1. AlphaFold's John Jumper Leaves Google DeepMind for Anthropic
Nobel laureate John Jumper, who led the AlphaFold 2 protein-structure breakthrough, departed Google DeepMind after nine years on June 20 to join Anthropic, where he is expected to drive the company's "AI for science" push. The exit landed days after Transformer co-author Noam Shazeer also left for OpenAI, and Alphabet stock fell roughly 7.2% as investors weighed the back-to-back losses of marquee research talent.
2. Noam Shazeer Joins OpenAI to Lead Architecture Research
Noam Shazeer — co-author of the 2017 "Attention Is All You Need" paper that introduced the Transformer — is leaving Google DeepMind for OpenAI, where he will serve as Lead for Architecture Research, owning the core neural-network structures behind OpenAI's models. The move continues an intense 2026 talent war among the frontier labs.
3. OpenAI Ships GPT-5.5-Cyber, a Gated Cybersecurity Model
On June 22 OpenAI released GPT-5.5-Cyber, scoring 85.6% on CyberGym (vs. 81.8% for standard GPT-5.5), 39.5% on ExploitGym, and 69.8% on SEC-bench Pro — described as the highest CyberGym result ever recorded for a single model. Access is restricted to vetted organizations via a "Trusted Access for Cyber" program, and OpenAI paired it with a "Patch the Planet" effort using Trail of Bits and HackerOne to put human review ahead of any vulnerability disclosure.
4. SpaceX Signs $6.3B Compute Lease With Reflection AI
SpaceX agreed to a $6.3 billion compute lease giving Reflection AI access to Nvidia GB300 chips at its Colossus 2 facility in Memphis through 2029 — the fourth major external client for a site now claiming over $80 billion in committed external revenue. Reflection separately raised at a $25 billion valuation, pitching itself as an American open-weight frontier alternative to closed US labs and Chinese models.
5. Zhipu AI Releases GLM-5.2 With a 1M-Token Context Under MIT License
Chinese lab Zhipu AI released GLM-5.2 in mid-June, featuring a 1-million-token context window and a permissive MIT license that lets developers use it commercially with few restrictions. It is one of several aggressive open-weight launches from Chinese labs this month, alongside Moonshot's Kimi K2.7 Code and MiniMax's M2.5.
6. China Unveils a $295B Five-Year AI Infrastructure Plan
Beijing announced a roughly $295 billion, five-year state AI infrastructure program — about $59 billion of government-directed AI spend per year — covering compute, data centers and model development. The plan underscores how national-scale capital, not just private labs, is now shaping the AI build-out.
7. US FERC Orders Grid Operators to Speed AI Data-Center Hookups
On June 18 the Federal Energy Regulatory Commission issued "show cause" orders to all six US regional grid operators, requiring each to defend its existing interconnection framework or propose reforms that let large-load customers — specifically AI data centers — connect to the power grid faster. The move signals that electricity, increasingly, is the binding constraint on AI growth.
8. ChatGPT Falls Below 50% Market Share for the First Time
ChatGPT's share of the consumer AI assistant market slipped to 46.4% by late May 2026 — its first time under 50% — as Google's Gemini climbed to 27.7% and Anthropic's Claude reached 10.3%. Meanwhile Morgan Stanley projects global AI-linked debt will nearly double to $570 billion in 2026 as hyperscalers and frontier labs finance massive infrastructure.
// KEY TAKEAWAYS
Two themes dominate this week: a frenzied talent war — with Jumper heading to Anthropic and Shazeer to OpenAI in the same stretch — and a hard pivot toward infrastructure and energy as the real bottlenecks, from SpaceX's multibillion-dollar compute leases to FERC pressing the grid and China committing $295B of state money. On the model front, the open-weight wave keeps accelerating (GLM-5.2, Kimi, MiniMax) while OpenAI pushes into gated, safety-screened specialist models. And the competitive map is shifting: ChatGPT slipping below 50% share shows the assistant race is genuinely multi-horse now.