In late July, the White House accused a Chinese AI startup of stealing frontier capabilities from an American lab, Congressional sanctions legislation advanced, and the prospect of federal restrictions on open-weight AI models drew nearer. More than 270 American technology companies responded with a letter urging Washington to leave those models unrestricted. The coalition’s breadth, which stretched from Nvidia and Microsoft through Palantir and Y Combinator to hundreds of startups, mapped a division within the industry that the national-security framing had obscured.
Open weights, closed ranks
Open-weight AI models publish their trained parameters for anyone to download, inspect, modify, and run on local hardware. Closed models keep those parameters proprietary and sell access through metered APIs. On July 24, a coalition led by Nvidia, Microsoft, and Meta published an open letter titled “Open Weights and American AI Leadership”, urging policymakers to avoid “premature restrictions” on open-weight releases. Jensen Huang shared the letter in his first-ever post on X, and Satya Nadella endorsed it the same day. The letter compared open-weight AI to the open-source software movement of the 1980s and argued that restricting model distribution would push development offshore and concentrate capability in a few providers. The signatory list grew from 25 companies at launch to more than 270 within two weeks, spanning chipmakers, cloud providers, enterprise software vendors, venture firms, and open-model labs.
OpenAI and Anthropic, the two labs most identified with closed proprietary models, were absent from the original signatories. OpenAI joined within days; Anthropic did not, and the silence fueled accusations that the company favored restrictions to protect its commercial position. Two days before the coalition letter, the Little Tech Association had sent its own letter to the White House on behalf of 179 startups, including Y Combinator, addressed to OSTP Director Kratsios and Commerce Secretary Howard Lutnick. That letter framed open access to foreign open-weight models as a matter of commercial survival for American companies that depend on cheaper alternatives to frontier API pricing.
Creating dependency
Palantir CEO Alex Karp opened the commercial front on July 1, telling CNBC that enterprise AI’s token-based pricing model had “gone completely wrong.” The frustration had acquired a label. Tokenmaxxing became shorthand for the escalating per-token costs that enterprises absorb while routing proprietary data through frontier API endpoints. Karp framed the arrangement as subsidizing a competitor’s training pipeline. Palantir announced an expanded partnership with Nvidia to build custom open-weight models for US government agencies, coupling the criticism with a product alternative.
On earnings day August 3, Karp escalated, telling CNBC that frontier labs were “trying to drug addict us to a future they believe they control.” Palantir’s own Q2 results reported 93% year-over-year revenue growth and 149% growth in US commercial revenue, a trajectory suggesting that enterprise AI spending increasingly favors the deployment layer. Karp signed the open-weight letter in late July, aligning Palantir with infrastructure providers whose commercial interests converge on maximizing independent model deployment.
The monopolist’s lament
Amodei published a response on July 27, writing that Anthropic had “never advocated for a ban on open-weights models.” He proposed three measures: tighter export controls on advanced chips, enforcement against industrial-scale distillation, and mandatory safety testing for sufficiently capable models regardless of whether their weights are published. These proposals tracked Anthropic’s commercial position. Chip controls and anti-distillation enforcement protect the investment required to train frontier models, while capability-based safety testing applies to open and closed models alike.