Follow us
Add us as a preferred source on Google
OpenAI CEO Sam Altman recently declared that the RAM sitting in your computer is a “terrible” allocation of scarce resources, arguing that the world’s memory belongs in data centers. If you want to know exactly how much that mentality is costing you, just look at the apocalyptic state of the PC hardware market.
On April 1st, I predicted that a market correction would keep RAM prices down by August. Now it’s October, it’s clear that I was spectacularly wrong, and the irony of publishing that on April Fool’s Day is not lost.
You can see that in the gigantic asking prices for the MacBook Pro with PROLED XDR display, as Nvidia’s RTX 5080 is now at over $2,000, and as a 32GB kit of DDR5 is now at $500 instead of the $400 it was when I wrote before.
Latest Videos FromTom the guide
We are currently in the midst of a brutal quadruple threat: relentless AI-driven demand, historically low inventory, geopolitical tariffs and meltdowns that are shifting advanced wafer capacity away from consumer storage and into enterprise servers.
If you buy a PC now, you are effectively paying a “cloud tax” on your local hardware. But before you remortgage the house and cut anything, the tech landscape has changed violently.
Altman wants all the silicon, but the AI bubble is showing serious signs of bursting — and relief might actually be on the horizon.
‘Short Resources’
OpenAI CEO Sam Altman recently went on the Cheeky Pint podcast and gave an interview that practically destroyed my BS detector. Speaking to John Collison, Altman claimed that the memory sitting in computers is a poor use of it, arguing that local inference only uses that memory about 5% of the time.
Look up
According to Altman, DRAM belongs to data centers that run flat, making local AI a “terrible” allocation of scarce resources.
Let me just unpack that audacity. The guy whose company is arguably responsible for eradicating the global memory shortage—locking up massive amounts of the world’s DRAM supply for hyperscaler cloud servers—is now telling you that the memory in your own laptop is inefficient?
Consumers want local AI because it offers privacy, zero rate limits, and independence from cloud subscriptions.
He is not wrong about scarcity, but he is wrong about allocation. It’s like the CEO of Chili tells you that supermarkets and cooking at home is a terrible allocation of food.
As we have seen with the rise of RTX Spark laptops, among others packing a ton of unified memory, and the meteoric rise of tools like Hermes Agent and Open Claw, consumers want local AI because it offers privacy, zero rate limits, and independence from cloud subscriptions.
And some people can use their memory 5% of the time. But that’s free market capitalism. If there is demand, it is not up to you to decide what people do with it.
The canary in the coal mine
The first AI company is out of the starting gate. Anthropic’s new IPO filing is old news from a few days, but the more I read, the more I see a $2 trillion gold rush wrapped in a sci-fi horror warning.
On the surface, Anthropic is printing money, pulling in $16.3 billion in revenue in just the first half of 2026. But the cost of keeping this cloud infrastructure running is a financial black hole. Anthropic spent nearly $9 billion on AI infrastructure in just six months, contributing to a whopping $80.7 billion net loss. And that’s before you look at the company’s $800 billion spending on computers over the next 10 years.
Worse than the infrastructure burn rate are the existential threats that Anthropic has uncovered. To protect itself from lawsuits, the company has warned that its AI models can actively sabotage code, manipulate information and act in ways that resemble blackmail.
They literally warned the federal government that using AI to speed up their own research could cause the models to overcome human safety guidelines.
The silver color
OpenAI and Anthropic (and the others) want all the memory in their data centers, but financial markets around the world are starting to reject the unlimited spending needed to keep those servers running.
The AI bubble is starting to burst, and in China it already has. The mainland’s STAR50 index collapsed by 40%, wiping out about $480 billion in market value as AI leaders like MiniMax fell.
And now the contagion has hit American shores. The S&P 500 took a pretty violent nosedive, dragged down hard by the very tech giants that stockpiled the world’s silicon.
Wall Street is finally waking up to the reality that building massive AI models requires burning unimaginable amounts of cash, and the consumer gains simply won’t materialize fast enough to justify the burn rate.
But it’s not just Wall Street—it’s the company’s customers themselves. A growing wave of customers is pulling back on their AI spending as bills spiral completely out of control.
Furthermore, many enterprises cite privacy as the single biggest barrier to adopting cloud AI. They did what I’ve been encouraging you to do for a while: ditch the data center and build your own local AI tomb.
What does that mean for you?
So let’s continue these trends. If the AI hyper-scaler bubble really bursts, the fallout for consumer PC hardware will be massive. But the end of RAMageddon is not guaranteed to be a smooth landing.
Depending on how large silicon furnaces react to energy budgets and consumer setbacks, we are looking at two very different scenarios for your next build. And I want to be as clear as possible here (so as not to get stuck on a bet like last time).
The best case scenario: The big pivot
Currently, Samsung, SK Hynix and Micron prioritize enterprise memory over the standard RAM sticks you find in a PC. But as the S&P 500 craters, IPOs expose infrastructure liabilities and hyperscalers cut their server budgets, that vacuum of demand will shrink.
To keep their cash flow moving, foundries will have no choice but to pivot their manufacturing lines back to standard consumer memory. The market gets flooded with DDR5, prices drop back down to reality, and you can finally build that new rig.
The worst case scenario: Silicon freezes
Instead of pivoting back to consumer memory, meltdowns hit the panic button. If enterprise orders dry up, Samsung, SK Hynix and Micron can simply lower their overall wafer production to artificially limit global supply and protect profit margins. It’s a brutal tactic, and they’ve done it before.
In this scenario, the AI bubble bursts, but consumer RAM remains scarce, and we end up trapped in hardware stagflation: the broader technology economy is bleeding, but you’re still paying a 500% markup for basic memory.
So is the AI bubble bursting?
The short answer is yes, but there are nuances.
The blank-check era of AI hyperscaling is officially over. Between the bleeding S&P 500, big companies boycotting cloud APIs, and the Anthropic IPO exposing the apocalyptic infrastructure costs needed to keep these data centers running (data centers pushing back on local governments), the market is starting to correct itself.
But whether bursting that bubble actually saves your wallet is a whole other story.
The tech industry desperately needed a financial shock to stop the churning out of silicon, and we’re watching the start of that shock happen today. If your current rig can survive just a little longer, you may see some benefit in holding the line.
The AI gold rush is over, and Fallout is about to completely reset the PC hardware market. Whether it’s for better or worse, no one knows.
follow Tom’s Guide to Google News in the add us as a preferred source to get our latest news, analysis and reviews in your feeds.