OpenAI and Broadcom unveiled Jalapeño on June 24, the two companies’ first jointly developed Intelligence Processor and OpenAI’s first custom silicon (OpenAI, TechCrunch). The chip is an ASIC designed exclusively for LLM inference rather than training, allowing optimization around the specific memory-bandwidth and arithmetic patterns of token generation at scale (CNBC). Its die measures approximately 840 mm², near the reticle limit of EUV lithography systems, and went from initial design to manufacturing tape-out in nine months (Tom’s Hardware). Engineering samples are running ML workloads including GPT-5.3-Codex-Spark at production target frequency and power; OpenAI states performance per watt is substantially better than current state-of-the-art, with a full technical report to follow in coming months (OpenAI). Broadcom handles silicon manufacturing and Celestica covers rack and system integration, with initial deployment in gigawatt-scale data centers targeted for late 2026 (CNBC).
Enterprise AI coding tool costs are forcing budget corrections at large deployments. Microsoft is canceling most internal Claude Code licenses in its Experiences and Devices division, the group responsible for Windows, Microsoft 365, Outlook, Teams, and Surface, directing engineers to GitHub Copilot CLI by June 30, roughly six months after the tool was introduced internally in December 2025 (Windows Central, The Next Web). Token-based billing drove per-engineer API costs to between $500 and $2,000 per month at large deployments (Windows Central). Uber reported a comparable overrun: after deploying Claude Code to roughly 5,000 engineers and watching usage climb from 32% of the engineering organization in February to 84% by March, the company burned through its entire 2026 AI budget within four months (Fortune).
On June 22, OpenAI expanded its Daybreak cybersecurity initiative with GPT-5.5-Cyber, a security-focused model that scored 85.6% on CyberGym, a benchmark measuring automated vulnerability reproduction, versus 81.8% for GPT-5.5 (OpenAI, Axios). The model can autonomously navigate large codebases, trace attack paths, validate exploitability, generate patches, and produce remediation evidence within a single workflow (Axios). Access is limited to vetted defenders through the Trusted Access for Cyber program, with Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler among the initial partners (OpenAI).