Musk, Zuckerberg, Sacks kill AI safety order; Anthropic hits $30B, NSA gets $9B chips

The same forces that killed an AI safety executive order are simultaneously pouring billions into the technology they claim needs no restrictions. A late-night phone call between three of tech's most powerful figures—Elon Musk, Mark Zuckerberg, and David Sacks—reportedly branded the draft order as 'doomer regulation' and derailed its progress. Meanwhile, Anthropic claims a $30+ billion valuation with $10.9 billion in annualized revenue and $1 billion in ARR for its Claude Code agent. The NSA will receive $9 billion in Blackwell chips from NVIDIA. These events are not contradictory; they reveal a strategic schizophrenia where safety is sidelined in favor of capability acceleration and national security dominance.

Anthropic's financial claims are seismic if true. OpenAI, its primary competitor, reported $3.4 billion in revenue for 2024. Google DeepMind and Mistral AI operate at far smaller scales. The $10.9 billion figure would imply Anthropic has captured a massive share of enterprise coding agents, surpassing GitHub Copilot's estimated $2 billion ARR. But industry analysts urge skepticism: the number likely reflects a multi-year deal annualized rather than sustainable recurring revenue. Microsoft's reported cancellation of an internal Claude Code pilot after token billing consumed its entire annual AI budget underscores the cost barriers that even deep-pocketed enterprises face. The $1 billion ARR for Claude Code may include interdepartmental accounting or a single government contract.

The government's actions crystallize the tension. The kill of the safety order was orchestrated by industry insiders who argued regulation would stifle innovation. Yet the Pentagon's objections to keeping Claude inside the NSA were overruled by the White House, and $9 billion was allocated for Blackwell chips—a direct investment in the same class of models the order sought to constrain. This is not hypocrisy; it is a realpolitik recognition that AI capability is now a national security asset. The CISA report of 15,000 exploit attempts against a Drupal SQL vulnerability within 48 hours of patch further illustrates the cybersecurity imperative driving this investment. The hidden risk is that by prioritizing deployment over safeguards, we are building an infrastructure that is both powerful and brittle.

The bottom line is that the AI industry has entered a 'prove it' phase. The kill of the safety order signals that self-regulation is the preferred model. But the $9 billion NSA deal and Anthropic's inflated revenue claims suggest that the real competition is not between companies but between the need for control and the fear of falling behind. Expect more scrutiny on revenue disclosures and a continued arms race in compute, regardless of safety concerns.