The big three just teamed up

Rivals joined forces. OpenAI, Anthropic, and Google are now sharing threat intelligence through the Frontier Model Forum after accusing DeepSeek, Moonshot, and MiniMax of large-scale adversarial distillation. Anthropic documented 16 million exchanges from the three Chinese companies across roughly 24,000 fraudulently created accounts. U.S. authorities estimate billions in annual lost revenue; OpenAI flagged DeepSeek’s extraction methods to Congress back in February. Three companies that compete on everything else just agreed they share one problem.

The company at the center of that coalition is also growing fastest. Anthropic’s run rate hit $30 billion—up from $9 billion at the end of 2025, a 3x+ jump in one quarter. To feed that growth, Anthropic expanded its compute deal with Google and Broadcom; a Broadcom SEC filing references 3.5 gigawatts of new TPU capacity coming online in 2027. The recent $30 billion Series G valued the company at $380 billion. Revenue is tripling while the infrastructure bill keeps climbing—and those two curves don’t bend in the same direction forever.

Not everyone is defending their moat with coalitions. Meta shipped Muse Spark—its first proprietary frontier model—breaking from the open-source playbook that defined Llama. Muse Spark is natively multimodal (voice, text, image in a single forward pass), scores 58% on Humanity’s Last Exam in contemplating mode, and uses 10x less compute than Llama 4 Maverick for equivalent capabilities. It was built under Meta Superintelligence Labs with $115–135 billion in 2026 AI capex behind it. Separately, Visa launched Intelligent Commerce Connect—a network-agnostic payment platform built for AI agents, supporting 4 agent protocols with AWS, Highnote, Mesh, and Payabli as pilot partners and a broader rollout planned later in 2026. Meta is closing the weights; Visa is opening the wallets. The models are no longer just answering questions—they’re spending money.