AI This Week: Distillation Wars Go Geopolitical
This was the week AI stopped being a product story and became a foreign-policy one. A White House official publicly accused a Chinese lab of copying Anthropic’s frontier model, OpenAI put more than $30 billion behind a single data center campus, and four separate enterprise “agent platforms” landed within weeks of each other. The through-line: capability is now cheap enough, and strategically important enough, that governments are treating model weights the way they once treated enriched uranium.
Model Releases & the Distillation Fight
White House accuses Moonshot of distilling Anthropic’s Fable model
OSTP Director Michael Kratsios publicly alleged that Moonshot AI distilled Anthropic’s Fable model to build Kimi K3 — a 2.8-trillion-parameter model claiming a 76% win rate on Frontend Code Arena and an 88.3 Terminal-Bench score, with open weights due July 27. Kratsios called it “large-scale covert industrial distillation.” It’s the first time a US official has named a specific model as stolen IP.
Independent analysis adds fuel to the claim
Redwood Research chief scientist Ryan Greenblatt published a cross-entropy analysis finding that Kimi K3 identifies itself as “Claude” disproportionately often across many prompts — the kind of statistical fingerprint that distillation tends to leave behind. Kratsios separately alleged Moonshot accessed export-restricted Nvidia GB300 chips through servers in Thailand.
DeepSeek V4 and Kimi K3 land within days of each other
The Chinese frontier is shipping fast and cheap: DeepSeek V4’s stable release arrived July 24 at $0.44 per million output tokens, with Kimi K3’s open weights following July 27. Both continue to narrow the capability gap with US labs at a fraction of the cost — the exact dynamic that made this week’s accusations so charged.
Agentic AI
OpenAI launches Presence, its enterprise agent platform
On July 22, OpenAI shipped Presence, a platform for deploying agents into customer support, sales, and other high-risk workflows, with BBVA, SoftBank, and IAG as early adopters. It joins a suddenly crowded field: Google Gemini Enterprise, Meta’s Business Agent Platform, and the NVIDIA/ServiceNow Project Arc all launched in the same window. The enterprise agent war is officially on.
Infrastructure & Business
OpenAI breaks ground on Project Camellia, a $30B+ data center
Announced July 22, Camellia is a 3.2-gigawatt campus across 1,400 acres in Effingham County, Georgia, with spending exceeding $30 billion and power delivery staged from 2028 to 2032. OpenAI paired it with $80 million in community benefits plus $71 million in Codex credits — a reminder that the constraint on frontier AI is now electricity and concrete, not algorithms.
BlackRock and MGX pour another $5B into data centers
BlackRock and Abu Dhabi’s MGX committed an additional $5 billion to Aligned Data Centers on top of a $40 billion acquisition, with total potential deployment reaching $100 billion including debt. Meanwhile, the Albany NanoTech Complex received components of a $400 million ASML High-NA EUV lithography system expected to be operational by year-end.
A federal judge approves Anthropic’s $1.5B copyright settlement
A San Francisco federal judge signed off on Anthropic’s $1.5 billion settlement with authors over pirated books used to train Claude. Over 91% of eligible authors and publishers filed claims, with $101 million set aside for attorney fees — one of the largest resolutions yet in the AI-training copyright fights.
Defense AI clears $3B in disclosed July funding
Defense-focused AI attracted well over $3 billion in disclosed funding this month alone, led by Shield AI’s $1.5 billion Series G and Helsing’s earlier €1.8 billion round, alongside a new Anduril–Archer partnership. Capital is following the same geopolitical logic driving the week’s headlines.
Science & Healthcare
AI drug discovery investment tops $2B as timelines collapse
AI-driven drug discovery pulled in more than $2 billion in recent investment, with technology cutting development timelines from 4–5 years to 12–18 months while roughly doubling clinical success rates. About 175 AI-originated drug programs have entered human trials since 2019, and 15–20 could reach pivotal Phase III this year. The FDA is expected to finalize guidance on AI in drug development during 2026.
Policy & Regulation
The US and China schedule formal AI talks for September
Washington and Beijing are preparing formal AI negotiations in September — before President Xi Jinping’s planned September 24 US visit — covering military AI, cyberattacks, model access, and open-weight releases. After a week of distillation accusations and chip-smuggling allegations, the diplomatic channel suddenly looks less like a formality and more like a pressure valve.
If there’s a single takeaway, it’s that the interesting AI questions are no longer just “which model scores highest.” They’re “who trained on what,” “who’s allowed to buy which chips,” and “who pays for the gigawatts.” The benchmarks are converging; the politics are diverging.