AI Newsletter

OpenAI and Broadcom have jointly unveiled their first proprietary AI chip called 'Jalapeño', which was developed in just nine months. The chip is designed to run ChatGPT, Codex, and future agents more efficiently and is set to be deployed by the end of 2026. This move reduces OpenAI's dependency on NVIDIA and fundamentally changes the power structure in the AI hardware market.

According to current reports, following John Jumper and Noam Shazeer, another wave of leading AI researchers is leaving Google DeepMind for Anthropic – a sustained brain drain that structurally threatens DeepMind's competitiveness. Prediction markets see Anthropic with 84% as the likely model leader by the end of July 2026. The strategic concentration of top talent at Anthropic suggests an imminent model breakthrough.

Polymarket shows 75% probability that Anthropic goes public before OpenAI – a clear shift from earlier expectations. Meanwhile, OpenAI's IPO market capitalization is being traded between $1.25 and $1.5 trillion (46% probability). The simultaneous preparation of both AI heavyweights for IPOs could make 2026 the most significant IPO year in technology history.

Companies worldwide are finding that uncontrolled AI spending is escalating, and are shifting from broad experimentation phase to targeted, cost-efficient deployments. The trend is moving toward smaller, focused agent solutions with clear SLAs and governance structures instead of company-wide rollouts. IBM CEO Arvind Krishna publicly warned that companies must now decide which AI tools are actually worth their price.

Micron Technology has concluded 16 Strategic Customer Agreements, of which 14 alone secure the lion's share of future HBM capacity. The CFO expects the overall HBM market to exceed the $100 billion threshold in 2027. This underscores that the AI boom is increasingly no longer just a GPU competition, but a memory infrastructure race.