The Signal — April 9, 2026
Meta shipped its first model from the Superintelligence Lab, and it's not open-source. OpenAI says enterprise is now 40% of revenue with consumer parity in sight by year-end. Two federal courts issued opposite rulings on Anthropic's Pentagon blacklisting, and the same Pentagon is pushing AI deployment with explicit instructions to deprioritize safeguards.
Meta's Superintelligence Lab Ships Its First Model. It's Closed-Source.
Meta released Muse Spark, the first AI model from its Superintelligence Lab formed in mid-2025. The model is proprietary, breaking from the open-weight Llama family that defined Meta's AI identity for the last three years.
Muse Spark includes a "Contemplating" mode that runs up to 16 parallel reasoning agents, scoring 58.4 on Humanity's Last Exam with tools. Rather than extending a single chain of thought, it splits problems across independent agents working simultaneously. Zuckerberg says future Muse models will include open-source releases, but Spark stays closed.
The signal here is strategic. Meta spent years positioning Llama as the open alternative to GPT and Claude. Shipping a proprietary frontier model says they think open weights can't win the race they're now running. The model also integrates Instagram, Facebook, and Threads content directly into responses, turning Meta's social graph into a data moat that no open-source project can replicate.
Sources: Ars Technica, Wired, The Decoder
OpenAI Says Enterprise Is 40% of Revenue
OpenAI's enterprise chief shared Q1 numbers: enterprise now makes up over 40% of revenue, on track to match consumer by end of 2026. Codex hit 3 million weekly active users. APIs process 15 billion tokens per minute. New enterprise customers include Goldman Sachs, Phillips, and State Farm.
Hard revenue breakdowns from OpenAI are rare. Two years ago enterprise was a footnote. Now it's approaching half the business, and the Frontier platform pitch positions OpenAI as company-wide agent infrastructure, not just chat or copilots. Fifteen billion tokens per minute is a number worth sitting with. That's the throughput of every enterprise customer's AI workload combined, running constantly.
Sources: OpenAI
Two Courts, Opposite Rulings on Anthropic's Pentagon Blacklisting
A DC appeals court refused to pause Anthropic's supply-chain-risk designation, saying it won't "lightly override" military judgments on national security. This directly contradicts a San Francisco judge who ruled last month that the Pentagon acted in bad faith. The government filed the designation under two different supply-chain laws, and each court is handling only one, producing two active preliminary rulings that say opposite things.
Acting AG Todd Blanche called the DC ruling "a resounding victory for military readiness." For the AI industry beyond Anthropic, the precedent being set is blunt: the executive branch can functionally blacklist an AI company for refusing military contracts. Whether that power holds up will depend on which ruling survives appeal.
Pentagon Goes "AI First" Under Hegseth
The Nation published a deep investigation into how Secretary of War Pete Hegseth's January AI acceleration memo is being put into practice. Seven "Pace-Setting Projects" are pushing AI into battlefield decision support, intelligence-to-action pipelines, and administrative operations on timelines of months, not years. Salesforce landed a $5.6 billion, 10-year AI contract.
The memo explicitly frames delays, risk aversion, and procedural safeguards as liabilities. Congressional oversight gets similar treatment. Read alongside the Anthropic court story, the picture sharpens: the same government rushing to deploy AI everywhere is also punishing the one major AI company that pushed back on military use. The $5.6 billion Salesforce deal and open-ended contracts with venture-backed startups show the spending is real and accelerating.
Sources: The Nation
