This was the week the AI industry went to the market for money at a scale it has never asked for before. Anthropic filed for an offering its bankers are pitching at roughly $100 billion, on a valuation near $2 trillion, and the filing reportedly names public backlash as a risk factor. Alibaba raised $10.2 billion at a discount and watched its shares fall 8%. Meanwhile OpenAI put its largest planned training run on hold over its own model's cyber capabilities, Alberta's regulator killed the power plant behind a $10 billion data centre, and Pew found that most Americans are now more worried about AI than excited by it. The capital and the consent are moving in opposite directions.
1. Anthropic filed for what could be the largest IPO ever attempted
Anthropic confidentially filed for a US listing and its bankers have told potential investors the offering could raise around $100 billion at a valuation near $2 trillion, which would surpass SpaceX as the biggest market debut on record, with Citigroup joining the underwriting bank syndicate and shares possibly trading in the autumn. The detail that matters most to anyone tracking sentiment is what the filing reportedly names as a risk factor: public backlash against AI itself, over data centres and job losses.
2. Alibaba raised $10.2 billion for AI and its shares fell 8%
Alibaba priced a $10.2 billion share placement at a sharp discount to fund its AI build-out, and the stock opened down roughly 8% in Hong Kong as investors questioned when the spending starts paying back. Days later the company launched Wan3.0, a new AI video model — the clearest test yet of whether public markets will keep funding AI capital expenditure on faith.
3. OpenAI put its largest planned training run on hold
OpenAI paused reinforcement learning training on its newest deployment-bound model for two weeks and said its largest planned frontier run remains on hold entirely, after internal evaluations found its upcoming Astra model may cross the "Critical" cybersecurity threshold in the company's own Preparedness Framework — the level at which a model can find and weaponise zero-day vulnerabilities without human involvement. Sam Altman said publicly that model progress is now outpacing the company's safety and monitoring standards.
4. OpenAI asked California to strengthen the AI safety law it once opposed
OpenAI publicly called on California lawmakers to expand SB 53, the state's frontier AI safety law — a reversal for a company that lobbied against the bill before its passage, and a shift that arrives days before the legislature adjourns on August 31. For anyone buying AI, the practical read is that model developers are starting to compete on governance credentials, which makes compliance posture a procurement question rather than a legal footnote.
5. Nvidia is raising AI server prices 15%, and the same squeeze is in your laptop
Nvidia has told its largest customers — reportedly including Microsoft, Google and Oracle — that servers built on its Grace Blackwell and Vera Rubin chips will cost more than 15% more in many configurations from early next year, because memory has become one of the most expensive ingredients in an AI server. The same shortage is repricing ordinary hardware: DRAM has gone from about $0.90 per gigabyte in August 2024 to $3.45 in July, and Apple, Amazon and Nintendo have all raised prices citing memory costs.
6. Nvidia paid $6 billion to start building its own open models
Nvidia agreed to pay $6 billion to license Poolside's "Model Factory" software, invest a further $1 billion in the startup at a $12 billion pre-money valuation, and hire 109 of its engineers to work on Nvidia's open-weight Nemotron models. The chipmaker's stated aim is a Western open-weight model competitive with DeepSeek and Kimi K3 — which also puts the industry's dominant supplier into direct competition with the frontier labs that are its biggest customers.
7. Stripe is buying OpenRouter for about $8 billion
Stripe finalised its acquisition of OpenRouter, the gateway that routes and meters API calls across hundreds of AI models, in a deal reported at roughly $8 billion — one of the largest AI acquisitions of the year and a return of close to $1.5 billion for Andreessen Horowitz alone. A payments company buying the meter for AI consumption is a bet that model spend becomes a line item every finance team has to reconcile, the way cloud spend did.
8. Alberta's regulator killed the power plant behind a $10 billion data centre
The Alberta Utilities Commission denied Synapse Real Estate's application for a 1.4-gigawatt gas plant in Olds, a town of 10,000, finding the site "simply not suitable" and granting standing to more than 900 residents who opposed it. Without the plant there is no data centre, and the province's technology minister called the rejection proof the system works — a reminder that in Canada the binding constraint on AI infrastructure is increasingly local consent, not capital.
9. Most Americans are now more worried about AI than excited by it
New Pew Research polling found 52% of Americans say they are more concerned than excited about AI's growing role in daily life, up from 37% in 2021, with young adults increasingly worried it will take jobs. Adoption has kept climbing through the same period — which means widespread use and public acceptance have decoupled, and companies deploying AI publicly should expect to explain themselves.
10. Purdue now requires every undergraduate to study AI to graduate
Starting with this autumn's incoming class, Purdue University requires AI-focused coursework in every undergraduate degree programme, not just technical ones, as part of its AI@Purdue framework. Northwestern and Wentworth are launching dedicated AI majors alongside it — the first cohort of graduates who arrive already fluent is roughly four years out, which is the real timeline for anyone planning workforce capability.
