One shortage, everywhere you look
Start with the story that touches all the others: the memory chip crunch. The RAM and storage that every phone, laptop, and server depends on is in shorter supply than the industry expected, and analysis from Deloitte suggests the squeeze — nicknamed "RAMaggedon" — could drag on until 2030. That is not a quarter of turbulence. That is the rest of the decade.
The knock-on effects showed up today in places you would not immediately connect to a DRAM spot price. Nvidia's next-generation Rubin Ultra has reportedly been scaled back on specs, and the reason floated in a report from odaily.news is memory. When the constraint reaches the flagship product of the most in-demand chipmaker on earth, it has stopped being a supply-chain footnote.
It is also reaching your wallet. A Yahoo Tech report says Nvidia GPU prices could climb by as much as 30%, and the headline framing is blunt: AI's bill is coming due at checkout. Consumer graphics cards have spent this cycle competing for capacity against data center buildouts. Shoppers are now the ones absorbing the difference.
Nvidia's problem is that everyone wants everything
The demand numbers remain difficult to comprehend. Nvidia is sitting on roughly $500 billion in bookings for its AI chips covering 2025 and 2026, according to a report published on MSN. Wedbush analyst Dan Ives, cited by Startup Fortune, puts demand ahead of supply by something like 12 to 1 — twelve buyers for every chip that exists.
Hold those two figures next to the Rubin Ultra report and the shape of the year becomes clear. Nvidia's ceiling is no longer set by how many customers it can win. It is set by what its suppliers can physically deliver, memory very much included.
Which makes the company's other move today read as strategy rather than press release. South Korea's SK Group and Nvidia are widening an existing partnership covering AI factories and memory, per the electronics trade publication Evertiq. When memory is the bottleneck, you get closer to the people who make memory.
The buildout, and who is paying for it
TSMC's $100 billion commitment to manufacturing in Arizona is increasingly being told as a jobs story, not just a capital story — the anchor figure in an argument about what advanced chipmaking means for American employment. Meanwhile Amazon's homegrown silicon has reached a $25 billion annual run rate, according to CFOtech Asia. Worth remembering that a run rate is a projection built from recent performance, not money already banked. Still, it is a serious number, and it signals that the biggest cloud buyers would rather design their way around Nvidia's queue than wait in it.
The froth is visible too. The Financial Times reports that James Dacombe, aged 25, has tripled the valuation of his AI chip start-up to $3.3 billion. Whatever you make of the fundamentals, the datapoint tells you how much capital is chasing anything with "AI chip" attached to it.
A reminder about physical risk
A magnitude 7.1 earthquake struck Kumamoto Prefecture in Japan's Kyushu region on July 28, Digitimes reports, putting a spotlight on TSMC's plant there alongside Sony and other manufacturers in the area. Geographic diversification is the industry's answer to concentration risk — but every new site comes with its own hazards, and Japan's fault lines are one of them. Consider it the first real stress test of a facility built precisely to be a backup plan.
And the long game
IBM CEO Arvind Krishna said quantum computing could begin making a measurable contribution by 2028 — a striking claim for a field that has spent decades as a science project with a business plan attached. The market appears to be taking the timeline seriously.
One closing note for anyone whose computer is used for fun rather than inference: Tom's Hardware reports that AMD's Zen 6 desktop processors may target stutter, the brief hitches that benchmark averages hide but every gamer feels. Small news on a day like today. Also the only item here that improves something you actually sit in front of.