AI Insider #45 -- The Courtroom, the Credits, and the Cold War

The biggest AI trial in history opened today, GitHub just killed flat-rate AI pricing, and China torpedoed a $2B acquisition. This week is a stress test for everyone building on AI infrastructure.

This Week’s Signal

Musk v. Altman goes to trial. Opening arguments began today in the case that could reshape OpenAI’s corporate structure. Musk’s legal team alleges OpenAI “stole a charity” when it pivoted to a capped-profit model after the October 2022 Microsoft deal. The jury will weigh whether OpenAI abandoned its founding mission. Meanwhile, a Wall Street Journal report reveals OpenAI missed internal revenue and user targets, falling short of its goal of one billion weekly ChatGPT users by end of last year. The company’s CFO has reportedly expressed concern about IPO timing and datacenter spending. Outcome here could set precedent for every AI non-profit-to-profit conversion that follows.

GitHub moves Copilot to usage-based billing. Starting June 1, GitHub will replace its flat monthly “premium requests” model with an AI Credits system tied to actual token consumption. The company says it can “no longer absorb the escalating inference cost” from heavy users. Under the new model, subscribers get credits matching their subscription cost, with overages billed at listed API rates per model. This is the clearest signal yet that the era of subsidized AI tooling is ending.

China blocks Meta’s $2B Manus acquisition. Beijing ordered Meta to unwind its deal for Manus, the Chinese-founded AI agent startup, after a months-long regulatory probe. The veto lands as US-China AI tensions escalate – Washington accused Beijing of “industrial-scale” AI theft just last week. For Meta, it is a setback in Zuckerberg’s push to dominate the AI agent layer. For the broader market, it confirms that cross-border AI M&A now carries sovereign risk.

David Silver raises $1.1B for Ineffable Intelligence. The former DeepMind researcher behind AlphaGo launched his lab just months ago. It is now valued at $5.1 billion, focused on AI systems that learn without human data. The bet: reinforcement learning and self-play can outperform the data-hungry LLM paradigm. If Silver is right, the entire training data economy shifts.

Google invests up to $40B in Anthropic. Following Amazon’s recent investment, Google has committed up to $40 billion in cash and compute to Anthropic. The hyperscaler arms race for frontier model access is now a three-way war between Google, Amazon, and Microsoft – each backing different horses while hedging with their own models.

Operator Playbook

1. Audit your AI tooling costs now, not in June.

GitHub’s move to usage-based billing is not an isolated event. Anthropic already tested removing Claude Code from its Pro plan due to “untenable demand.” Every AI tool provider is doing the same math: heavy users cost more than their subscription covers. If your team relies on AI coding assistants, code completion, or agent workflows, map your actual token consumption today. Build internal dashboards. Set usage budgets per developer. The companies that understand their AI unit economics before the pricing shift will adapt; the rest will get a surprise bill.

Your move: Pull your team’s Copilot usage data this week. Calculate what June 1 pricing would cost at current consumption. Identify your heaviest users and evaluate whether their output justifies the spend.

2. Treat geopolitical risk as a supply chain variable.

The Manus block is not a one-off. Any AI company with Chinese founders, Chinese investors, or Chinese data dependencies now carries regulatory risk in both directions. The US is tightening export controls while China is blocking acquisitions. If you are evaluating AI vendors, add a “sovereignty risk” column to your comparison matrix. Where are the models trained? Where is the company incorporated? Who are the investors? These questions now affect uptime, not just compliance.

Your move: Review your top three AI vendor dependencies. For each, document incorporation jurisdiction, training data geography, and major investor nationality. Flag any that sit in a regulatory crossfire zone.

3. Watch the OpenAI trial for structural signals, not drama.

The Musk-Altman courtroom fight will generate months of headlines. Ignore the personality theater. What matters: if the jury finds OpenAI deviated from its charitable mission, it sets legal precedent for clawback risk on every non-profit-to-profit AI conversion. That affects Anthropic’s structure, any university spin-out, and the growing wave of “public benefit” AI companies. The ruling could also affect OpenAI’s IPO timeline and valuation, which in turn affects the API pricing and capacity that thousands of businesses depend on.

Your move: If your product depends on OpenAI APIs, scenario-plan for two outcomes: OpenAI IPOs on schedule with stable pricing, or OpenAI faces structural upheaval that delays capacity expansion. Have a fallback provider tested and ready.

Steal This

The Silver raise proves a pattern worth watching: the next wave of AI labs is not competing on scale – it is competing on learning efficiency. Systems that can train without massive human-curated datasets sidestep the licensing battles, copyright lawsuits, and data moats that define the current LLM landscape. If you are building on top of AI, keep an eye on self-play and synthetic data approaches. The cost structure of the models underneath your product could change dramatically within 18 months.

The Bottom Line

The AI industry is entering its accountability phase. Flat-rate pricing is dying. Cross-border deals carry sovereign veto risk. The biggest company in the space is on trial for abandoning its mission. And a new $5B lab thinks it can make human training data obsolete. The operators who survive this transition are the ones who stopped assuming stability three months ago.

AI Insider is published by Digital Forge Studios Inc.