Anthropic Sets $900 Billion Deadline.

Mazen Al Sabbagh

— 8 min read
Share

TL;DR

  • Anthropic accelerates its final private raise: The company issued a 48-hour window for investor allocations, targeting a $50 billion raise at a $900 billion+ valuation, expected to close within two weeks. This round likely precedes an October IPO. (TechCrunch)
  • Pentagon excludes Anthropic from classified AI deals: Seven other firms—SpaceX, OpenAI, Google, Microsoft, Nvidia, AWS, and Reflection—secured contracts to deploy models on classified DoD networks. Anthropic remains "blacklisted" due to supply chain concerns, despite the NSA's separate use of its Mythos model. (Reuters)
  • Musk admits xAI distilled OpenAI models: Elon Musk testified that Grok was "partly" trained using OpenAI's models, impacting his Polymarket win odds in the ongoing lawsuit. This admission complicates his core complaint against OpenAI. (TechCrunch)
  • Apple posts record quarter, flags AI memory costs: Reporting $111.2 billion in revenue and $29.6 billion in profit, Apple also authorized a $100 billion share buyback. However, the company warned of rising memory expenses, a direct consequence of the escalating AI chip market. (CNBC)
  • Anthropic surpasses OpenAI in LLM revenue: Counterpoint Research indicates Anthropic now holds 31.4% of global LLM revenue, narrowly exceeding OpenAI's 29%. This is achieved with a significantly smaller user base, yielding $16.20 ARPU compared to OpenAI's $2.20, suggesting a market shift from user volume to monetization efficiency. (The Register)

Lead Story: Anthropic's Capital Ascent: A $900 Billion Private Round Accelerates

Late Wednesday, TechCrunch disclosed that Anthropic has set a 48-hour deadline for investor allocation requests in what is poised to become the largest private funding round in history. The targeted $50 billion raise, at a valuation exceeding $900 billion, could finalize within two weeks. This trajectory positions Anthropic to potentially surpass OpenAI's $852 billion post-money valuation from March, securing its status as the world's most valuable private enterprise.

The market tension is palpable. Business Insider reported aggressive secondary market activity, with existing Anthropic shareholders fielding numerous inquiries and prospective investors deploying intermediaries to secure allocations. Even institutional players prepared to commit $5 billion faced challenges in securing direct engagement by Wednesday evening. This suggests a unique demand profile, reflecting both confidence in the asset and a scarcity of access.

The financial metrics underpinning this valuation surge are substantial. Anthropic’s annual revenue run rate now exceeds $30 billion, approaching $40 billion, propelled by its Claude Code and the Cowork enterprise platform. Counterpoint Research data for Q1 2026 shows Anthropic capturing 31.4% of global LLM revenue, marginally ahead of OpenAI's 29%. Crucially, Anthropic's efficiency is distinct: it extracts $16.20 in average monthly revenue per active user, significantly outpacing Microsoft's $5, OpenAI's $2.20, and Google's $1.10. This indicates a highly effective monetization strategy within a concentrated user base.

This round is widely anticipated as Anthropic’s final private capital infusion before a projected October 2026 IPO. Fortune highlighted a notable structural dependency: approximately half of Google’s and Amazon’s recent "blowout AI profits" derived from unrealized gains on their Anthropic stakes, rather than operational performance. Both are major Anthropic investors, with Google committing up to $40 billion and Amazon $25 billion. This feedback loop—hyperscalers investing in Anthropic, Anthropic consuming hyperscaler compute, and hyperscalers subsequently booking valuation gains—is a critical component of the current AI market architecture. Its sustainability hinges directly on continued underlying revenue expansion, which, for now, remains robust.

The Pentagon's announcement on Friday (detailed below) introduces a strategic paradox: Anthropic is poised to close the largest private funding round ever, yet remains excluded from U.S. military contracts. The leading LLM revenue generator, with demonstrated cybersecurity capabilities, is concurrently classified as a supply chain risk by its own government.

In Other News

Pentagon Signs Classified AI Deals with Seven Companies — Anthropic Still Blacklisted. The Department of Defense announced agreements with seven AI companies—SpaceX, OpenAI, Google, Microsoft, Nvidia, Amazon Web Services, and Reflection—to deploy their models on classified DoD networks at Impact Levels 6 and 7. Anthropic was conspicuously absent. CNBC confirmed that Pentagon CTO Emil Michael reiterated Anthropic's "blacklisted" status as a supply chain risk, even while acknowledging the unique national security utility of its Mythos model. The NSA is reportedly employing Mythos for Microsoft vulnerability assessments, as per a Wednesday Bloomberg report. This dichotomy underscores a policy impasse: Anthropic's refusal to permit "all lawful purposes," including autonomous weapons, for Claude has barred it from direct DoD contracts, even as its capabilities are leveraged indirectly. CNN suggested this move to onboard Anthropic's competitors could enhance the Trump administration's leverage in the dispute. These tools will be accessible via GenAI.mil, which serves over 1.3 million DoD personnel. (Reuters)

Apple Posts Record Quarter, Warns of AI-Driven Memory Costs. Apple announced a record fiscal second quarter, achieving $111.2 billion in revenue—a 17% increase—and $2.01 EPS, surpassing consensus estimates. Services revenue reached a record $31 billion, and the company authorized its largest-ever share buyback at $100 billion. CEO Tim Cook highlighted the MacBook Neo and strong iPhone 17 performance. However, Cook also pointed to "significantly higher memory costs" anticipated for the current quarter and beyond, directly attributable to the global AI chip boom’s impact on DRAM and HBM supply. Mac supply constraints are expected to persist for several months. Post-market, shares rose 3.6%. These results reveal a nuanced market dynamic: Apple, not a primary driver of the AI infrastructure buildout, is nonetheless incurring a cost burden from its accelerated demand for critical components. (Bloomberg)

Musk's Distillation Admission Turns the Trial Narrative. Day 4 of the Musk v. OpenAI trial delivered a critical admission from the plaintiff. Under cross-examination, Elon Musk conceded that xAI had "partly" utilized OpenAI's models to train Grok, a practice of model distillation. Wired reported Musk's defense that this is "standard practice," yet the admission undermines his foundational complaint regarding OpenAI's alleged betrayal of its original mission. The incongruity of suing OpenAI for commercialization while replicating its intellectual property was not lost on market observers; Polymarket win odds for Musk subsequently dropped to 42%. Musk concluded his testimony after three days. His wealth manager, Jared Birchall, was then called, with questions centering on the documented conditions of Musk's initial donations to OpenAI—a key aspect of the breach-of-trust claim. Separately, WIRED reported that trial evidence revealed Shivon Zilis, a long-time Musk associate, had acted as a covert liaison between Musk and OpenAI for years post-2018. (TechCrunch)

OpenAI Restricts GPT-5.5 Cyber — Days After Criticizing Anthropic for Doing the Same. Sam Altman announced that OpenAI’s forthcoming cybersecurity model, GPT-5.5 Cyber, will be rolled out exclusively to "critical cyber defenders" through a vetted, restricted-access program. This mirrors the strategy Anthropic employed with Mythos, a decision Altman publicly criticized as "fear-based marketing" just ten days prior. TechCrunch's headline captured the reversal, highlighting OpenAI's adoption of the very playbook it had derided. Altman stated OpenAI would engage "public institutions and industry stakeholders" for a gradual expansion of access. This policy pivot is the clearest indication yet that frontier cybersecurity models are reaching a capability threshold necessitating restricted deployment, irrespective of their developer. India Today observed Altman "doing the same with Cyber, for very similar reasons." (TechCrunch)

X / Social Pulse

The Pentagon's contract announcements dominated Friday's social discourse. The strategic dissonance—Anthropic leading LLM revenue, securing a $900 billion valuation, being used by the NSA against Microsoft vulnerabilities, yet simultaneously "blacklisted" by the Pentagon—drew sharp critique regarding the coherence of national AI policy. Defense and policy commentators underscored Anthropic as the sole frontier lab to decline the Pentagon’s "any lawful use" terms, leading to its exclusion. The inclusion of SpaceX, given Elon Musk's concurrent lawsuit against OpenAI while his company signs a classified AI deal alongside OpenAI, amplified the perceived ironies of the week.

The Atlantic's midday analysis, "Maybe AI Isn't a Bubble After All," garnered significant attention, identifying Claude Code and AI agents as pivotal revenue drivers. This, combined with Counterpoint data revealing Anthropic's $16.20 ARPU versus OpenAI's $2.20, effectively shifted the market conversation from the feasibility of AI monetization to its current distribution and efficiency. Industry unity was briefly observed as Anthropic and OpenAI jointly backed the Warner-Budd workforce data bill (Politico), a rare alignment on regulatory issues. Musk's distillation admission continued to circulate, prompting legal speculation regarding potential counterclaims from OpenAI.

One to Watch

The $1 Trillion Year Is Coming Into View

CNBC reported Thursday that both Evercore and Bank of America now project Big Tech AI capital expenditures will exceed $1 trillion in 2027. This follows revised capex guidance from this week's earnings calls. The 2026 figures are already substantial: $725 billion across Alphabet, Amazon, Meta, and Microsoft alone, marking a 77% increase from 2024. Including Apple, Nvidia’s own investments, and second-tier cloud providers, Forbes estimates the 2026 total approaches $750 billion.

Analyst consensus is shifting from questioning sustainability to contemplating an endpoint. Alphabet and Microsoft explicitly signaled further capex increases for 2027. Tom's Hardware quoted one analyst dismissing the bear thesis on AI infrastructure as "garbage." However, the fundamental question posed by Fortune—"No one knows where the buildout ends"—persists. Memory chip shortages now contribute billions to budgets; Microsoft attributed $25 billion of its record capex directly to memory price inflation, indicating that a portion of the spending increase is cost-driven rather than purely capacity-driven. The trillion-dollar threshold, once a theoretical projection, is now a foundational planning assumption for the global technology architecture.

Quick Hits

  • Congress passed a 45-day FISA Section 702 extension just before midnight Wednesday, averting a lapse and deferring critical debate on warrantless, AI-enabled surveillance until mid-June. The Senate had rejected the House’s three-year bill, citing an attached CBDC ban. (ABC News)
  • The Senate Judiciary Committee unanimously advanced the GUARD Act, proposing age verification and parental controls for AI chatbots used by minors. This represents the latest legislative push for AI safety measures impacting younger users. (Let's Data Science)
  • A Chinese appeals court ruled against AI-driven dismissal, deeming the replacement of a tech worker with AI as unlawful. This marks one of the earliest judicial precedents globally concerning AI’s impact on workforce restructuring. (NPR)
  • Google is exploring advertising within its standalone Gemini app. PCWorld reported executives are "circling the idea," a monetization strategy that could fundamentally reshape the economic model for conversational AI platforms. (PCWorld)
  • Nvidia B300 server prices in China have nearly doubled to $1 million, primarily due to U.S. restrictions tightening gray-market chip supply. Concurrently, Huawei projects $12 billion in AI chip revenue for 2026, a 60% increase from 2025, as Chinese firms pivot towards domestic Ascend processors. Chinese AI models now constitute 32% of global token usage, a significant rise from 5% a year prior. (Reuters)

April concluded with the AI sector exhibiting dynamics few could have forecast even six months prior. Anthropic is poised to finalize the largest private funding round in history, already outpacing OpenAI in revenue, all while simultaneously being designated a supply chain risk by its own government. On Friday, the Pentagon executed classified AI agreements with seven of Anthropic's competitors, including OpenAI and SpaceX, even as the NSA continues to leverage Anthropic's own models to assess Microsoft vulnerabilities. The entity that previously cautioned against Mythos's capabilities is now observing its primary rival adopt an identical restricted-access deployment strategy. Furthermore, Elon Musk, whose SpaceX just secured a Pentagon AI contract, spent the week under oath admitting his AI venture copied the very company he is suing. The trial reconvenes next week, as do these broader market forces.

Sources

Lock in. M. mazen@thorterminal.com

Read more