Alphabet raises $80B for AI, as the IPO window cracks open

Alphabet is raising roughly $80bn through equity to fund AI infrastructure, in one of the largest such fundraisings on record. CNBC pins the structure: "The offerings consist of $30 billion in concurrent underwritten public offerings and a $40 billion at-the-market offering program, plus a $10 billion private placement to Berkshire Hathaway split between Class A at $351.81 and Class C at…

Anthropic faces a spending backlash right before its IPO

Axios

Anthropic filed to go public just as corporate America hits its AI sticker-shock phase. The catch Axios flags: companies are Anthropic's biggest customers, so the people deciding whether to keep paying are the same ones underwriting the IPO story.

SpaceX, Anthropic and OpenAI IPOs could add $4T within months

The Economist via Techmeme

The Economist tallies the three pending listings at up to $4 trillion in new public-market value, and warns that scale could pull even more capital-raising in behind it.

Zhipu AI plans a Shanghai listing on an $83B market cap

Reuters via Techmeme

Zhipu's Hong Kong shares are up more than tenfold since its January IPO, and it now wants a Shanghai listing too — a reminder the IPO rush isn't only a US story.

The prospectus itself

SEC EDGAR

Alphabet filed its 424B5 prospectuses on June 2. If you want the raise in primary-source form rather than the headline, this is the filing.

Chinese defense-linked universities are seeking Nvidia H200s

Bloomberg via Techmeme

Procurement records show at least seven Chinese universities tied to the country's military and defense industry are seeking access to Nvidia's H200 chips — the export-control boundary getting tested in public.

ARM names Oracle and ByteDance as data-center CPU customers

Reuters via Techmeme

At Computex, ARM CEO Rene Haas listed Oracle and ByteDance among buyers of the company's new data-center chips — ARM moving further up the stack into the gear the labs actually run on.

China adds data and algorithms to its trade-secret rules

Bloomberg via Techmeme

Beijing is folding data and algorithms into its trade-secret protections, part of an effort to stop tech leaks as competition with the US sharpens. The raw material of AI is now treated as a guarded secret.

Altman on Stargate, and what's actually driving demand

CNBC via Techmeme

In a CNBC interview, Sam Altman talks through OpenAI's Stargate data center in Saline, Michigan, and names coding models as the single biggest driver of AI demand.

Office workers, not just developers, are adopting Codex

Axios

Per a new OpenAI report, knowledge workers are now about a fifth of Codex users and are growing more than three times as fast as developers — a coding agent turning into a general work tool.

Cloudflare's Agents SDK ships durable workflows and skills

@whoiskatrin

Version 0.14.0 adds skills, messengers, schedules, and durable workflows on Cloudflare. It also brings recurring tasks, chat recovery, and MCP support — the pieces an agent needs to survive a restart and keep running.

“you can now build agents with skills, messengers, schedules, and durable workflows on cloudflare out of the box support for recurring tasks, think workflows, chat recovery, mcp”

Intel and Perplexity demo hybrid local-plus-cloud inference

@intel

At Computex, Perplexity's Aravind Srinivas showed orchestration that keeps sensitive data on-device while the cloud adds scale and context — the split-inference pattern getting a public demo.

“Keeping sensitive data on device while cloud AI adds scale and context, @perplexity_ai @AravSrinivas demonstrates hybrid local server inference orchestration at Computex.”

Florida sues OpenAI and Sam Altman over safety claims

NBC / Today

Florida is suing OpenAI and Altman, accusing the company of putting profit over safety. The legal exposure around frontier products is moving from think-pieces to dockets.

Sampling research directions no human community would propose

arXiv

The paper breaks ~7,500 recent NeurIPS, ICLR, and ICML papers into reusable 'idea atoms', then trains a coherence model and an availability model to find directions that are logical yet unlikely to be proposed. It validates the reconstruction step; it doesn't show the sampled directions yield real results.

“By fusing a coherence model and an availability model, the system samples 'alien' research directions that are logical but highly unlikely to be proposed by human scientists.”

What sits outside any existing community's reach

@gravity7

Adrian Chan frames the same paper's question well: large language models recombine familiar ideas, but coherent research directions no existing community is positioned to explore are a different kind of target.

“LLMs recombine familiar ideas — but what about coherent research directions no existing community is positioned to explore?”