Capital markets flinch at AI costs as OpenAI misses its raise and Google's AI Mode dents Search.
The money side of AI took three hits at once. WSJ reports OpenAI raised less than half of its $30B target. Google admitted AI Mode is cutting into Search revenue, and Latent Space puts the drop at 15%, about $500B, the worst since its 2004 IPO. Blackstone told the FT that Anthropic may not earn an investment-grade rating after it goes public. US tech stocks also fell on AI cost fears (BBC). The market is now asking whether AI products can pay for the infrastructure built to run them.
The WSJ reports OpenAI secured well under half of the $30B it sought in its latest round. The shortfall points to investor caution or disagreement over valuation.
OpenAI's compute commitments assume capital keeps flowing at this scale. A miss on the industry's flagship round resets expectations for every lab raising behind it.
Latent Space reports that Google acknowledged its AI Mode in Search cannibalizes the revenue it was built to protect. The company's drop of roughly 15%, about $500B, is its worst since the 2004 IPO.
This is the first big-incumbent admission that an AI-native product monetizes worse than the one it replaces. Every ad-funded AI search product now has to answer for it.
The FT reports that Blackstone projects Anthropic may not secure an investment-grade credit rating after listing, citing its economics. The comment treats an Anthropic IPO as a given.
A sub-investment-grade rating means pricier debt for a company that needs heavy compute spending. It is a major credit investor questioning frontier-lab unit economics in public.
Multiple reports, including DoublePulsar and Phorion, describe a Claude Code VS Code extension on the Cursor marketplace containing a backdoor. DoublePulsar says it delivered a remote access trojan to developer laptops.
Developer machines hold source code, API keys, and cloud credentials, so a trojan there is a full compromise. Teams should audit installed Claude Code extensions now and treat third-party marketplaces as untrusted supply chains.
Cohere is committing C$4B to Canadian AI expansion, covering research infrastructure and more than 1,500 jobs through 2031.
It stakes Cohere's position as Canada's sovereign-AI champion, a capital-light alternative to US hyperscaler dependence. The commitment also lands in a week when US lab fundraising looks shaky.
OpenAI will respect robots.txt signals that let websites opt out of having their content used for model training.
A frontier lab is adopting a concrete, machine-readable opt-out, which pressures competitors to follow. OpenAI also published court briefs arguing that fair use covers AI training, so it is offering an opt-out while defending the legality of training without one.
OpenAI's undersubscribed round, Blackstone's view on Anthropic's credit, Google's AI Mode drop, and the tech-stock selloff all point the same way.
The Cursor marketplace backdoor hit Anthropic's developer tooling while Anthropic reportedly tightened terms on game rewrites.
Cisco handed an engineer unlimited tokens for legacy rewrites, OpenAI matched Anthropic's API credits for subscribers, and Spawn and YC Labs ran autonomous agents in live game worlds.
SkyPilot's Grace Blackwell scheduler yields 27% more usable GPUs from the same racks, which matters when capex is the market's central worry.
Agent swarms playing Minecraft and Mindustry make a good demo but say little about reliability on production work.
Next week, watch for Cursor's and Anthropic's disclosure on the backdoor's scope, final terms of OpenAI's round, and any Anthropic IPO filing detail on debt and margins.