Google DeepMind suffered an unprecedented talent haemorrhage during the week of 24 June, losing four senior researchers in six days to direct competitors. Noam Shazeer, co-author of the foundational Transformer paper, joined OpenAI as Lead for Architecture Research. John Jumper, the Nobel Prize-winning creator of AlphaFold, departed for Anthropic. Two additional Gemini researchers — Jonas Adler and Alexander Pritzel — also announced moves to Anthropic, collectively representing the AlphaFold team's core leadership.
Alphabet stock fell 7.2 per cent intraday as investors questioned whether Google can maintain its AI competitiveness. The concern is compounded by Gemini 3.5 Pro's failure to ship within June despite CEO Sundar Pichai's explicit May commitment to a June launch. With six days remaining in the month, leaked benchmarks suggest the model struggles in advanced reasoning compared to Claude Fable 5 and GPT-5.5, raising questions about whether the delay reflects quality concerns rather than routine scheduling.
Simultaneously, SpaceX signed a compute lease with Reflection AI for $150 million per month — approximately $6.3 billion over the full contract term through 2029 — providing access to NVIDIA GB300 chips at the Colossus 2 Memphis facility. This is SpaceX's fourth major external compute lease, following agreements with Anthropic ($45 billion), Google ($30 billion), and Cursor.
OpenAI's GPT-5.5-Cyber, launched on 22 June, achieved 85.6 per cent on CyberGym — the highest single-model cybersecurity score ever recorded and higher than Anthropic's Mythos 5. The model is restricted to vetted organisations through OpenAI's Trusted Access for Cyber programme, directly competing with Anthropic's Project Glasswing.
For context engineers, Google's talent exodus raises a critical question: can a company maintain frontier AI capabilities when its most foundational researchers are leaving for competitors? The answer will shape the competitive landscape for years to come.
