The trial between Elon Musk and OpenAI has cracked open the industry's most profitable secret: everyone is copying everyone else's models, and the companies suing over it are doing the exact same thing. Musk testified under oath that xAI trained Grok using OpenAI's models through distillation, a technique he simultaneously claims is standard practice. OpenAI, facing this admission, responded by restricting access to its own cybersecurity tool GPT-5.5 Cyber to "critical cyber defenders" only, mirroring the exact access controls it criticized Anthropic for implementing on Mythos. The pattern is transparent. When a competitor limits your access, it's anticompetitive overreach. When you do it, it's security. The real fight is not about theft or ethics but about who gets to control the moat around frontier models before the gap between leaders and followers closes entirely.
Capital is flooding into legal AI with the kind of velocity that suggests the market itself is the product being built. Legora hit a five-point-five-billion-dollar valuation on a fifty-million-dollar extension led by Nvidia's venture arm, while Harvey, its rival, is matching it with dueling ad campaigns. Meanwhile, Anthropic is running a fundraise that could value it above nine hundred billion dollars, with investors submitting allocations within forty-eight hours. First-quarter 2026 venture funding in AI companies hit three hundred billion dollars across six thousand startups worldwide, a quarterly record. The speed matters more than the amount. Companies are being valued on growth trajectories, not profitability or even product-market fit. Legora and Harvey are both raising massive sums while competing for the same customers in a market where legal services are traditionally sticky and slow to change. The venture firms are betting that one will win and the other will become a footnote, but they are funding both anyway.
The infrastructure layer is quietly becoming the real leverage point. Cloudflare is giving AI agents full autonomy to create accounts, pay subscriptions, register domains, and deploy code with only initial human consent required. Stripe is building Link, a digital wallet that autonomous agents can use to spend money with approval flows. Google is embedding Gemini into millions of vehicles. Meta is facilitating ten million business conversations per week through its GenAI tools. None of these moves are framed as competition with OpenAI or Anthropic. They are infrastructure plays that make frontier models useful only if they integrate with the platforms that control distribution, payment, and deployment. The companies winning the AI era may not be the ones building the best models. They are the ones making it impossible to use anyone else's.
Sloane Duvall