Blog

Google’s next frontier model slipped — and the market took back $200 billion. Model roadmaps are now market-moving events: for anyone heading into tech or investing, this is the clearest evidence yet of how much AI execution is priced into big-cap valuations. Bloomberg reported July 16 that Gemini 3.5 Pro — announced in May and expected in June — has slipped months behind schedule after its coding performance missed Google’s own internal targets, forcing the company to restart parts of training; Alphabet shares fell about 4% on the report, erasing roughly $200 billion in market value in a day.

Microsoft is coaching its sales force to sell against OpenAI and Anthropic. The defining partnership of the AI era keeps unraveling — a live case study in what happens when a platform decides its partners are now competitors, and a preview of the model bake-offs anyone selling or buying enterprise software will sit through this year. At a July 14 internal strategy meeting first reported by Bloomberg, executives told salespeople to pitch Microsoft’s in-house MAI models as more efficient and cost-effective than OpenAI’s, Google’s, and Anthropic’s (“Everyone else is selling parts — we’re selling the full end-to-end system,” EVP Jay Parikh reportedly said). It follows reporting earlier this month that Microsoft has been swapping OpenAI and Anthropic models out of flagship apps like Word and Excel in favor of its own.

The first major lawsuit over AI-assisted layoffs landed — 26 Meta workers say algorithms flagged them while on protected leave. If you’ll manage people in an AI-instrumented workplace, this is the legal frontier arriving: courts are about to test whether “the model picked the list” survives employment law. The suit, filed Monday, July 13 in federal court in Oakland, alleges Meta used internal AI systems, keystroke and activity-monitoring data, and AI token-usage dashboards to help select some of the 8,000 layoffs it began in May — and that those scores structurally penalized employees on medical, parental, or family leave. Meta denies the claims, saying workforce decisions “were and are made by people, not AI”; legal commentators describe it as the first suit of its kind against a major U.S. tech company, and the allegations are so far just that — allegations.

China’s Moonshot released the largest open-weight model ever — and it wins some coding benchmarks outright. Open-weight models this strong change the buy-side math: any company weighing frontier API contracts now has a credible near-free alternative, which pressures U.S. labs’ pricing even where the open model isn’t best overall. Kimi K3, a 2.8-trillion-parameter model from Beijing-based Moonshot AI, took first place in Arena’s blind Frontend Code evaluation ahead of Anthropic’s Claude Fable 5, though Moonshot itself concedes K3 still trails Fable 5 and GPT-5.6 Sol on overall performance; full weights are due for release by July 27.

TSMC posted another record quarter — the AI buildout keeps showing up in real earnings. Whatever you make of AI-bubble talk, the cash is real at the chokepoint of the supply chain, and TSMC’s results are the cleanest single indicator of whether AI capex is still accelerating. Second-quarter revenue came in at US$40.20 billion, up 33.7% year over year, with net profit surging 77% to a single-quarter record, gross margin at 67.7%, and the company’s first revenue from its newest 2nm process — all driven by AI and high-performance computing demand.

Beijing institutionalized its AI diplomacy: 29 countries joined a new China-headquartered AI governance body. AI governance is becoming a two-bloc world, and where developing markets land will shape which models — and whose standards — global businesses build on. On July 16, 29 countries including Pakistan, Russia, and Kazakhstan signed an agreement with China establishing the World Artificial Intelligence Cooperation Organization, headquartered in Shanghai; Xi Jinping keynoted the World AI Conference the next day, calling for AI development that is “not a solo performance by any single country” and pledging 5,000 AI training opportunities for developing countries — an explicit pitch to the Global South as U.S. export curbs bite.