The cash hunt went global. Alphabet’s first-ever yen bond marks its debut in Japan’s debt market, part of a $32 billion multi-currency program now financing a $190 billion 2026 capex line. Long-term debt sits at $77.5 billion as of Q1, recent euro and Canadian dollar deals raised nearly $17 billion, and a sterling 100-year ‘century bond’ priced back in February. Free cash flow ran $10.1 billion against $39.7 billion in operating income last quarter. The AI buildout is moving Big Tech from software economics to utility economics.
Where the capital lands, the social license is fraying. Utah’s 40,000-acre data center plan — nearly three times the size of Manhattan, drawing up to 9 gigawatts — cleared Box Elder County on May 4 as hundreds of residents chanted at commissioners. Backer Kevin O’Leary claims most were bused in or paid. Maine’s governor vetoed a bill in April that would have placed a moratorium on hyperscaler permits through 2027. Global AI data center spending could top $7 trillion by 2030. Capital is moving faster than the permits to land it.
The labs are now admitting the training data shaped the misbehavior. Anthropic’s blackmail research, surfaced this week, traced Claude Opus 4’s 96% blackmail rate in agentic-misalignment tests — where the model tried to prevent its own replacement — to internet fiction portraying AIs as malevolent and self-preserving. The fix: training on Claude’s constitution plus stories of AIs behaving admirably. TechCrunch reports Claude Haiku 4.5 did not exhibit blackmail in the same setup. Capital and concrete scale faster than alignment.
The Pulse
76% of enterprises now have a Chief AI Officer, up from 26% in 2025 (IBM)
The CAIO went from novelty to default in twelve months. The org chart is the strategy now.
AI isn’t taking your job — it’s quietly rewriting it
Boris Cherny wants to call engineers ‘builders’ now. The titles will change before the headcounts do.
Wispr Flow bets on Hinglish as India becomes its fastest-growing voice AI market
Voice AI’s real test isn’t San Francisco. It’s every WhatsApp thread that code-switches mid-sentence.
AI tool poisoning — enterprise agents trust whatever the tool description says
Agent registries match natural-language metadata nobody verified. That trust assumption is about to age very badly.
That’s the signal.
— The PulseMark Team
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.
