The AI Landscape: April 2026
The major labs enter Q2 in very different positions.
The field looks different again. OpenAI enters April with a fully assembled IPO narrative, backed by the largest private funding round in history and operational numbers to match. Anthropic is recovering from a confrontation with the Pentagon that cost it a $200 million defense contract and from a month of operational failures that tested whether the company can run a reliable service at scale. Google, Meta, and DeepSeek are each on trajectories of their own.
Ready to launch
OpenAI closed $122 billion at an $852 billion valuation on March 31. The three largest commitments came from Amazon, Nvidia, and SoftBank. Amazon alone pledged $50 billion, with $35 billion contingent on an IPO or the achievement of artificial general intelligence, while Nvidia and SoftBank each committed $30 billion. For the first time, OpenAI raised $3 billion from individual investors through bank channels and secured inclusion in several ARK Invest ETFs, broadening the shareholder base in advance of a public offering that the company has signaled for later this year.
OpenAI reported $2 billion per month in revenue, having generated $13.1 billion last year, with 900 million weekly active users and more than 50 million subscribers. The enterprise business accounts for 40 percent of that total, up from roughly 30 percent last year, and the company expects parity with the consumer side by year-end. The ads pilot, which launched only weeks ago, has already crossed $100 million in annualized recurring revenue, a line of business that did not exist two months ago.
OpenAI also acquired TBPN last week, a daily tech talk show with roughly 70,000 viewers per episode that has become a regular stop for CEOs including Mark Zuckerberg and Satya Nadella. The show will report to chief global affairs officer Chris Lehane, and its advertising business will wind down under the new structure. CNN drew the parallel to RCA creating NBC in 1926 to sell radios, and The Information founder Jessica Lessin offered her own summary: Musk has X, and Altman now has TBPN. GPT-5.5, codenamed Spud, completed pretraining in late March, with Altman indicating release within weeks.
Cracking under pressure
Anthropic held two red lines in its Pentagon contract, insisting that Claude could not be used for autonomous weapons or for the mass surveillance of American citizens. Defense Secretary Pete Hegseth gave Dario Amodei a deadline of February 27 to drop the restrictions and allow the model to be used “for all lawful purposes.” Amodei refused. Trump directed federal agencies to cease using Anthropic products, and Hegseth designated the company a supply-chain risk, a classification that the Pentagon had previously reserved for firms tied to foreign adversaries. OpenAI announced its own Pentagon contract hours later.
Anthropic sued the government on March 9, arguing that the designation violated its First Amendment rights and jeopardized hundreds of millions of dollars in contracts. A federal judge in California blocked the designation on March 26, finding the Pentagon’s actions “troubling” and noting that the government’s own lawyers admitted that Hegseth’s enforcement language had “absolutely no legal effect.” The public dispute also triggered a wave of Claude downloads that pushed the app to the top of the charts in multiple markets. Anthropic’s infrastructure buckled under the load.
On March 26, the same day as the injunction, a misconfigured content management system left roughly 3,000 internal files publicly accessible. Among them was a blog post in draft form about Claude Mythos, a new model tier above Opus. The post warned that Mythos posed “unprecedented cybersecurity risks.” Five days later, a source map accidentally included in a routine Claude Code release exposed 512,000 lines of source code across 1,900 files, revealing internal model codenames, an “Undercover Mode” that strips Anthropic attribution when employees contribute to external projects, and further details about Mythos. Service outages hit on March 2, 11, 25 through 27, 31, and April 4, with the late-March disruptions coinciding with the user surge from the Pentagon controversy.
On April 4, Anthropic cut off subscription-based access for third-party agent tools, starting with OpenClaw. An estimated 135,000 OpenClaw instances were running at the time of the announcement, and users received less than 24 hours to adjust. OpenClaw’s creator, Peter Steinberger, had joined OpenAI in February, and the timing drew accusations that Anthropic had absorbed popular features into its own closed harness before locking out the open-source alternatives. In the same week, Anthropic formed AnthroPAC, its first political action committee, described as employee-funded and bipartisan. Amodei tried to hold the moral high ground with the Pentagon on autonomous weapons and surveillance, and the PAC suggests that he has concluded that moral authority alone does not confer political protection.
The best of the rest
Gemini 3.1 Pro leads 13 of 16 benchmarks on the Artificial Analysis Intelligence Index, tied with GPT-5.4 Pro at the top. Google expanded Personal Intelligence across Search, Chrome, and the standalone Gemini app during March, connecting securely to Gmail, Photos, and other Google services for personalized results. The company also shipped migration tools that let users import chat history and preferences from competing AI assistants directly into Gemini.