The MacBook Pro on which I’m writing this piece needs memory that can keep up with a powerful processor running many programs at once: so it uses a standard called DDR, “double data rate,” which runs at a reasonably high voltage and offers high bandwidth. The processor on my iPhone is less powerful, so it needs less data at any given moment; but voltage matters enormously, since every milliwatt allocated to memory is drained from the battery. So smartphones use LPDDR, “low-power double data rate,” a variant of DDR engineered to operate at lower voltages.

> So modern DRAM manufacturing is an extraordinarily complex and expensive process. Building a single state-of-the-art DRAM fabrication facility, a “fab,” will cost you about $15 to $20 billion; acquiring all the necessary equipment, like lithography tools and etching machines, will cost you another few billion; and then it’ll take you a few years of producing substandard and defective memory chips before your yields start to look competitive.

Extraordinarily complex and expensive! And yet I look at all the money being shuffled around between Nvidia and Google and Microsoft and Amazon and Apple and can't help but think that this is a tiny amount in comparison to what they're moving around on the stock market buying shares in each other.

Apple in particular has $20B in its couch cushions and is very vertically integrated and hardware-focused. Apple silicon is currently made by TSMC, but it seems they'd be a prime candidate to spin up their own memory fab.

I suppose the biggest problem to current executives at each company is the "few years" until that investment yields results, in the short term it's better to pay through the nose and buy GPUs with HBM at any price.

If the Big 3 had been more prudent, they could have kept profits lower and avoided this, but now they've probably dug their own grave and will see massive decline in profits starting from 2027/2028 when Chinese capacity really comes online.

So, that's another way we are financing the LLM machine and the trillion-dollar valuations of those corporate behemoths.

The Iran war is spiking the price of oil and will likely cause shortages of pretty much everything if it isn't ended.

The Ukraine war is helping with that by destroying Russian refining capacity.

The memory shortage is set to do the same to consumer electronics, which are absolutely essential to the modern economy.

Meanwhile the AI fad is seeing huge layoffs. At the same time as the AI Big Cos are beginning to show signs of ending the subsidised free lunch phase and moving to a utility model, which will raise prices for every company that is hooked on AI.

Also tariffs. Although I'm not sure if anyone knows what's happening there.

And farms are failing. Climate change will accelerate that, so there will be food shortages within a few years.

If it's not cynical and deliberate, it's an astounding confluence of (literally) catastrophic mismanagement.

  If a dominant buyer locks up most of the supply chain through exclusive contracts, it prevents rival companies from getting the materials they need to survive, which violates laws like Section 2 of the Sherman Antitrust Act.

As other commenters have pointed out but I might have missed in the article, compute maturation is amplifying memory constraints right now and making it worse. Device upgrade cycles are getting longer because most compute-based products have matured, with CPUs not seeing substantial gains and memory usage really only expanding at the absolute top end of workloads pre-LLMs (3D and HPC in particular). An iPhone 14 still has almost all the features of the iPhone 17, because the compute capabilities are remarkably similar; Geekbench shows a performance delta of ~25-30% between the 14 and 17 Pro Max models, which is pretty paltry considering the devices are separated by four years of manufacturing improvements. This extends into desktops, laptops, tablets, STBs, and more, with only VR devices and larger ARM/RISC-based kit seeing more substantial uplifts as general designs improve.

So with compute stagnating and memory constrained, my money is on vendors taking this as an opportunity to gradually shift away from a yearly release cadence and slow down to a biennial cycle that alternates between budget and flagship launches every other year. Even if LLMs fail spectacularly and all that memory capacity becomes available, HBM memory likely isn't to find its way into many consumer devices (just ask AMD how it worked out for them on consumer GCN GPUs).

The name of the game, especially for consumers, is efficiency - "potato builds", as I've been calling them. Software and services optimized for lower power, smaller-specced devices of increasing age instead of pandering to flagship devices with poorly optimized code or engines for the sake of new shinies (like Raytracing). Between the memory shortage, shifting geopolitics, rising costs, and stagnant wages, consumer purchasing power is going to be squeezed like a vice for the foreseeable future, and businesses will need to adapt around that reality.

What about the RAM consumption trend of the last 10 years? I think it is very feasible to produce phones with the same amount of RAM as was the norm 10 years ago. The only compromise would be using older algorithms and features that consume less of it, and to take a bit of effort on keeping an eye on memory consumption in the development phase. There's a lot of opportunity. We can even leverage AI these days to optimize existing software for RAM usage.

"The reason why RAM has become four times more expensive is that a huge amount of RAM that has not yet been produced was purchased with non-existent money to be installed in GPUs that also have not yet been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that may never appear, to satisfy demand that does not actually exist and to obtain profit that is mathematically impossible."

The current crunch and constrain in compute is great time to show once again some ingenuity. The lowest level of smartphones from couple of years ago have more computing power than XBOX 360. That should be enough to run Whatsapp smoothly.

Maybe it is time not just shrink transistors but also software bundles. I can see decades of possible progress hiding in plain sight behind a browser screen.

The people responsible for throwing the consumer segment under the bus will see no repercussions for this - nobody is going to jail, consumers get screwed.

The memory prices will never return to the pre-shortage prices. They’ll use this as an excuse to subtly increase the base prices and the orgs that sell the devices will also do a similar increase, also past what the BOM changed by.

I'm not one of those people who chases all the new great things. I wait until things wear out or become completely obsolete before upgrading. I just get comfortable doing things the same way every day and see no reason to waste money on SaaS shit or anything else wastes my time or money.

I think the memory shortage will present opportunities for those willing to take advantage of the situation. A lot of DRAM is going into GPUs for data centers in AI work. Those units have a limited lifetime online and they will be rotated out and replaced with new units as performance degrades. I think this will be a lot like Li-ion batteries in that many of these GPUs will be perfectly fine for home pcs or small business workstations or for other less intensive use cases and the RAM will be performant enough that a viable recycling industry should arise from this AI buildout.

Funny enough, one day the local AI noise-making, power-wasting, water-wasting data centers will be the best places to score high-tech components and many of us will have one right down the road. That should set a lot of people up as recyclers redistributing reconditioned components to those who build their own systems.

For anyone who doesn’t follow the market closely, this is about a good a primer as you could hope for.

Any prediction on when it'll end? Can Chinese companies scale up to scare the big 3 into increasing capacity or lose price control?

"Because of all the peripheral circuits and all the vertical channels, a single gigabyte of HBM consumes more than three times the wafer capacity that a gigabyte of DDR or LPDDR does. Every gigabyte of HBM produced is, in effect, three gigabytes of commodity memory not produced."

Is this true or conflating wafer area with process steps?

If the bottleneck isn't process steps but wafer area, you'd expect the semiconductor boule growers to rake in all the profit, not the memory makers themselves (although they would see more revenue). I'm not hearing about Umicore or other semiconductor corps raking in big under a wafer shortage.

Thats why I ask the question if this is actually correct.

If correct, it really makes one wonder what would be more productive for machine learning community X PB of HBM vs 3X PB of conventional (LP)DDR RAM, or somewhere in between. By giving a couple of big players carte blanche financially, has allowed the decision to divide in 3 global memory capacity (in units of terrabytes manufactured per day say)? The singularity that slows itself down?

Poco X3 has excellent support for Ubuntu Touch, PostmarmetOS and LineageOS etc.

I think there's a reason that University of Michigan researcher jumped after that ferroelectric breakthrough.

I am hopeful that some of this massive pressure is going to push us into the next paradigm, which is likely some kind of "true" compute-in-memory system. That may increase performance and efficiency for AI by up to 100 X.

My 4090 and 12900k are gonna have to last till 2029 at this rate won’t they…

>In 1985, if you were a reasonably affluent American, the best computer that you could afford was the IBM PC AT. The PC AT would cost you about $6,000—$19,400 in 2026 dollars—and thus represented about a quarter of the median American’s annual income;

In early 1980s a PC AT was SOTA. I guess that if you buy a SOTA computer today it might cost you close to $19,400.

my 2 cents

As a complete know nothing about the fab industry, I am always puzzled by this. Do the fabs need to be seasoned like an iron cast skillet or something?

I have a 14900KS, a CPU from 2 years ago and is still amongst the strongest in benchmarks like Cinebench. And 128 GB DDR4 RAM. I feel like I can wait a few more years to upgrade.

I hate it that we had decades of progress to have computers become a very expensive hobby because some dudes high on fentanyl think some text prediction model that destroys the planet is worth a trillion dollars.

From 2008 to around 2015, upgrading every two years could make a meaningful difference. From ~2015 to ~2020 upgrading every three years might be worth considering. I just upgraded my top of the line flagship after nearly six years. And I actually looked for compelling reasons to buy a new phone every year since 2023. There just weren't any.

Frankly, this latest flagship phone is pretty underwhelming. It's slightly faster at a few things. The battery lasts a little longer. The screen can get a little brighter. The camera is supposed to a little better. But those are just the claimed improvements. I haven't actually noticed any of them in daily use because they weren't issues with my 2020 flagship phone either. Otherwise, the new phone is almost exactly the same size, same weight, same resolution, same look and same capabilities. I only upgraded because I was long out of contract and it was a only a couple hundred bucks for a $1400 MSRP phone with a new contract and a trade-in of the old phone.

China has an advantage here, once domestic DRAM production finally gets going. DRAM policy can be set strategically. China's economic planners may choose to provide DRAM to domestic manufacturers rather than export parts, even if exporting parts would be more profitable in the near term. That's already being done in raw materials. Conversely, if external suppliers have lower prices, there may be a policy decision to buy domestically to keep the domestic manufacturers going. Done with the goal of leveling production, this can work. Done stupidly, it becomes a money drain, of course.

Probably China controls the DRAM market around 2030 or so.

I would have thought in the coming years they'll build more chip plants or the AI bubble will crash or some such. If there's an effect it should be greatest now.

Pics:https://duckduckgo.com/?ia=images&origin=funnel_home_website...

Two other underappreciated handset brands are Doogee and Blackview. Gorgeous devices and solidly built. From what I recall they're friendly to root.