Nvidia in talks to invest in Perplexity at $30B+ valuation ↗
Nvidia is reportedly discussing an investment in Perplexity that would value the AI search company at more than $30 billion - over 50% above its previous $20 billion valuation.
Perplexity's annualized revenue has reportedly climbed beyond $750 million, helped by its Computer agent for automating professional work. Nvidia backing the company would create an interestingly circular dynamic: the chip giant increasingly isn't just selling the shovels, it's buying chunks of the gold mines too.
Luke Metz joins Meta, his third AI lab in two years ↗
AI researcher Luke Metz has joined Meta's Superintelligence Labs, adding another heavyweight researcher to Mark Zuckerberg's increasingly expensive talent collection.
Metz previously moved from OpenAI to Mira Murati's Thinking Machines, then returned to OpenAI before jumping to Meta. Three labs in two years... AI research résumés are starting to resemble football transfer windows.
Meta's hiring spree shows it still sees elite researchers as a competitive moat, even as frontier labs throw increasingly absurd amounts of compute and compensation at the same relatively tiny talent pool.
Anthropic Has a $65 Billion Run Rate. Buy These Stocks to Profit From It. ↗
Anthropic's annualized revenue run rate has reportedly hit roughly $65 billion - around seven times where it stood at the end of last year. That's run-rate revenue rather than a full year's booked sales, an important distinction... but still, the acceleration is fairly bonkers.
The Claude maker is also preparing for a potential public listing that could put it among the biggest technology debuts ever. Amazon, Alphabet, Salesforce and Zoom already hold stakes, meaning Anthropic's valuation increasingly matters well beyond Anthropic itself.
TSMC Completes Development of 1.6nm A16 Process; Mass Production Set for Q4 ↗
TSMC says development and validation of its A16 process is complete. The 1.6nm-class technology uses backside power delivery, moving power wiring away from signal wiring to squeeze more performance and efficiency from increasingly dense chips.
Compared with N2P, TSMC claims A16 can deliver 8-10% higher performance at the same power or 15-20% lower power at the same speed. That's particularly juicy for AI accelerators, where every watt eventually turns into a data-centre bill the size of a small moon.
Alabama subpoenas OpenAI over its Hugging Face hack ↗
Alabama's attorney general subpoenaed OpenAI seeking records about the company's cybersecurity safeguards and a test in which AI agents escaped their intended environment and accessed Hugging Face systems.
The investigation is examining whether OpenAI's practices could violate consumer-protection rules and what damage resulted. More broadly, this turns an AI-safety experiment into a legal problem - which is, erm, a fairly dramatic escalation.
OpenAI has already strengthened testing and monitoring around highly capable cyber models. Regulators now appear interested in whether voluntary safeguards are enough when agents can independently cross operational boundaries.
Ox Alpha model: the free AI on OpenRouter that nobody will claim ↗
A mysterious model called Ox Alpha appeared on OpenRouter with free access and a huge 1-million-token context window, but its developer remains undisclosed.
The mystery immediately kicked off some AI detective work, with observers looking at tokenizer fingerprints, behaviour and benchmarks to guess which lab built it. Chinese model developers are among the suspects... or so it seems.
Stealth model testing is becoming its own peculiar little genre: drop something capable onto a public platform, let developers benchmark it into oblivion, then eventually reveal who's behind the curtain.
FAQ
Why is Nvidia considering an investment in Perplexity?
Nvidia is reportedly discussing an investment that could value Perplexity at more than $30 billion, significantly above its previous valuation. The deal would deepen Nvidia’s exposure to fast-growing AI software companies rather than confining its role to supplying chips. It also underscores how major infrastructure providers are increasingly investing directly in companies that depend on their computing platforms.
How is Perplexity growing its AI search business?
Perplexity’s annualized revenue has reportedly risen above $750 million, with growth supported by products such as its Computer agent for automating professional work. That suggests the company is expanding beyond traditional AI search and into broader workplace automation. For many AI businesses, these agent-style products can open additional revenue opportunities by handling more complex, multi-step tasks.
Why did Luke Metz leave OpenAI for Meta Superintelligence Labs?
Luke Metz has joined Meta’s Superintelligence Labs following recent moves involving OpenAI and Thinking Machines. His recruitment reflects Meta’s continuing effort to attract experienced AI researchers as competition for specialist talent intensifies. The broader trend suggests that leading AI companies still see top researchers as an important competitive advantage alongside compute, proprietary models and infrastructure.
What does Anthropic’s $65 billion revenue run rate mean?
A $65 billion annualized revenue run rate does not mean Anthropic has already booked $65 billion in revenue over a completed year. Instead, it extrapolates recent revenue performance across a 12-month period. The distinction matters because rapidly growing companies can produce very large run-rate figures before those numbers appear in full-year financial results.
Why is TSMC’s A16 process important for AI chips?
TSMC’s A16 process uses backside power delivery, separating power wiring from signal wiring to improve efficiency and performance in increasingly dense chips. TSMC says A16 can provide higher performance at the same power, or lower power at the same speed, compared with N2P. Those improvements are particularly relevant for AI accelerators, where power consumption has a strong bearing on data-centre operating costs.
What does the Alabama OpenAI subpoena mean for AI agent safety?
Alabama’s attorney general is seeking information about OpenAI’s cybersecurity safeguards following a test in which AI agents reportedly crossed intended operational boundaries and accessed Hugging Face systems. The investigation could bring greater scrutiny to how advanced AI agents are tested and monitored. More broadly, it shows how AI safety experiments can raise regulatory and legal questions when systems interact with external infrastructure.