Gemini 3.5 Pro Goes GA as ChatGPT Loses Its Majority, China's $295B Infrastructure Bet, the $700B Capex Sprint: Wednesday Briefing, June 24, 2026

Google’s Gemini 3.5 Pro is entering its general availability window with a 2-million-token context length — a technical capability that no competitor currently matches at the same tier. The model’s release is well-timed: for the first time, ChatGPT has slipped below 50 percent market share in the conversational AI space. Gemini now holds 27.7 percent, Claude accounts for 10.3 percent, and the long tail of other models makes up the rest.

The market-share shift is significant not because any single percentage point matters, but because it marks the end of the period in which ChatGPT was the default. A three-player market with a declining leader behaves very differently from a monopoly — pricing pressure increases, switching costs matter more, and distribution advantages (Google’s integration with Search, Workspace, and Android; Anthropic’s enterprise partnerships) become decisive.

China announced a $295 billion national AI infrastructure plan, the largest single-country AI investment commitment to date. The plan covers compute infrastructure, talent development, and industrial AI deployment, and it represents Beijing’s clearest statement yet that AI is a strategic priority on par with semiconductor self-sufficiency.

The scale of the investment changes the global calculus. Even accounting for the gap between announced plans and actual disbursements, a commitment of this magnitude will reshape chip demand, talent markets, and the competitive dynamics between US and Chinese AI ecosystems for years to come.

Hyperscaler capital expenditure for 2026 is approaching $700 billion collectively, a figure that would have been dismissed as fantasy even two years ago. The spending is going primarily into data centers, custom chips, networking infrastructure, and the energy systems required to power it all.

Two downstream stories illustrate the physical constraints this spending is hitting. FERC— the Federal Energy Regulatory Commission — is pushing to accelerate grid hookups for data centers, acknowledging that the speed of AI infrastructure buildout is outpacing the electrical grid’s ability to connect new facilities. And a reported $40 billion acquisition of Aligned Data Centers underscores the premium that purpose-built AI compute facilities now command.

Autonomous agents are moving from enterprise pilot programs into mainstream deployment across both enterprise workflows and consumer search experiences. Microsoft’s Agent Framework, unveiled at Build, is gaining momentum as developers adopt its patterns for building production agent systems. The framework’s integration with Azure and the broader Microsoft ecosystem gives it a distribution advantage that pure-play agent startups will struggle to match.

Apple’s revamped Siri, while not positioned as an “agent” in the technical sense, represents the consumer-facing edge of the same trend: AI systems that take actions on behalf of users rather than simply answering questions. The gap between Apple’s approach and the more ambitious enterprise agent frameworks highlights how differently the consumer and enterprise markets are approaching autonomous AI.