The real story today isn't what companies are building. It's who gets to extract value from what already exists, and what happens when that power shifts.
Anthropic faces a Sony lawsuit over staff chat messages praising piracy, OpenAI is cutting off Cursor's access to its models following SpaceX's acquisition of Cursor's parent company, and Apple is presenting evidence of alleged data theft by a former employee now at OpenAI. These aren't separate incidents. They're markers of a consolidation phase where incumbents are using legal and contractual leverage to constrain competitors and control access to training data, models, and talent. The pattern is clear: build moats, not just products. Meanwhile, the EU's Digital Services Act is now applying to ChatGPT and Reddit, imposing compliance costs that scale with user growth. The regulatory burden lands heaviest on companies that have already won distribution. Smaller players and open-source alternatives face friction, but the rules hit fastest where adoption is highest.
Nvidia's $3.5 billion investment in MediaTek reveals the real anxiety underneath all this. Big Tech is building its own AI chips because relying on Nvidia creates a chokepoint they can't tolerate. Neoclouds are capturing $25 billion in revenue and Gartner expects them to claim 20 percent of the $267 billion AI cloud market. The hyperscalers are losing pricing power. Cursor loses OpenAI models. SpaceX gets a coding platform. OpenAI gets contractual control back. The Pentagon now runs ChatGPT, Grok, and Gemini on a central portal. These are all moves to secure supply chains and lock in dependencies before the market fragments further.
The pressure cascades down. Wall Street banks are telling law firms to cut fees because AI made routine work faster. Insurance claims adjusters report 98 percent negative Glassdoor reviews mentioning AI. Blue Voice raises $6 million to build domain-specific models for police departments because general-purpose models can't compete on local knowledge. Clipto hit profitability at $15 million ARR by solving a specific problem at scale. The market is sorting into layers: incumbents defending territory through legal and contractual control, specialists building defensible positions in narrow verticals, and open-source alternatives serving cost-conscious users willing to accept less polish. None of these trends resolve the underlying tension. Regulatory compliance keeps rising. Chip costs keep falling. Data access keeps tightening. The winner isn't whoever builds the best model anymore. It's whoever controls the input, the output, or the platform the output runs on.
Sloane Duvall