Talent keeps flowing to Anthropic while OpenAI and Google prune old models and chipmakers pour billions into inference and memory. US-China friction and school-level AI bans round out a day of consolidation and containment moves.
Top Stories
Anthropic Accuses Alibaba of ‘Illicitly’ Accessing AI Models — Anthropic says Alibaba used thousands of fraudulent accounts to reach its Claude models despite the company’s decision to block China access. The move exposes how hard it is to enforce geographic restrictions on frontier models. https://www.bloomberg.com/news/articles/2026-06-24/anthropic-accuses-alibaba-of-illicitly-accessing-its-ai-models
Google Poised to Lose Two More High-Profile AI Staffers to Anthropic — Two senior Google researchers are reportedly joining Anthropic, extending a pattern of departures from Alphabet. The poaching highlights Anthropic’s edge in attracting top talent. https://www.bloomberg.com/news/articles/2026-06-24/google-poised-to-lose-two-more-high-profile-ai-staffers-to-anthropic
Qualcomm to buy startup Modular for $4 billion in AI software push — Qualcomm is acquiring the AI software startup to strengthen its position beyond chips. The deal shows hardware players racing to own more of the stack. https://www.reuters.com/business/qualcomm-buy-ai-startup-modular-2026-06-24/
Micron reports 346% sales surge on AI memory chips — Micron’s revenue jumped sharply on high-bandwidth memory demand, helping ease a broader AI sector selloff. US manufacturing expansion funded by the CHIPS Act is accelerating. https://fortune.com/2026/06/26/micron-technology-346-percent-revenue-surge-ai-memory-chips-hbm-earnings-2026/
Groq Raises $650M After Nvidia’s $20B Deal — Groq closed fresh funding months after Nvidia licensed its LPU architecture for $20 billion. Inference specialists continue to draw serious capital. https://memeburn.com/groq-raises-650m-after-nvidias-20b-deal/
OpenAI retires GPT-5.2 models — GPT-5.2 Instant, Thinking, and Pro versions were removed from ChatGPT on June 12, with existing threads moved to GPT-5.5 equivalents. Providers are actively cleaning up older model lines. https://help.openai.com/en/articles/6825453-chatgpt-release-notes
Google deprecates Gemini image preview models — gemini-3.1-flash-image-preview and gemini-3-pro-image-preview will shut down June 25. Another round of model sunsetting from the major labs. https://ai.google.dev/gemini-api/docs/changelog
Trump says he no longer views Anthropic as national security threat — The former president told Axios he would not rule out emergency powers but currently sees no need. Policy signals around frontier labs remain fluid. https://www.reuters.com/technology/artificial-intelligence/
NVIDIA announces Vera Rubin platform — The new supercomputer platform targets scientific workloads and was unveiled at ISC High Performance 2026. https://nvidianews.nvidia.com/
Captain’s Take
Talent and capital are concentrating fast around a handful of names while the big labs keep retiring models they just shipped. That churn forces every builder to treat model choice as temporary infrastructure, not a stable foundation.
For anyone shipping real products the pattern is clear: fight for scarce researchers, watch your inference costs get arbitraged by new hardware plays, and assume any closed model can vanish with two weeks notice. The Alibaba clash and school bans both point to the same thing — containment is getting messier and more political at every layer.
Next watch the post-deprecation migration pain and whether the next wave of funding goes to wrappers or actual new silicon. We keep betting on local control because the centralized story keeps proving brittle.