$9 Billion to $65 Billion in Eight Months. That's Anthropic's Year.

Morning, folks. Anthropic just posted revenue numbers that are almost hard to believe, Nvidia's backing an OpenAI data center to the tune of $105 billion, and OpenAI's own president just admitted the cybersecurity clock is ticking. Let's get into it.

Anthropic's annual revenue run rate just surpassed $65 billion in July, according to Bloomberg, up from $47B in May and just $9B at the end of 2025. Investors expect it to keep climbing at roughly the same pace, landing between $100B and $120B by year's end. Meanwhile, rival OpenAI has doubled its own revenue to $40 billion, up from $20 billion at the end of 2025, still less than half of where Anthropic already stands.

Both are racing toward the public markets too, confidential IPO paperwork already filed on both sides. But Anthropic looks likely to get there first, possibly by fall, and it's not aiming small either, reports put its target valuation north of $2 trillion. If that holds, it'd be the biggest public debut ever.

OpenAI is teaming up with SB Energy and Nvidia to build a massive AI data center in Pike County, Ohio, on the site of a former uranium enrichment plant. The project will bring 8 gigawatts of IT capacity, with the first 800 megawatts expected online in 2028.

The real number is what Nvidia's guaranteeing: up to $105 billion in lease and power payments, plus a separate $1.5 billion investment straight into SB Energy. In return, Nvidia becomes the site's exclusive chip supplier. Jensen Huang already jumped on X to deny it's circular financing: "OpenAI will pay the lease." Whether that holds is the actual thing worth watching.

Greg Brockman just published a policy essay on last month's OpenAI-Hugging Face breach, where an AI agentic collective broke into both OpenAI's own research infrastructure and Hugging Face's production systems, mixing unknown security flaws with leaked credentials. He's calling it a "watershed moment," and buried inside it is a real warning: a competitor's about to release an open-weight model with near-frontier cyber capabilities by the end of this month.

His ask is blunt. Every security team needs AI defensive agents running now, not eventually. There's a narrow window where defenders can actually stay ahead of attackers, and once that model ships, the window starts closing fast.

Groq just raised $350 million at a $3.5 billion valuation, roughly half what it was worth a year ago. The drop traces back to December, when Nvidia paid Groq around $20 billion to license its chip technology and hired away founder and CEO Jonathan Ross along with key staff. Now Nvidia's back, joining this same funding round as an investor.

Groq has repositioned itself as a data center operator focused on AI inference rather than chipmaking. The round was led by Disruptive, the Dallas firm whose founder is now Groq's executive chairman.

Dharma AI tested a smarter way to schedule GPUs, weighing priority, workload type, and demand instead of just running jobs in the order they show up. Same hardware, zero new GPUs, just a better order. Best case: an 8-GPU setup went from 53.6% to 87.0% utilization, and output value more than doubled.

Ran it 7 times too, not a fluke. 6 out of 7 improved on utilization, and all 7 improved on value, even when they stripped priority out entirely. Kind of makes you wonder how much of the "GPU shortage" is actually just bad scheduling.