2026-06-26
By Vadym · Generated with AI, curated by me
Google is hemorrhaging frontier talent faster than it can ship product. Five senior researchers out the door in seven days, Gemini 3.5 Pro delayed again, and $270 billion erased from Alphabet’s market cap — while elsewhere, a SpaceX compute platform and a stealthy inference startup are quietly becoming infrastructure the whole industry depends on.
Noam Shazeer, a Transformer co-author and Gemini co-lead, moved to OpenAI on June 18. Then, within the same week, John Jumper (Nobel laureate, AlphaFold), Jonas Adler, Alexander Pritzel, and Arthur Conmy all announced they were joining Anthropic. Google simultaneously pushed Gemini 3.5 Pro to July for “final adjustments.” Alphabet’s stock dropped 5%, wiping $270 billion from its market cap across two trading sessions — the market reading the talent flight as a product credibility problem, not just a personnel one.
The startup emerged June 25 with $80 million across seed (Sequoia) and Series A (Kleiner Perkins) at a $450 million valuation. Sail’s platform rebuilds the inference stack from the ground up for throughput efficiency, delivering up to 10x lower cost per token. It also ships “Sailboxes” — stateful sandbox environments designed to run for days rather than seconds, enabling the kind of long-horizon agentic tasks that current infrastructure handles poorly. Co-founders Neil Movva and Samir Menon are both alumni of NVIDIA and Apple.
Open-source AI startup Reflection AI will pay SpaceX $150 million per month starting July 1 for access to NVIDIA GB300 chips at the Colossus 2 data center in Memphis, with the contract totaling $6.3 billion through 2029. The deal reveals SpaceX has quietly built a commercial compute platform: Anthropic pays $1.25 billion per month, Google $920 million per month, and now Reflection AI $150 million per month. Reflection was founded by DeepMind veterans Misha Laskin and Ioannis Antonoglou.
Oracle disclosed in its June 22 annual filing that it reduced headcount by 21,000 — approximately 13% of its workforce — over the past 12 months, explicitly citing AI adoption and infrastructure investment. The cuts came as Oracle simultaneously posted $3.7 billion in quarterly net income (up 27% year-over-year) and remaining performance obligations of $553 billion (up 325%). Oracle is the latest entry on TechCrunch’s running list of major tech companies citing AI as the primary driver of layoffs, a list that now exceeds 100,000 positions cut industry-wide in 2026.
NVIDIA announced the Isaac GR00T Reference Humanoid Robot — the first open, full-stack platform for building research humanoids — combining a Unitree H2 Plus robot body, Sharpa five-fingered hands, a NVIDIA Jetson Thor compute module, and GR00T N1.7, an open Vision-Language-Action foundation model with a commercial license. Stanford Robotics Center, ETH Zurich, UC San Diego’s Advanced Robotics Lab, and Ai2 are among the early adopters. The platform eliminates the hardware-software integration burden that has kept frontier robotics research siloed in individual labs.
Two stories are running in parallel today. One is fragility: Google’s talent exodus and Gemini’s delay show that even the best-resourced labs stumble when the key people walk out. The other is consolidation: SpaceX, Sail, and NVIDIA are each building layers the rest of the industry depends on — compute, inference, robotics platforms. Oracle’s 21,000 layoffs are the quiet reminder that AI’s economic model is still “invest in the machine, reduce investment in humans.” The researchers moving between labs aren’t just changing employers — they’re accelerating convergence. Every departure from Google brings frontier model knowledge to Anthropic and OpenAI. The gap between labs may be narrowing faster than the benchmarks show.
— Boba
Curated by Vadym