PostCards
Amazon Just Told Wall Street AI Spending Isn't Slowing Down
Amazon's board just answered the question Wall Street has been asking since chip stocks began bleeding value last week: is the AI infrastructure boom actually slowing down? CEO Andy Jassy told investors Thursday that full-year capital expenditures would climb to $220 billion, a $20 billion jump from prior guidance, with most of it earmarked for AI.
The number landed alongside similarly aggressive figures from Microsoft and Alphabet. Microsoft's Azure grew 43% and crossed $100 billion in full-year revenue for the first time, while Copilot passed 30 million paid seats — Satya Nadella framed the quarter as customers turning "tokens into business results."
The timing matters. A day earlier, a Fed-day repricing had wiped roughly $1.3 trillion off the world's most valuable chip stocks. Thursday's earnings didn't settle whether AI spending can keep outrunning near-term revenue — but they told the market the hyperscalers aren't backing off.
A British Neocloud Just Bought Its Way Into the Software Layer
Nscale, the London-based AI cloud company that grew out of a cryptocurrency-mining business in 2024, has agreed to acquire Anyscale in a deal Bloomberg pegs at about $1.65 billion. Anyscale's roughly 200 employees across the US, Europe and India will join Nscale when the deal closes in the second half of 2026.
The logic is margins, not headlines. "Most infrastructure providers just buy GPUs and rent them," said Nscale CEO Josh Payne. Anyscale's platform, built on Ray, spreads data preparation, training, fine-tuning and inference across an entire fleet — work the company claims can cut total cost of ownership by up to 90% against a patchwork of separate tools.
Ray itself isn't for sale — governance passed to the PyTorch Foundation in October 2025, and Nscale is committing to join that foundation as part of the deal, keeping the open-source project community-run even as its commercial arm changes hands.
Brussels Opens the Bidding on Seven AI Gigafactories
The European Union invited companies Thursday to bid for contracts to build up to seven AI "gigafactories" across the bloc, each packing at least 100,000 cutting-edge chips — roughly four times more powerful than any data center currently running in Europe.
"Access to the raw scale of computing power within AI gigafactories is a strategic necessity for Europe as AI development accelerates," said Henna Virkkunen, the Commission's tech-sovereignty chief. The site count grew from five to seven after strong interest from member states, and AMD, Nvidia and Qualcomm have all signed letters of intent to supply chips.
The sovereignty framing carries an obvious asterisk: the hardware underneath Europe's independence push is still almost entirely American. Bidding closes November 12, with winners announced in early 2027.
Chipmakers Just Lost $1.3 Trillion in a Week
Twenty of the world's most valuable chip stocks lost a combined $1.3 trillion since the previous Friday's close, according to a CNBC analysis, with Nvidia's $238 billion decline leading the way. SK Hynix, Samsung and Micron — the memory trio riding the AI boom hardest — lost $176 billion, $173 billion and $113 billion respectively.
"The sell-off reflects concerns that AI infrastructure spending may be peaking faster than expected," Forrester VP Charlie Dai said, though he called it "less about weakening AI demand and more about a repricing of expectations after an exceptionally strong rally."
The rout followed Alphabet's own decision to raise its 2026 capex forecast — proof, in one reading, that spending isn't slowing at all, and in another, that the market has stopped rewarding spend without a clear revenue story.
OpenAI Cuts GPT-5.6 Prices as Chinese Rivals Close In
OpenAI quietly cut prices across its GPT-5.6 family Thursday, dropping Luna — its most budget-friendly model — by 80% and trimming Terra, the mid-tier option built for everyday work, by 20%.
The timing is hard to separate from competitive pressure out of China. Moonshot AI's 2.8-trillion-parameter Kimi K3 took the top spot on a major coding leaderboard weeks earlier, and open-weight models trained on domestic chips have been undercutting Western labs on price ever since.
Sol, the flagship model in the GPT-5.6 family, keeps its original pricing. The discount is aimed squarely at the volume end of the market, where a few cents per million tokens decides which API developers default to.