Sunday, June 28 June 28, 2026
Google capped Meta's Gemini AI access after demand outpaced compute supply — the clearest sign yet that the $650 billion infrastructure buildout can't keep pace with appetite. Meanwhile, Gemini 3.5 Flash gains computer use, Microsoft ships MAI-Code-1-Flash to GitHub Copilot, and a cascade of new model releases signals the industry's release velocity is only accelerating.
Good morning, and welcome to the MorningAI Daily Briefing for Sunday, June 28th, 2026.
Let's start with the story everyone is talking about this morning. Google has placed hard limits on Meta's use of its Gemini AI models. According to reporting from the Financial Times and confirmed by Bloomberg, Meta had been requesting far more computing capacity from Google Cloud than Google could actually deliver. Around March of this year, Google quietly imposed caps on Meta's API access after Gemini requests roughly doubled in just five months. The result? Some of Meta's internal AI projects have slowed down. This is a significant moment. Meta — one of the richest companies on earth — hit a compute wall it could not buy its way through. It signals that even top-tier cloud deals are no longer guaranteed unlimited throughput when the whole industry is sprinting at the same time.
And speaking of that sprint — the $650 billion figure is not a typo. Amazon, Google, Meta, and Microsoft are collectively on track to spend $650 billion on AI infrastructure in 2026. Nvidia's Jensen Huang added another wrinkle this week: China revenue for Nvidia has dropped to effectively zero due to US chip export restrictions, but Huang says a new $20 billion CPU expansion plan is the comeback. The AI hardware race is moving fast in multiple directions at once.
On the model front, this past week was unusually packed. Google quietly shipped computer use capabilities directly into Gemini 3.5 Flash, meaning developers can now point the Gemini Flash model at a screen and have it click, type, and navigate autonomously. That puts Gemini squarely in competition with Anthropic's computer use feature. Meanwhile, Microsoft launched MAI-Code-1-Flash — its own in-house coding model — directly into GitHub Copilot for Business and Enterprise teams. OpenAI is also experimenting with gifting Codex credits to users, a potential growth strategy that hints at the broader commoditization of AI coding tools.
Also making waves in the model space: this past week also brought GPT-5.6 variants named Sol, Terra, and Luna from OpenAI, a new Claude Tag capability from Anthropic, Sakana's Fugu model, Mistral OCR version 4, Qwen's AgentWorld, and a preview of Seedance 2.5 for video generation. This release velocity is not slowing — it is accelerating. The question for developers is no longer which model is best. It is which model is right for each task.
On the regulatory and geopolitical front, India signed on to the US-led "Pax Silica" AI pact, aimed at securing a shared semiconductor supply chain among allied nations. Separately, the Indian government is reportedly set to take a one to two percent equity stake in Sarvam AI, a Bengaluru-based AI startup — a rare move that signals sovereign interest in backing homegrown AI capacity. Asian AI startups are also capitalizing on US export restrictions by launching rival platforms to Anthropic and other American providers, marketing themselves as "sovereign AI" alternatives.
On the governance side, the Vatican held the first meeting of its new Interdicasterial Commission on Artificial Intelligence. Separately, TrustEvals and Accorian are warning enterprise finance teams about what they're calling "control drift" — the gradual degradation of AI compliance guardrails in live production systems, which they say could expose financial institutions to massive regulatory fines.
And one stat worth filing away: more than half of Georgia's teachers are now using AI to prepare for class. Ninety-five percent of those users said they use it for instructional planning at least a few times a month. Education adoption is happening faster than most policy frameworks can track.
Here is the business idea I'll leave you with today. The Google-Meta compute cap story reveals a real market gap: enterprises using cloud AI at scale have no early warning system when their compute allocation is throttled or at risk. A lightweight monitoring layer — think of it as a "compute credit health dashboard" for AI API consumers — could alert engineering teams before projects grind to a halt. It is a B2B SaaS wedge into every company that relies on third-party AI compute, and this week's news just made the pitch write itself.
That's your briefing for June 28th. Stay curious, stay building — and we'll see you tomorrow.