The Anthropic blackout has upended the calculus of AI leverage in ways that transcend the immediate export control. When the White House ordered Anthropic to revoke foreign access to Fable 5 and Mythos 5, it didn't just restrict a few users. It revealed that American AI dominance rests on a fragile social contract: foreign governments and companies tolerate U.S. control of the models they depend on only until they don't. The order has already triggered the response it was meant to prevent. DeepSeek closed a $7.4 billion funding round while Chinese labs slashed token prices up to 99 percent. Cohere reports drowning in government inbounds. The export control meant to protect America's AI lead is accelerating the very alternatives it sought to prevent. JPMorgan Chase and Goldman Sachs have already cut Claude access for Hong Kong staff. Meanwhile, Dario Amodei is telling G7 leaders to resist splintering over AI, but the splinter is already happening. The White House wants Anthropic to guarantee that Fable 5's guardrails cannot be circumvented before allowing rereleases, a demand security experts say is technically impossible. The policy is driving vendors toward other options and competitors toward capital markets that now see China's labs as safer bets than American companies subject to sudden revocation.
The real tension isn't between safety and capability. It's between tokenmaxxing and token costs. Uber blew through its annual AI budget in months. Meta killed its internal leaderboard. Some companies cut Claude licenses for entire departments. The enthusiasm that drove CEOs to push AI usage "as far as it would go" earlier this year has collided with bills that don't reflect the hype. Enterprises are still figuring out their AI ROI, according to NEA's Tiffany Luck, which is another way of saying they spent heavily on pilots that haven't scaled into value. AWS is now betting the bottleneck isn't code generation but release management, testing, and safety review. Databricks is pitching context layers and ontologies as the key to trusted agents. Pramaana Labs raised $27 million to bring formal verification to AI in law, drug discovery, and tax preparation, where errors have real costs. The market is correcting from hype toward specificity. Vendors are moving upstream from "write better emails" to domains where AI failures have measurable consequences and where enterprises will actually pay for reliability.
Hardware and infrastructure are where capital is flowing hardest. Odyssey, a world model startup, just hit $1.45 billion valuation with backing from Amazon and others. Canadian pension funds are acquiring stakes in Indian data center operators. SK Telecom was cut off from Claude, but the same geopolitical pressure driving data center investment in India also means the supply of compute will fragment by region, making local models and local infrastructure more valuable than centralized American platforms. Nvidia's robots are learning to install GPUs and cut zip ties through teams of AI coding agents, a signal that the company is solving the data problem for embodied AI at scale. Robot training data collection through vendors like XDOF is becoming a distinct business. Physical AI still trails LLMs in capability but capital is moving to solve that gap. The smart speaker market, dormant for years, is suddenly active again because Google believes Gemini makes conversational interaction valuable enough to justify a $99.99 device. The hardware layer is where the next wave of defensibility lives, not in models that can be turned off by government order.
Sloane Duvall