Top 7 Stories
Nvidia in Talks to Invest in Perplexity at $30 Billion-Plus Valuation Reuters and The Information report that Nvidia is discussing an investment in AI search startup Perplexity at a valuation above $30 billion. A deal would extend Nvidia’s strategy of backing fast-growing AI applications — following earlier bets on companies that consume its GPUs at scale — and give Perplexity a powerful ally ahead of a rumored funding round.
For Nvidia, the move pairs with a landmark week: Q2 earnings land Tuesday and will test Wall Street’s faith in the AI boom. The Perplexity talks signal that Nvidia sees value beyond selling chips — in the applications and agents those chips power.
OpenAI Urges California to Strengthen Its Newly Passed AI Safety Law OpenAI is publicly asking California to tighten the AI safety law it recently enacted, per Mashable and TechCrunch. Rather than opposing regulation, the company is pushing for stronger provisions — including more robust oversight of frontier models and clearer obligations for developers.
The stance marks a notable policy posture for OpenAI, which has often been cast as a skeptic of binding rules. By endorsing tougher standards, it may be seeking to shape the regulatory template that other states — and Washington — are likely to copy.
Anthropic Hires Former Google TPU Chief Amir Salek, Signaling In-House Chip Ambitions Anthropic has brought on Amir Salek, the former leader of Google’s TPU program, according to multiple reports. The hire is widely read as a step toward designing its own AI accelerators, reducing reliance on Nvidia and external cloud suppliers as training costs balloon.
In-house silicon is the frontier-lab endgame: OpenAI, Google, and Meta all pursue custom chips, and Anthropic now appears determined to join them. The move could reshape its cost structure and bargaining power with cloud partners.
Meta Hires OpenAI Veteran Luke Metz Axios reports that Meta has hired Luke Metz, a longtime OpenAI researcher, to work on its AI efforts. Metz is known for work across large-scale model training and systems — a profile Meta has been aggressively recruiting as it races to close the gap with OpenAI and Anthropic.
The hire continues a pattern of top-tier talent flowing between frontier labs. For Meta, it bolsters the research bench behind Llama and its consumer AI push; for OpenAI, it highlights the intensifying competition for elite researchers.
Nvidia Minicomputers Are Powering Moscow’s AI Drones, NYT Finds The New York Times reports that small Nvidia computing modules have been found inside AI-guided drones used by Moscow’s forces, despite export restrictions. The components reportedly reached Russia through intermediary supply chains, underscoring the difficulty of enforcing chip controls.
The findings add urgency to the export-control debate just as Taiwan reportedly indicted Nvidia employees over shipments of prohibited AI servers to China. Hardware traceability is becoming a first-order national-security issue for the entire AI supply chain.
Taiwan Reportedly Indicts Nvidia Employees Over Prohibited AI Server Exports to China Engadget reports that Taiwanese prosecutors have indicted Nvidia employees accused of helping export AI servers to China in violation of export bans. The case is one of the first criminal enforcement actions targeting individuals in the AI hardware chain rather than just companies.
If confirmed, the indictments signal a tougher enforcement era for AI export controls, with personal liability extending deep into vendor and partner networks. Companies across the industry are likely to tighten compliance and channel screening in response.
Frontier AI Labs Still Won’t Say How They’d Contain a Rogue Model TechCrunch reports that leading AI labs remain unwilling to disclose concrete plans for containing a rogue or out-of-control model — even as agentic systems take more autonomous actions. Internal safety documentation, where it exists, is vague on triggers, escalation paths, and kill-switch mechanisms.
The opacity is drawing scrutiny from policymakers, especially as incidents of unexpected agent behavior multiply. Calls are growing for standardized, externally verifiable containment commitments before frontier models are deployed at scale.
Trend Watch
| Story | Impact | Why it Matters |
|---|---|---|
| Nvidia’s Perplexity investment talks | Deepens chipmaker’s reach into the AI app layer | Nvidia is investing in the GPU consumers that drive its own growth |
| OpenAI backs stronger California AI law | Sets a pro-regulation precedent for frontier labs | State rules are becoming the de facto national template |
| Anthropic hires ex-Google TPU chief | Signals a shift toward in-house silicon at frontier labs | Custom chips could rebalance cost and leverage vs. Nvidia and clouds |
| Export-control enforcement intensifies | Adds legal risk across the AI hardware supply chain | Criminal cases could reshape how chips and servers reach markets |
What to Watch
- Nvidia Q2 earnings (Tuesday): guidance, HBM supply, and China commentary will move the whole AI complex — and any Perplexity deal news could surface on the call.
- California’s AI safety law: watch for OpenAI’s specific proposals and whether other labs join the call for stronger rules.
- Anthropic’s chip roadmap: early signals on silicon partners, tape-outs, and timeline will determine how serious the in-house push is.
- Export-control fallout: follow the Taiwan indictments and NYT findings for new enforcement actions and compliance shifts.
- Rogue-model containment: pressure is building for labs to publish verifiable kill-switch and escalation protocols.