Trillion Dollar Anthropic Overtakes OpenAI.

Mazen Al Sabbagh

— 8 min read
Share

TL;DR

  • Anthropic's market valuation: Surpassed $1 trillion on secondary markets, notably Forge Global and Caplight, eclipsing OpenAI's $880 billion. This rise is underpinned by a 233% quarterly revenue surge to an annualized $30 billion.
  • OpenAI's GPT-5.5 "Spud" launch: The first fully retrained base model since GPT-4.5, released to paid subscribers. It registered 82.7% on Terminal-Bench 2.0, narrowly outperforming Mythos Preview's 82.0%.
  • DOJ pauses Anthropic appeal: The Department of Justice requested a pause in its legal challenge against Anthropic concerning the Pentagon standoff, signaling a potential shift toward political resolution over continued litigation.
  • White House accuses China: Allegations surfaced of "industrial scale" AI intellectual property theft by China from US labs, weeks before a scheduled Trump-Xi summit. Concurrently, new export control measures targeting Nvidia chip loopholes advanced in the House.
  • Tesla's strategic capex increase: Tesla disclosed a $25 billion capital expenditure plan for 2026, a 25% increase over prior guidance, intensifying its pivot into AI and robotics. Intel was also secured as its inaugural 14A chip client.

Lead Story: Anthropic Crosses $1 Trillion, Overtakes OpenAI for the First Time

The AI landscape has recalibrated, as Anthropic now holds the position of the industry's most valuable private entity. This valuation ascendancy is particularly notable given the strategic decision to restrict public access to its most advanced model.

Secondary market transactions this week saw Anthropic's valuation exceed $1 trillion on platforms like Forge Global, with some offers reaching $1.15 trillion. Concurrently, OpenAI's shares on these same platforms traded at approximately $880 billion. This marks the first instance Anthropic has surpassed OpenAI in any valuation metric, signifying a fundamental shift in market perception.

This revaluation is grounded in exceptional financial performance. Anthropic's annualized revenue accelerated from $9 billion at the close of 2025 to $30 billion by March 2026—a 233% increase within a single quarter. Claude Code alone contributes over $2.5 billion annually. The company's client portfolio now includes more than 1,000 entities spending over $1 million annually, alongside eight Fortune 10 companies. This aggressive market penetration and rapid scaling define a new benchmark for hyper-growth business architecture.

It is crucial to contextualize these secondary-market dynamics. Glen Anderson of Rainmaker Securities noted to QuantoSei that offers for shares at valuations up to $960 billion are "snapped up within a day," indicating extreme scarcity and high demand rather than solely fundamental valuation. Reports even suggest property has been offered in exchange for shares. Despite this fervor, Anthropic's planned October 2026 IPO targets a valuation between $400 billion and $500 billion, implying a significant disparity between current secondary market pricing and institutional investment bank projections.

This divergence between public market anticipation and disciplined IPO guidance is itself instructive. The confluence of the Mythos controversy, the Amazon $25 billion investment, the Glasswing enterprise initiative, and the ongoing Pentagon standoff has cultivated a narrative premium. Investors are actively pricing in this strategic narrative before the underlying economics fully mature. The sustainability of this premium through October will depend on both revenue trajectory maintenance and whether Mythos evolves into a defensible strategic moat or a regulatory liability.

In Other News

GPT-5.5 "Spud" Ships — OpenAI's First Fully Retrained Base Model Since GPT-4.5.

The 93% Polymarket odds proved accurate. OpenAI launched GPT-5.5 late Wednesday, deploying the model to all paid ChatGPT and Codex subscribers (Plus, Pro, Business, Enterprise), with API access slated to follow. OpenAI President Greg Brockman characterized it as "more intuitive," capable of achieving more with reduced human oversight. A GPT-5.5 Pro variant is also available for Pro subscribers at $100/month.

Benchmarking results are competitive, yet not decisively dominant. GPT-5.5 recorded 82.7% on Terminal-Bench 2.0, marginally surpassing Mythos Preview's 82.0%, while comfortably exceeding Opus 4.7 (69.4%) and GPT-5.4 (75.1%). Broader evaluations, as reported by Inc., indicated GPT-5.5 matched or exceeded human performance on approximately 85% of benchmarked tasks, compared to Opus 4.7's 80% and GPT-5.4's 83%. OpenAI classified the model's cybersecurity capabilities as "High" under its preparedness framework; a notable designation, but falling short of the "Critical" threshold that precipitated Mythos's restricted deployment. The New York Times framed the launch as OpenAI adopting "a more open approach to cybersecurity than its chief rival, Anthropic." Startup Fortune termed it "a precision upgrade that bets reliability will beat raw scale." This model arrives seven weeks post-GPT-5.4 amidst a challenging environment: a criminal probe, a looming trial, a competitor that just surpassed its market valuation, and a Mythos-centric narrative requiring immediate counter-positioning.

DOJ Blinks: Trump Administration Asks to Pause Its Own Anthropic Appeal.

Two distinct filings late Tuesday have significantly altered the legal posture of the Anthropic-Pentagon dispute. First, the DOJ petitioned U.S. District Judge Rita Lin to halt its appeal—an appeal initiated after Lin's temporary injunction last month blocked the Pentagon's supply-chain-risk designation. Separately, Anthropic informed an appeals court that it possesses no "kill switch" for Claude once deployed within classified Pentagon networks, directly refuting the Pentagon's previous claims. The Boston Herald reported Anthropic's contention that the Pentagon's designation constitutes "illegally retaliating" with a measure intended for foreign adversary sabotage. Collectively, these actions indicate a trajectory towards a negotiated settlement. The administration is de-escalating litigation while White House-brokered talks (which President Trump himself deemed "possible" on Tuesday) proceed. The May 19 Ninth Circuit oral arguments remain scheduled but may ultimately become moot.

White House Accuses China of "Industrial Scale" AI Theft.

The White House accused China of "industrial scale" theft of AI intellectual property from American research facilities. This allegation, conveyed in an OSTP memo by Director Michael Kratsios and reported by the Financial Times Wednesday, carries significant geopolitical weight. Axios indicated that this accusation could complicate the upcoming Trump-Xi summit in Beijing, with Fox Business suggesting the timing is deliberately pre-summit. This claim coincides with the House Foreign Affairs Committee advancing new export control measures to close Nvidia chip loopholes, and follows testimony before the Senate Judiciary Committee by a former Google engineer convicted of stealing AI trade secrets for China. The administration's dual stance—advocating for AI containment from China while de-escalating a dispute with the developer of America's most advanced cyber-capable model—underscores a complex and potentially contradictory policy framework.

Microsoft Integrates Mythos into Its Security Development Lifecycle.

Microsoft announced the embedding of Claude Mythos Preview within its Security Development Lifecycle (SDL), the foundational framework for building and hardening all Microsoft software. Reuters confirmed this integration, and CSO Online reported that this move signifies "a clear signal that powerful AI models are making inroads into real software security work." This represents the first public commitment by a Glasswing partner to integrate Mythos into a production security pipeline, moving beyond mere testing. In parallel, Google at Cloud Next unveiled AI-driven security agents designed to mitigate Mythos-era threats, implicitly establishing the model as a new competitive benchmark for adversaries.

X / Social Pulse

GPT-5.5's late-afternoon release effectively reoriented the narrative from what had been an Anthropic-centric day. Initial reactions immediately polarized: developers lauded the Terminal-Bench improvement, while skeptics pointed to GPT-5.5's marginal lead over Mythos on a single benchmark, emphasizing Mythos's continued restricted access. The New York Times' framing—positioning OpenAI as the "open" cybersecurity player against Anthropic's restrictions—furnished Altman with a counter-narrative to his "fear-based marketing" broadside earlier this week. The long-term efficacy of this positioning depends on the model's performance in real-world application over the coming 48 hours.

Tesla's $25 billion capex disclosure dominated earnings-related X discussions overnight. Bloomberg highlighted the 25% increase over previous guidance. Reuters noted it "tests investor faith in unproven AI bets." Musk also confirmed Intel as Tesla's first major 14A chip customer, representing a critical lifeline for Intel's struggling foundry division and a foundational compute strategy for Tesla's Optimus robot program.

One to Watch

April 27 is converging as a nexus of critical legal and political deadlines. FISA Section 702—a surveillance authority whose capabilities civil liberties groups warn are critically amplified by AI—expires on the same day that jury selection commences in Musk v. Altman. Congress recessed after a 10-day extension, having failed to resolve whether to grant a long-term renewal, an 18-month bridge, or allow the authority to lapse. The AI dimension is significant: the Lummis-Wyden coalition has cautioned that Section 702, combined with contemporary AI, facilitates "unprecedented mass surveillance." Should Congress reconvene Monday without a resolution, the program will lapse at midnight Sunday—precisely as twelve Oakland jurors are being impaneled to adjudicate whether the foundational premise of the company that just shipped GPT-5.5 was a deception.

Quick Hits

  • Tesla's 2026 capex disclosure: $25 billion, a 25% increase over prior guidance, to fund initiatives in self-driving taxis, Optimus robots, and AI infrastructure, as per Bloomberg. While revenue exceeded expectations, the stock saw minimal movement.
  • DeFi protocol security breaches: $606 million has been lost across 12 hacks on Ethereum and Solana since April 1, according to Motley Fool. Analysts, however, attribute these exploits to conventional vulnerabilities, not Mythos-class threats.
  • Sunak's AI employment warning: Former UK PM Rishi Sunak asserted that AI is "already leading to fewer jobs for young people," validating graduates' concerns regarding entry-level employment, as reported by the BBC.
  • Pentagon's AI budget request: The Pentagon requested $54 billion for AI-powered warfare capabilities in its latest budget, per The Guardian. Experts simultaneously warn of the military's insufficient preparedness for autonomous drone program risks.
  • Anthropic's European infrastructure scaling: Anthropic is actively hiring for European data center capacity negotiation to support its UK expansion to 800 staff, as reported by CNBC.

Wednesday delivered the predictable collision of market and geopolitical forces the calendar had promised. Anthropic secured a $1 trillion secondary market valuation by morning; OpenAI countered with GPT-5.5 by afternoon. This new model narrowly outperforms Mythos on Terminal-Bench but enters the market devoid of the access restrictions that have made Mythos both strategically formidable and commercially elusive. The DOJ's decision to unilaterally pause its appeal, without judicial compulsion, is the strongest indication yet that the Pentagon standoff is transitioning from a courtroom battle to a negotiated settlement. This aligns with the White House's accusation of industrial-scale AI theft by China weeks before a Trump-Xi summit: the political rationale for insulating America's most advanced cyber model from national defense networks becomes unsustainable when simultaneously alleging Beijing is actively acquiring the technology through illicit means. Microsoft's SDL integration provides Mythos with its first production security deployment, establishing a new benchmark for enterprise risk management. Meanwhile, Brockman's diary entry—"it was a lie"—will inevitably cast a long shadow over the trial commencing Sunday. April 27—FISA expiration and jury selection—is now just four days away.

Sources

Lock in. M. mazen@thorterminal.com

Read more