Enterprise AI economics are cracking: Claude's new pricing model is pushing CIOs toward offshore alternatives, OpenAI's Codex gains Windows computer use with unresolved security questions, and Microsoft's expanded ISO 42001 audit reveals how fast governance is being stress-tested. Today's briefing maps where the $700B infrastructure bet is already showing fractures.
Audio is available on Spreaker — see link below.
Hyperscalers are on track to spend somewhere between six hundred fifty billion and seven hundred fifty billion dollars on AI infrastructure in twenty twenty-six alone. That number isn't a forecast.
Anthropic's shift to usage-based tokenizer pricing for Claude is already triggering cost alarms inside major companies. CIOs aren't waiting to see how this plays out.
OpenAI moved quickly on a different front. Codex now supports Computer Use on Windows as of May twenty-ninth, extending autonomous desktop control beyond Mac for the first time.
OpenAI also announced Rosalind Biodefense on May twenty-ninth. It's a restricted-access program giving select U.S. government and allied biodefense partners access to GPT-Rosalind for high-stakes life sciences applications.
Microsoft cleared its ISO forty-two thousand and one recertification for Copilot in March twenty twenty-six with a meaningfully broader scope than the previous audit. The twenty twenty-five baseline covered single-model Copilot.
GPT-five point five Instant also received a quality update on May twenty-eighth. Cleaner, more concise responses, and canvas replaced with in-chat writing and coding blocks.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.