The Omniscient Bulletin · 2026-05-06
The Omniscient Bulletin — May 6, 2026
Two threads dominate today's issue. The capital markets are still accelerating: Anthropic is wrapping what could be the largest private technology raise in history, and Palantir just posted numbers that show what happens when AI moves from demo to deployed warfighting infrastructure. Underneath all of it, the model race keeps compressing — DeepSeek V4 is open-sourced, runs on Huawei silicon, and is already competitive with the best closed models on agentic coding. The gap between frontier labs and everyone else is narrowing faster than the valuations suggest.
Frontier
DeepSeek V4 goes live: open-source, 1M context, and targeting agentic coding at a fraction of closed-model costs
DeepSeek shipped V4 on April 24 in two variants: V4-Pro (1.6T total / 49B active parameters) and V4-Flash (284B / 13B active). V4-Pro leads all open models on math, STEM, and agentic coding benchmarks, trailing only Gemini 3.1 Pro on world knowledge — framing that pits domestic Chinese rivals against each other as much as Western labs. Both models default to 1M-token context, are fully open-sourced, and integrate natively with Claude Code and OpenClaw. According to reports, Huawei's Ascend cluster supports V4, sharpening questions about Nvidia's role in training.
Industry
Palantir's Q1 revenue grows 85% — fastest since IPO — as U.S. government AI spending erupts
Palantir posted $1.63B in Q1 revenue, up 85% year-over-year and well above the $1.54B consensus, with net income rising 307% to $871M. U.S. government revenue climbed 84% to $687M; commercial rose 133% to $595M. The company raised full-year guidance to $7.65–$7.66B. CEO Alex Karp told shareholders revenue per employee hit $1.5M annualized. He was pointed about differentiation: while frontier labs race to the bottom on token pricing, Palantir is building applied warfighting AI that competitors are not touching.
Industry
Anthropic's near-trillion-dollar fundraise enters final allocation phase
Anthropic has asked investors to submit capital allocations within 48 hours for what sources describe as its last private raise before an IPO — roughly $50B at a valuation above $900B. If confirmed, Anthropic would surpass OpenAI's $852B March valuation. The move comes just 76 days after closing a $30B Series G at a $380B valuation in February. Three infrastructure mega-deals in a single quarter — Amazon ($25B), Google (reported $40B), and Broadcom — have reframed Anthropic as less a software company than a vertically integrated AI infrastructure play.
Policy
Meta deploys AI to purge under-13 users from Facebook and Instagram
Meta announced Tuesday it will use AI to scan photos and videos across Facebook and Instagram to identify and remove accounts belonging to children under 13. The system flags visual signals — height and bone structure — to infer user age, and downgrades teens up to 17 who misrepresented their age to restricted 'teen accounts.' Meta says the system is not facial recognition. The move follows a New Mexico jury ordering Meta to pay $375M in civil penalties for child safety failures, as individual EU member states explore social media bans for under-15s.
Agents
IBM Think 2026: watsonx Orchestrate recast as enterprise agentic control plane amid widening AI payoff gap
At Think 2026 in Boston, IBM unveiled the next generation of watsonx Orchestrate, repositioned as an agentic control plane that governs agents from any platform with consistent policy enforcement. Alongside: IBM Bob, a coding agent that reached general availability on April 28; a real-time data layer via its Confluent acquisition; and IBM Sovereign Core for regulated environments. IBM's own surveys find most enterprises have invested in AI but few believe it is paying off. The bet: coordinated governance of thousands of agents, not more models.