The AI market is consolidating around two incompatible conclusions: that models will become commodities and that dominant players are worth trillions. This tension is playing out in real time across pricing, acquisition, and enterprise strategy.
DeepSeek's price increases of more than 1,100% on some V4 API tiers demolish the narrative that cost leadership is sustainable. The company that built its entire positioning on ultra-low pricing is now rationing capacity through peak and off-peak rates, a classic move when demand exceeds supply. Simultaneously, Anthropic commands a $2 trillion IPO valuation and OpenAI faces a price war with Chinese rivals, yet both are releasing cheaper models to defend market share. Writer built its new system as a post-training variation on an open source model to reduce token costs. This is not competition driving prices down. This is competition driving margins down while founders and early investors extract value through exit events. Databricks wanted to raise $1 billion but settled for $5 billion at a $190 billion valuation when investors fought to get in, not because the company proved unit economics work at scale, but because AI infrastructure is perceived as a necessary tax on every deployment. Google released Gemini 3.7 Flash just three weeks after 3.6 Flash, each claiming substantial improvements, a release cadence that suggests feature velocity matters more to market positioning than genuine differentiation. OpenAI's new "Ultrafast" mode for GPT-5.6 Sol runs at 14x speed to court enterprise users, another signal that speed and availability are becoming the real product when models converge on capability.
The enterprise sales machine is accelerating even as product maturity stalls. IBM is training tens of thousands of consultants on OpenAI's stack. OpenAI replaced its chief revenue officer after nine months and hired Dali Rajic from Wiz, a company that mastered the enterprise sales playbook. Microsoft is consolidating Copilot apps and killing failed features like AI-generated podcasts and Deep Research, a retreat dressed as simplification. Adobe launched AI Collaborators in Workfront to embed agents directly into workflow management. Databricks acquired Electric to run Postgres databases locally within agentic applications, solving a real infrastructure problem but also deepening lock-in. Nvidia is financing GPU purchases through new financial instruments to ensure older chips retain value as new models emerge. These are not product innovations. These are distribution and lock-in strategies layered onto models that are becoming faster, cheaper, and harder to differentiate. The trillion-dollar valuations and the price wars are not contradictions. They are different ways to extract value from the same bottleneck: the transition from experimentation to production, where switching costs, integration depth, and installed bases matter far more than raw model performance.
Sloane Duvall