut. Semiconductors at the leading edge take very large, very expensive, very rube goldbergian machines to make them, and a very skilled workforce to make those work.
If this stuff stays this expensive/profitable, industry will grow capacity to boost earnings until they overshoot, there's not some grand conspiracy to make you rent everything. We're just in a massive dislocation right now because we've discovered this awesome new tool that's boosting demand by a ridiculous degree. Last year we were at ~$300B of capital investment in DCs in the US, this year it's looking like ~$750B.
> If this stuff stays this expensive/profitable, industry will grow capacity to boost earnings until they overshoot, there's not some grand conspiracy to make you rent everything
Production cartels exist to prevent individual greed from "ruining" it for the rest of the group. OPEC does it in the overtly, memory manufactures do it covertly, and have been convicted of conspiring in multiple jurisdictions, on multiple occasions.
Good point re cartels. But I still don’t think there’s a shadowy group trying to make owning things impossible for normal people as an explicit goal.
It's not a goal, but a secondary effect. Companies that build long-lasting products for sale to customers will have lower profits than those that rent-out cheaply made goods. Over a multi-decade window, with some bankruptcies, consolidation roll-ups and astute leadership, you can guess which model will win, based on nothing but market forces and imperfect information. The collusion is based purely on P&L performance.
I remember buying a 80GB Intel SATA SSD (which was $450 USD at the time) and putting it into the DVD drive bay in a 2009 17" Macbook Pro. At the time there was no official SSD support, but there was an aluminum holder you could buy to remove the slimline slot loading DVD drive, and put a 2.5" form factor (7.5 or 9.5mm?) height SSD or HDD in its place. Was a huge performance increase for the OS/boot drive.
I just bought a few more Intel 310 Series 40GB mSATA ssd’s on eBay for $10 each. They are ancient and comparatively slow, but reliable and extremely low power, so much so they can be perfectly powered by a USB 2.0 port in an enclosure if needed, and don’t need a heat sink. My original 80GB version ($190) that I put in my W520 thinkpad in 2011 is still going strong.
Basically I like to use them as a cheap “better” thumb drive, putting ventoy on them with no concern for space, along with a fat32 partition for bios updates.
These along with a nanoKVM box my siblings can attach to the computers I built for them makes remote troubleshooting relatively painless when rustdesk isn’t sufficient.
When people talk about AI investments of trillions of dollars, its never mentioned how high a price we are paying at a planetary scale in externalities. Billions of people are now paying a lot more for these components, memory, ssd, cpu; all these costs must be factored into the total cost of AI. AI companies are being subsidized at the expense of consumers. The current supply chains were built for consumer demand, which is hijacked by the AI grifters.
Same a consumers (car drivers) subsidizing businesses (trucks). Road damage is almost entirely caused by trucks, 4th power of weight, but the cost is borne by us.
True, I've been searching for 2 weeks now for components to get a PC back online. I suspect the VRM blew in my motherboard, and trying to find x570 or b550 old Gen tech is drying up because that's the only thing affordable. Meanwhile decent 3D cache CPU + motherboard combos afe going cheap because they can't move stock. The price looks nice but the second you add DDR5 RAM it is just a gut punch. Scalpers are taking advantage of this as well, as well as Chinese "Brand New" x570 motherboards listed as gaming consoles on Amazon.
This is the first time I've had a computer have a complete failure and no way to fix it as I look around at the used market and the latest generation market. I'm kind of just disillusioned.
So you pay for parts that are all very expensive due to AI in order to get the machine online so that you can pay for AI subscriptions to do work that is disappearing due to job cuts due to AI.
When I was 18 I was hoping we'd have 500TB hard drives by now.
the interesting thing is that scaling laws (at least Moore's and the like) are complete lies. (Moore is fine if you treat it as a weak observation that shrinking features to half gives you 4x as many devices in the same area.)
what happens is that the industry goes through certain discrete technical upgrades. for instance, EUV in fabs, or GMR disk heads. none of these are really planned, none of them are exponential. and they usually interact with other phenomena (such as Dennard scaling).
in a sense, the phenomenon is more like "expectations are exponential, and this motivates manufacturers to schedule updates".
hard disks are still improving, arguably similar to how they have in the past, but there are limits to demand. the consumer market has mostly dropped out, for instance due to flash.
even in flash, there is no exponential scaling in devices. people got excited in the initial startup, when for instance, mature TLC is so much better than early SLC. but all that's over: it's both mature and we'll probably never see PLC. even QLC is interesting in that it illustrates that most of our storage is very cold.
And you are now how old? 19? 79?
36 ;)
Last fall I bought a 4tb Samsung 990 pro drive for $260. Now it retails between $800-900.
Are you sure it’s the same ssd. The new nvme5 disks are 3x faster
and 2TB SSD's in 2020 were expensive too, after dropping to lows late last decade
ebbs and flows
Meanwhile I'm still dreaming about any consumer and affordable 32TB or even 16TB portable SSD. Innovation and market for consumers are going backwards.
Funny thing is that one of the best you can get is the Crucial (Micron) 8TB one but even that one gets more expensive. I have the feeling it will be gone completely soon.
Enterprise NVMe on the high end is now starting to ship batches at $1000/TB with existing stock around $500/TB. No consumer is going to pay that.
But if you're buying a $500k GPU server putting 100TB of nvme in there for $50-100k is justifiable.
There was once a 2.5" SSD Mushkin Source 16TB SATA drive. At its cheapest it was ~1700 USD (or 1500 EUR). That was mid 2023 (like 3 years ago!).
Nowadays it feels like that this time and price region is like decades away in the future. I was hoping I can store more data in future on modern tech like SSDs and not less.
Yeah it sucks :( Almost exactly a year ago, I got a brand new 15.36TB Kioxia CD-6R (u.3 pcie4x4 drive) for $1450+tax from serverpartdeals.com - that same drive is now listed for ~$4600 (and it’s also out of stock there)
Nobody is stopping anyone from buying a USB-C powered and connected very portable 2 or 4 slot external NVMe enclosures.
The old SATA SSD form factor is dead and wont come back.
OWC ThunderBlades exist, but 32TB will set you back 9 grand.
You should be able to assemble something with USB-C for under 5k. That's not a mass consumer market thing, but perfectly doable, if your use case warrants it. We are stuck with 2021 pricing, but now with options of 8TB per NVMe drive at way higher speed.
36TB+ HHD external WD drive combos were always around EUR 1000 over last 5 years. With a short low end around EUR 600 in 2023
https://www.owc.com/solutions/thunderblade?sku=OWCTB3TBL8X32
>Nobody is stopping anyone from buying a USB-C powered and connected very portable 2 or 4 slot external NVMe enclosures.
That's more expensive than buying a single large capacity drive. It's also a terrible idea. I would never trust a low cost chinese controller with terabytes of my data.
>The old SATA SSD form factor is dead and wont come back.
That's true and very unfortunate.
> buying a single large capacity drive
Now that's a terrible idea. You shouldn't trust any single device with terabytes of your data, regardless of whether it is low-cost or Chinese.
I will not miss the awful, half-duplex protocol that never should have won over SAS. I just wish that PCIe switches and cheaper eMMC/UFS flash on M.2 were available for more flexible and cost effective storage options.
Yeah, the problem is most consumer motherboards don't come with that many PCIe lanes. Building storage servers for home use with M.2 SSDs is just not feasible.
PCIe is a switched protocol. We ought to have had better options to make better use of the limited lanes by now. There are a few cards with switches that can give you multiple M.2 slots on x4, but that's only with gen3, which the latest GPUs won't init on, and trying to cable that is hell. Better PCIe switch options for external enclosures exist, but aren't available to mortals that I've seen.
I find it very odd that there is so much faith in "innovation" (and probably "economies of scale").
there is no sign of any impending breakthroughs that would change flash economics much.
slc-mlc-tlc-qlc was very nice but plc will not happen. layer-based flash was also nice but it is ultimately linear (more layers, more cost, lower yield). dimensional shrinks are already stalled because of a tragic electron shortage (per cell).
I guess there's no harm in pining for some other NVRAM technology (spins, etc).
I'm still pining for Optane to make a comeback.
Didn't help that they used "Optane" for two very different products. I agree on the good one though !
The prices aren't going down for large consumer drives because the market is so small, and because the AI DC market is swallowing up everything. There's little demand from your average consumer to have 30TB of storage, let alone specifically SSDs. The average user doesn't have that much data, and if they do a HDD is fine for any practical purpose.
Despite the recent AI bubble you can still buy HDDs in the tens of TBs for a few hundred EUR/USD and you still don't see them in every computer. How high could the 30TB SSD demand be to justify the kind of volumes that drive price down?
In the DC it's the opposite, large and efficient drives are a must to save support all those fancy workloads while driving down space, power, cooling needs.
A few months ago I finished building a new media server based on UnRaid. I populated it with WD 26TB drives. At the time they were about $400 (steep, but a decent capacity/dollar buy). Now they are nearly $1000 on Amazon, a 250% increase. I just hope I don't have a drive failure.
With regards to the new Micron SSD - I wonder how they keep it cool? I don't see coolant ports on it so they must strap a heatsink on.
The product brief says maximum 30W and it looks like the whole enclosure is a heatsink, even has ribs on the back. The expected operating temp is 50C but it's probably rated to operate at higher than that.
P.S. I had to shuck 20TB WD drives that cost 350EUR on sale (now at 400EUR). 26TB drives are now ~700EUR. These external drives were the cheap option. Standalone drives usually cost more.
>The average user doesn't have that much data
The average user consumes that much quite regularly. They've been taught to stream it off of someone else's computer, mostly so that the next time they stream it they can be compelled to pay for it again. It's fun going back to dumb terminals.
But they don't local store that much data, which is what would be relevant for a discussion above local storage costs.
Consuming and needing to store are very different things. Most media is disposable, one-time consumption. How many people stored the newspapers they read?
Why would you want to store every movie or series you watch? 30TB of data is something like 1 year of uninterrupted streaming at average Netflix 4K bitrates. Even more at HD bit rates. How many people would ever store years worth of movies on SSDs no less? Enough for it to drive huge sales in the market?
>Most media is disposable
It isn't. The number of books I've read again, that I've wanted to have my children read. Music. Shows. The old VHS tapes were only disposable because the format was so deteriorating. It was inconvenient because of the physical size.
But, if people realize that media could be non-disposable, how could the recording industry get rich selling you the same albums you'd already bought a half dozen times before? I've been storing everything as FLAC... it'll be good for centuries. Books already have a "good for centuries" file format. I suspect very strongly that 4K is the ultimate res for all media made up until today (no way to do 8k remasters). It doesn't need to be disposable, it just is because there are people who want to get rich off of you renting your own culture back from them.
>How many people would ever store years worth of movies on SSDs no less?
Send me a boxfull of these 245TB SSDs, I'll show you. My film count surpassed Netflix's about 12 years ago, and I haven't slowed down once.
Your Aunt Debbie recorded 37TB of video last year on her iphone and had to delete most of those precious memories to save space for bacon jesus memes.
I checked the specs here: https://www.micron.com/content/dam/micron/global/public/prod...
The interface looks equiv to 4x PCIe 5.0.
That is pretty awful write performance. Does anyone know more about this? I assume all of these hyperdense SSDs suffer from the same drawback. Also, I heard that the E3.L interface can support up to 16x lanes, but there are no practical commerical products at this point.Consumer and data center drives play by different rules. The super high write speeds you see for consumer SSDs are usually achieved through tricks like using sections of the drive as a high-speed buffer and then using a background process to rewrite the data into the drive's high-density NAND storage during downtime. They can also use caching techniques that aren't resilient to power loss. They might allow burst performance that heats the drive up until it throttles.
This is all fine and even desirable for a consumer who will only be writing for at most a minute or two, but with a 245TB server drive you need to assume the performance will be needed constantly. They target sustainable and predictable performance.
Extremely dense QLC chips. Still it's 2700-3000MByte, ie ~3GByte/second.
What should worry way more is DWPD which is abysmal... on the first glance. But if you punch it in the calc it still would take ages to wear it out.
https://wintelguy.com/dwpd-tbw-gbday-calc.plDWPD was the boogey man 10 years ago. everybody worried about it.
now, nobody cares. I have over 500 NVMe drives in our deployment and the drive deaths are not due to wear.
then either your drives are overprovisioned or read-mostly.
it's not that hard to hit 300 cycles on flash.
it's all 4TB or larger plus the drives do wear leveling internally.
A more convenient (and dare I say, faster) tape drive replacement for backups? They do make a good point, it would take 10*24TB drives working in the worst raid configuration to even come close to these speeds.
65 hours to restore a full backup
Yes, but with all that data, how much heavier does it get?
2.231705*10^-13 gram
:)
A single speck of dust could throw off that measurement (~ 1.6 x 10^-7 grams)
It depends on the data. I heard that ones are heavier than zeros.
Yes, this why drives come in the mail pre-wiped - it saves on shipping.
I look forward to have my favourite hyperscaler grant me 1000 "premium" IOPS per VM on this monster.
IOPS? This thing has slower IOPS than an old SATA SSD (~40k / QLC). I think it is meant for sequential operations only.
Note how that is still well in excess of what e.g. AWS EBS GP3 volumes offer (or at least used to, though even now their "80K IOPS" is measured with 64 KiB random transfers, whereas Micron measured that 42K IOPS with 4 KiB random transfers), which is what the person above is gesturing towards.
The same EBS GP3 used to be specified with 16K max IOPS at 16 KiB random transfers until pretty recently.
What’s the intended block size of these things? I thought 4KB was normal, but that doesn’t make sense at 40K IOPS, and doesn’t align with the benchmarks I’ve seen.
Also: price is expected to be $80k. I suppose density is the selling point here, not speed.
https://web.archive.org/web/20260505162256/https://investors...
Rather silly of them to hide investor relations material behind an anonymity-hostile CDN.
PDF for those who want it. https://web.archive.org/web/20260506084407if_/https://invest...
The u.2 form factor is slightly larger than a 2.5" drive. I can imagine the entire space in it taken by Flash chips. I can't imagine what cooling scheme do they employ for the chips in the middle.
The U.2 form factor is a 2.5" drive, not larger than it.
"U.2" does not change anything in the mechanical characteristics of a 2.5" drive, it just replaces the SATA or SAS electrical interface with a NVMe electrical interface.
You can mount a U.2 drive in any location intended for 2.5" drives, as long as its height can fit there.
However, 2.5" drives come in various heights. Many laptops and mini-PCs that accept 2.5" drives accept only some of the smaller heights and they do not accept the greater heights, like 15 mm, which are typical for enterprise SSDs and HDDs, regardless whether they have a NVMe, i.e. U.2, or a SAS interface or a SATA interface.
This new high-capacity U.2 SSD has the standard 15 mm height of the 2.5" form factor.
Apparently TDP is 30 watts¹, according to the product brief. I would imagine it's a single PCB with flash chips on both sides then thermally bonded to the aluminum chassis. That should keep all chips at approximately the same temperature. On its own it could be easily air cooled, but with 24 in a 2U chassis you'll be having some decently hefty forced air over the drives.
1. For comparison, an HDD usually comes in around ~10 watts
Given the cost of 24 of them, you can probably buy solid silver heatsinks watercooled with tears of sysadmins.
Hey! You leave me out of your twisted fantasy!
I just want....I just want hard drive prices to come back down. *sniffle*
I was going to say blood of virgins, but tears are probably better heat conductors.
I tempted to say that blood is better one. Among other things blood has iron, while tears just salt. Last, but not least it's for thermoregulation of the body.
If we're evaluating blood and tears for cooling, I'd argue that sweat is significantly better as a renewable resource, and also specifically adapted towards evaporative cooling.
The tears of sysadmins are fairly cheap though.
It may be the only resource that truly scales with demand.
It's not just a single PCB, but a sandwich of several.
The 4th Earl of Sandwich disagrees.
The transfer rates limit how much each chip can be active at any given time, so a heat-aware writing allocator can pick the least active blocks for the next writes and distribute the heat accordingly. Even if it’s not heat-aware, the tendency will be that the writes will be distributed over as many chips as there are, and so will be the heat generated.
Now, I would LOVE to see this much SLC flash on a direct to bus attachment setting.
the u.2 form factor indeed evolved from chassis designs that were originally 2.5" drives. It's now kind of becoming obsolete with new designs using things like E1S, E1L (exactly the correct height to be slotted into a 1U server, it's like a slightly wider M.2, but meant to be insertable and removable), and E3S and E3L.
Note that the 245TB is an E3L, the half size version of it come in smaller size.
https://americas.kioxia.com/en-ca/business/ssd/solution/edsf...
https://www.exxactcorp.com/blog/storage/edsff-e1s-e1l-e3s-e3...
https://www.simms.co.uk/tech-talk/e1s-e1l-the-new-server-for...
Over the past few years the main improvement in SSD capacity has been due to them stacking an ever-increasing number of NAND layers in a single chip, with state-of-the-art SSDs already having over 300 layers.
No need to worry about cooling when each layer in the sandwich is only a fraction of a micrometer thick!
Access Denied
You don't have permission to access
"http://investors.micron.com/news-releases/news-release-detai..." on this server.
High security on this press release.
Your IP address might be on a blocklist.
works for me. akamai doesn't like you
even my AWS IP is let in without trouble
No problems here ...
The press release is missing the key specification — how many Libraries of Congress fit on this thing?
Usual estimate is 10TB of compressed text-only. So 24x LoC would fit on the drive.
How much is it?
They haven't released details but I was able to find a Solidigm D5-P5336 122.88TB drive for around 40,000 USD, as a guideline. So ... more than that.
Okay, so that 122TB drive costs about $330/TB.
I haven't bought a hard drive or an SSD in at least a decade (I get stuff for free, basically) but…that seems a bit high, right?
Seems like well-rated consumer-level SSDs cost around $250 for 1TB right now.
What accounts for the premium price/TB of these extremely high capacity enterprise-targeted drives?
> What accounts for the premium price/TB of these extremely high capacity enterprise-targeted drives?
Spare capacity, mostly. That’s why they have higher endurance. If you want to double the endurance of a given drive, tell the controller to allocate twice as many spare blocks and report less capacity than you would otherwise.
In this case, you are also paying a premium for the PCIe attachment instead of SAS, and a lot for price elasticity. You see, with drives like these you slash space and energy consumption in relation to HDDs by a large number, and that allows you to pay a premium for the device, because, at the end of its lifetime, it’ll have more than covered the cost difference in saved space and energy.
What accounts for the premium price/TB of these extremely high capacity enterprise-targeted drives?
The word "enterprise".
What accounts for the premium price/TB of these extremely high capacity enterprise-targeted drives?
The extremely high capacity and the enterprise targeting.
I fondly remember when i could buy a well-rated consumer-level SSD for a lot less per TB...
I paid $300 each for my last two SSDs, 4 TB Samsung 990 Pros.
They’re currently selling for $942.72 on Amazon.
> They’re currently selling for $942.72 on Amazon.
The price graphs of these are wild, wow. https://camelcamelcamel.com/product/B0CHGT1KFJ
Looks like prices for NAND spiked at the end of last year, a couple of months behind DRAM.
Happy not to be building or buying any new computers these days.
Density, power efficiency, write endurance, sustained write speeds under continuous load, power-loss protection.
And out of band management, hot plug capable form factors, and a bunch of other things described in the OCP NVMe SSD spec.
https://www.opencompute.org/documents/datacenter-nvme-ssd-sp...
I was quoted $18K for a 3.7 TB Dell NVMe disk the other day. I'm gonna guess these drives are literally a quarter million each
> I was quoted $18K for a 3.7 TB Dell NVMe disk
surely you don't actually think that's realistic pricing?
What is "realistic" in this context?
It is very real, that is the price they quote. You can buy it through Dell at that price.
as in that it reflects market prices in any way. I feel like anybody who works in this field knows that Dell etc. rip off naive customers this way and you can either negotiate it back to reality or just buy 1 small drive and order your own separately
Various Dell prices from the US website:
I don't see a 'write intensive' option (I only looked around for a few minutes), but I can imagine them being 2-4x those prices.$200/TB is reasonable. $300 if it is VERY fast. That is just robbery.
You're getting ripped off. NVMe SSDs are expensive, but not THAT expensive. A 4Tb drive should be around $1k even with some "enterprise" markup.
Apparently $80k, not that terrible in comparison
4-5x times what it would have been if not for the demand from AI. According to my rough calculation 4-8tb ssd drives were going to reach parity with hdd this year
Likely $90k USD MSRP with a wholesale price around half that.
Dell is getting first dibs.
If you have to ask...
I don't think he wants to buy one
‘Contact us’
What is this thing that all pictures of new devices need to come with this black background?
Dark mode.
“For AI workloads: The 245TB Micron 6600 ION provided up to 84 times better energy efficiency”
How big of a deal is this part in relation to the initial upfront costs? I’m not privy to the cost of power for SSD
With a modern CPUs hitting 400W it's already a problem to fill a rack top down with servers like you could do before: too many heat to dissipate and transfer, too much power to provide in the first place.
Just imagine something like 2S 9565 in at least 2U machines: with 10 server x 2U x 2 CPU you would have 8kW in the processors alone and you didn't even fill half of the 42U rack.
https://www.amd.com/en/products/processors/server/epyc/9005-...
It means you don't have as much to cool.
Getting rid of 30 watts of heat is trivial compared to say, 300 (I don't quite know how to read that ratio since a 2.5kW SSD seems a little high to me).
Given that 2.91TB SSDs are a common enterprise size, perhaps they're saying the 1x245TB SSD uses 1/84 the power of 84 2.91TB SSDs ;p
A big consideration for efficiency and TCO calculations is the number of servers required to house the drives. NVMe drives tend not to be in external JBOF enclosures.
Fewer servers means fewer cpus, less RAM, fewer fans, and maybe fewer switches.
(Im)patiently waiting for this AI-generated memory crisis to pass (or the bubble to pop) so SSD prices can crash back down again. Been dreaming of replacing my RAID6 HDD setup with a RAID1 of SSDs and a hot spare.
Can someone who knows explain what is the benefit of having all that data in one ssd instead of splitting it up into hundreds of individual drives? Does the single ssd benefit is more performance or does it really tuen out to be cheaper than hundreds of individual drives?
It’s about density in a datacenter. With this you have 1PB in 4 drives, fitting in a 1u rack, which is just incredible. Also these drives don’t use regular SATA or SAS, they use PCIe, so these drives are also quite fast in comparison. Density has a power efficiency aspect as well both in just having fewer drives and requiring fewer servers to put drives into.
A 42U rack filled with 1u servers with 8 drives each, will have 84PB of data. It feels like it was a few month ago where you could buy a rack with 1PB of storage, and that was awesome. Not anymore.
For when you need to store a copy of the internet, and have been granted immunity for your copy of Anna's Archive.
Power consumption is the single biggest data center cost. This thing takes only 30W. An average 4 TB SSD pulls 6W, so that's a 12x improvement.
Furthermore, 15-60x density improvement reduces server and infrastructure costs because it requires vastly less of everything per EiB.
Cooling and power are the limiting factors of density.
You’re actually right, it’s just that datacenters like density and will gladly split your data onto hundreds of these little amazing magical bits of technology rather than hundreds of less magical ones in the same physical volume.
They’ll still have hundreds of individual drives. Of these drives.
And thanks to the density, they won’t need as many racks as they used to.
Higher density, less power. Those are the bottlenecks in current and new data centers that are built out.
So it's not exactly about cost savings, but having the option to do more, faster.
Also, you could also get much higher bandwidth density out of this vs HDD, and this is great for AI training
Probably for a similar reason why I would rather buy a single 4TB SSD than fourty 100MB SSDs.
DENSITY. Hyperscalers want to store as much data per rack and per data center as possible. They will eventually have hundreds of thousands of these drives.
God damn. I know somebody that became a multi-millionaire from web hosting in the 2000s and his entire data center back then could have been replaced with just one of these SSDs.
Wait until you see what CPUs have done since then...
Want, but then need two for reduncancy... then a spare for recovery... why not 3 raid or zfs... imagine the resilver time on this. It's hit the limit of data surety surely.
A 245TB SSD completely changes the density math for database nodes. The reduction in rack space alone makes this a huge win.
It is wild that a single 2.5-inch-ish device can hold more data than many companies had in total not that long ago.
Would like to see what the internals of this look like, how many flash packages and PCBs are in that tiny chassis?
QLC NAND
The datasheet shows 3GB/s sequential write, which for 245.76TB means writing the whole drive takes around 22h45m. Odd that the endurance is specified as "1.0 SDWPD", which is almost meaningless since the drive takes roughly that long to write at full speed.
At scale, 1.9 times more energy is required for an HDD deployment
...but those HDDs are going to hold data for far more than twice as long. It's especially infuriating to see such secrecy and vagueness around the real endurance/retention characteristics for SSDs as expensive as these.
On the other hand, 60TB of SLC for the same price would probably be a great deal.
Perhaps their usual buyers just care less about retention?
Those drives aren't going to be used for cold storage, and it is basically a guarantee that there will be checksums and some form of redundancy. Who cares whether the data is retained for 10 or for 15 years after writing when you can do a low-priority background scrub of the entire drive once a month, and when there are already mechanisms in place to account for full-drive failure?
QLC retention reported to be around 1 year in unpowered state. I would assume, that drive does background refresh, though. No idea what effect it has on total drive lifetime. It is still mean that if you use it for cold storage it has to be powered.
A drive's write endurance rating is derived at least in part from the JEDEC standard data retention requirements: 1 year at 30C for consumer drives, 3 months at 40C for enterprise drives, IIRC. Thus, a drive that has reached the end of its rated write endurance can be expected to have those retention characteristics. A drive that hasn't been subjected to that much wear will have significantly longer retention.
Why is it mean? Why would you want to use a technology that is unsuitable for cold storage for cold storage? You won't even get the power / IOPS benefit if all it does is an infrequent replication of data and is then switched off.
What kind of usage do you envision for 245TB drive with read speed of 3GB/sec?
I believe it has read speeds of 13GB/s, not 3 (unless you are referring to an equivalent array of 10 HDD). It will almost certainly be used to store training datasets and model weights. Which I assume are good use cases for fast sequential reads.
Right, I misread the spec.
You can trivially modulate flash endurance by tweaking the reported space - the less space you report, the more spares you have.
This is an ad, not a spec sheet. The vast majority of the people buying this understand the endurance and retention characteristics of this type of device. It isn't going on Amazon or Best Buy, and the target market knows how to ask the right questions.
You're right. Facebook / Meta has advocated for QLC SSDs https://engineering.fb.com/2025/03/04/data-center-engineerin...
The word AI can be safely deleted wherever it occurs in this press release.
Very cool bit of tech.
2 TB NVMe SSD on the German market:
* October 2020: around 200 EUR
* May 2026: around 230 EUR
Cost? Durability? Iops do we know?
In the fall, I bought a 4TB OWC Thunderbolt 5 external SSD for my Mac Studio 512 GB for $500. It was cheaper and faster than any other solution. Now the price of both have at least doubled. I've seen Mac Studio M3 Ultra 512 GB go for $20k+ but I don't know if those are real prices or scams. We are going on 6 years of global supply crunches, I'm really getting sick of it!
Data centers are winning.