Anthropic Buys Stainless, Hires Karpathy, Lands in CIA Briefings

Three Anthropic stories share the top of today's batch. The rest of the day is X chasing creator ad dollars and Nvidia walking into an earnings test with the AI-spend question hanging.

1. Anthropic acquires Stainless for $300M+

Anthropic acquired Stainless, the SDK generation platform that builds the official client libraries for OpenAI, Google, and Cloudflare, for more than $300 million. The deal is roughly double Stainless's December 2024 Series A valuation. The tooling that turns competitor APIs into developer-usable code is now owned by a direct competitor; OpenAI and Google have to rebuild that pipeline themselves.

My take: This is infrastructure capture, not a talent acquisition. The interesting question is not whether OpenAI and Google can rebuild SDK generation in-house (they can), it is how much developer mindshare leaks during the rebuild. Every week a competitor's SDK lags, Anthropic's first-party Claude SDK is the smoother one to integrate. Worth watching: how Stainless's open-source posture changes (if at all) under the new owner.

2. CIA flags Mythos as a "reflection point"

Dan Richard, Associate Deputy Director of the CIA's Digital Innovation Directorate, told reporters that Anthropic's Mythos model brings federal agencies handling sensitive information to a "reflection point" given its hacking capabilities. Security researchers split between excitement about defensive use cases and concern about lowering the bar for attackers.

My take: This is the first time I have seen a sitting US intel official publicly label a specific commercial AI capability as a regulatory inflection point. The framing matters more than the model itself. Once "reflection point" enters the vocabulary, the next round of administrative actions has a ready-made hook.

3. X launches Creator Connect, an AI brand-matching ad product

X (now controlled by Musk's SpaceX) rolled out Creator Connect, an AI-driven product that pairs brands with creators on the platform. The pitch is automated matching that bypasses the influencer-agency layer currently sitting between advertisers and creators.

My take: Platforms have been trying to commodify the creator-to-brand match for a decade. The AI angle is interesting because it threatens the influencer agencies that extract most of the margin today. Whether X has the trust or the first-party data to actually do this matching well is the open question. The platform with the messiest brand-safety reputation is a strange place to launch an automated brand-matching product.

4. Karpathy joins Anthropic's pre-training team

Andrej Karpathy, OpenAI co-founder and former Tesla AI lead, started this week at Anthropic on the pre-training team under Nick Joseph. An Anthropic spokesperson said Karpathy will lead a new team focused on using Claude to accelerate pre-training research.

My take: "Use the model to train the next model" is the bet I think is most underrated right now. The shape of the next frontier model is increasingly determined by how good its predecessor is at helping you train it, not by raw compute. Putting Karpathy on this signals Anthropic thinks the talent moat is at the meta-research layer, not in application-level fine-tuning.

5. Nvidia heads into earnings with the AI-spend question hanging

Nvidia is expected to post 79% revenue growth on Wednesday's earnings report, but its stock is up only 19% year-to-date versus a 2x move in AMD and 27% in Alphabet. The shift in how AI workloads run (custom silicon at the hyperscalers, more inference-heavy mixes) is feeding doubts about how long the GPU dominance lasts. Amazon's "Titus" project, designed to future-proof its data centers for Nvidia GB200 racks while also accommodating Trainium, is a concrete example.

My take: The market is repricing the dominance question quarter-by-quarter now. The earnings number will be huge on absolutes, but the multiple is compressing because the biggest customers (AWS Titus, Meta MTIA, Google TPU) are all building around the GPU rather than buying more of them. That story will not show up in one earnings cycle; it shows up in the trajectory of long-term purchase commitments.

Closing thought

The thread tying today is the speed of Anthropic's moves. Three discrete events in one day, all pointing at the same posture: own the developer tooling layer, get into the federal conversation on capability, hire the people who shape how the next model gets trained. Worth watching tomorrow whether OpenAI and Google respond publicly or sit on it.