The Pentagon and the Market Have Simultaneously Sentenced Anthropic | Rewire News Evening Report

The DOJ qualitatively characterized Anthropic as an "unacceptable national security risk" in a 40-page document, while on the same day Axios data showed that Anthropic's enterprise market share had surpassed that of OpenAI. The government is punishing it in principle, while the market is rewarding its product.

1｜DOJ Characterizes Anthropic as an "Unacceptable Risk," Yet Enterprise Customers Are Voting with Their Wallets

The U.S. Department of Justice filed a 40-page document with the Northern District of California federal court on Tuesday, stating that Anthropic has the ability to "disable or modify technology in wartime according to its interest," thus posing an "unacceptable risk" to the national defense supply chain. The document used the term "Department of War," a preferred term of the Trump administration, redefining Anthropic's technical ethics clause as a strategic vulnerability. Hearing on March 24.

On the same day, according to Axios, Anthropic's enterprise AI spending share has climbed from less than 40% in early December to around 40% currently, while OpenAI has dropped from 50% to 27%. Anthropic has an annualized revenue of $19 billion, OpenAI $25 billion, but according to Epoch AI estimates, Anthropic's annual growth rate is nearly three times that of OpenAI and is on track to surpass it by mid-year.

The morning reported the industry's support for Anthropic in court documents. The evening update was the government's official response and the market data that emerged on the same day. Both "judgments" arrived simultaneously, pointing in completely opposite directions. The government is saying that "a bottom-line AI company is a security risk," while enterprise customers are saying that "a bottom-line AI company is a more reliable supplier."

(Source: Bloomberg / Axios / Epoch AI / Al Jazeera)

2｜NVIDIA GTC Day Three, Jensen Huang Personally Hosts Open-Source Camp Gathering

The last day of GTC is not about hardware but about the ecosystem. Jensen Huang personally hosted a roundtable on open models, with LangChain founder Harrison Chase, A16Z, AI2, Cursor, and Thinking Machines Lab representatives sitting across from him. The core question under discussion was whether open models can catch up to closed-source cutting edge.

A more concrete action was the formation of the Nemotron Coalition. This alliance brings together eight institutions: Cursor, LangChain, Mistral AI, Perplexity, Black Forest Labs, Reflection AI, Sarvam, and Thinking Machines Lab. The first project is a basic model jointly developed by Mistral and NVIDIA, trained on DGX Cloud, and released to the open-source ecosystem.

On Monday, we talked about hardware specs; on Tuesday, we discussed the inference grid; on Wednesday, we covered the open-source alliance. Over these three days, NVIDIA is not just building a chip business but creating a complete AI ecosystem based on its own computing resources. The underlying implication of the Open Model Alliance is that all participants must run on NVIDIA's computing power.

(Source: NVIDIA Newsroom / Tom's Hardware / TechCrunch)

3｜Mistral Makes Waves at GTC, Forge Empowers Enterprises to Train Their Own Models from Scratch

According to TechCrunch, Mistral unveiled the Forge platform at GTC, allowing enterprises to train custom models from scratch using Mistral's open-weight model library. Initial customers include Ericsson, the European Space Agency, and Singapore DSO. Simultaneously released was Mistral Small 4 with a 119B parameter MoE architecture, with only 6B activated parameters, under the Apache 2.0 license, achieving 40% faster inference speed than its predecessor.

The logic of Forge is completely opposite to OpenAI and Anthropic's API model. The latter sells black-box inference, while Forge sells end-to-end control from data to weights. For sovereignty-sensitive clients in Europe and Asia, the phrase "I own the model" is more valuable than performance scores.

(Source: TechCrunch / VentureBeat / Dataconomy)

4｜AI Layoffs Shift from Singular Events to an Industry Template

According to Fortune, analysts predict that Zuckerberg is now "finishing what Dorsey started." In late February, Jack Dorsey laid off 40% of Block's employees (4,000 people), foretelling that "most companies will make the same structural decision within a year." Two weeks later, Atlassian cut 1,600 jobs, with over 900 in R&D roles. CEO Mike Cannon-Brookes stated that "pretending AI doesn't change the required skill set and job count is dishonest." Meta's plan to lay off 20% of its workforce, about 16,000 people, surfaced last week.

From 2026 to date, over 55,000 tech industry workers have been laid off. Dorsey's template is the most radical version, cutting nearly half of the workforce, resulting in a 17% stock price increase, indicating a rewarding market response. When layoffs become synonymous with "AI transformation," the replacement of labor costs with computing resources is no longer an isolated decision but a structural reset being replicated across the industry.

(Source: Fortune / TechCrunch / Bloomberg)

Also Worth Knowing ↓

Bitrefill confirms North Korea Lazarus Group breach, 18,500 purchase records leaked. According to CoinDesk, the attack began with an old credential on an employee's laptop, allowing the attacker to access the production environment and steal hot wallet funds. Bitrefill stated that it will cover the losses with operational funds. The security vulnerability of the crypto infrastructure was not on-chain but on the employee's computer that was not promptly decommissioned. (Source: CoinDesk / Bitcoin Magazine)

Bitcoin ETF sees nearly $1 billion inflows since March 9, with BTC consolidating around $74,000. The price has risen by 12% during the same period, but on-chain data shows a surge in exchange inflows with selling pressure near the $75,000 resistance level. Bitcoin Core released the v31.0rc1 beta on Wednesday. (Source: Fortune / CoinGecko)

The robotics sector raised over $12 billion in a single week, expected to surpass $200 billion by 2026. The number of AI funding rounds above $100 million generated in the first two and a half weeks of March has exceeded any comparable period in venture capital history. Capital is shifting from language models to real-world automation. (Source: AI Funding Tracker)

The FOMC interest rate decision will be announced at 2 a.m. Beijing time tomorrow. The market expects the rate to remain unchanged at 3.5%-3.75% (probability over 92%), with focus on the dot plot and Powell's press conference wording. Powell's term ends on May 15, marking his second-to-last press conference as Fed Chair. (Source: CNBC / Kiplinger)

Google Gemini saw a 258% YoY growth in paid users, surpassing Claude's 200%. However, in the enterprise market share, Anthropic is widening the gap. The competition in AI between the consumer and enterprise segments is evolving into two distinct battles. (Source: Axios / UCStrategies)