AI Briefing

Today's theme is where the AI money and power are pooling. AMD just wrote Anthropic a check big enough to dent Nvidia's monopoly, while Alphabet's earnings show Gemini has quietly grown into a near-billion-user habit. Regulators are starting to notice: France warns that three companies already own the agent economy. And the reality check of the day is a cybersecurity model claiming a benchmark score four times higher than anyone has independently reproduced, plus Elon Musk promising a whole AI-generated Homer epic by New Year's.

01FundingCNBC ↗

AMD Bets Up to $5B on Anthropic and 2 Gigawatts of Its Own Chips, Aiming Straight at Nvidia

AMD announced a strategic partnership to supply Anthropic with up to 2 gigawatts of Instinct MI450 series GPUs, alongside an equity investment of up to $5 billion tied to deployment milestones. The first gigawatt of Helios rack-scale systems, built on MI455X GPUs and EPYC 'Venice' CPUs, starts rolling out in the first half of 2027. It's one of the clearest signals yet that Anthropic wants a serious alternative to Nvidia, and that AMD is willing to buy its way into the frontier-compute race.

02Industry9to5Google ↗

950 Million People Now Open Gemini Every Month as Alphabet's Cloud Revenue Jumps 82%

Alphabet posted Q2 2026 revenue of $119.8 billion, up 24% year over year, with Google Cloud surging 82% to $24.8 billion and a backlog approaching $460 billion. The Gemini app hit 950 million monthly active users with daily users tripling over the past year, and the models now process 22 billion API tokens a minute. It matters because it shows AI has moved from demo to genuine, revenue-scale infrastructure, even as Alphabet's capex hike spooked some investors.

03ResearchSakana AI ↗

Sakana's Fugu-Cyber Claims 86.9% on a Hacking Benchmark Where Rivals Score Around 20%

Sakana AI released Fugu-Cyber, a defense-focused endpoint on its multi-agent orchestration platform that presents as a single model while routing tasks across agents built on Claude, Gemini, and GPT. It claims 86.9% on UC Berkeley's CyberGym benchmark, which tests exploit generation across 1,507 real vulnerabilities. The catch: CyberGym's own creators found leading systems clearing only around 20%, and Sakana disclosed no methodology, trial counts, or scaffolds, making this one of the largest unexplained vendor-versus-independent gaps in recent AI.

04PolicyPPC Land ↗

France Warns Three Companies Already Control 84% of the AI Agent Market

The French Autorité de la concurrence published an opinion finding that OpenAI, Google, and Anthropic together hold more than 84% of the global AI agent market. It flags platformisation, disintermediation, and even algorithmic collusion as risks, and calls for enforcement on data access, interoperability, default placement, and open standards. It's notable as the regulator's first move downstream from model training into the deployment layer, where agents plan and execute tasks with little human oversight.

05ProductThe Hollywood Reporter ↗

Elon Musk Says Grok Imagine Will Generate a Full 'Odyssey' Movie Before Year's End

Escalating a running feud with Christopher Nolan's box-office hit, Elon Musk pledged that Grok Imagine will produce a full-length, 'historically accurate' AI-generated film of Homer's Odyssey before the end of 2026. He made the claim while reposting a fan-made three-minute AI clip of the epic on X, and floated giving Mel Gibson $100 million to make a version in Homeric Greek. It's a headline-grabbing test of whether generative video is anywhere near feature-length coherence, and the industry is skeptical.