The day in one read

The Anthropic IPO and the Economics of AI Infrastructure

The dominant narrative of the day centers on the imminent initial public offering of Anthropic, which is expected to trade between November 10 and 20, and the broader implications for the AI infrastructure market. The company’s S-1 filing reveals a business model characterized by extreme efficiency: while serving compute spend increased 2.3x year-over-year, revenue grew 14x, indicating a 7x efficiency gain in inference. Anthropic reported 85% gross margins, 20% free cash flow margins, and 8% GAAP operating margins. However, the balance sheet presents a complex picture, listing approximately $108 billion in convertible notes payable. This figure is not new debt but the mark-to-market appreciation of roughly $11 billion in converts issued to strategic investors like Google and Amazon, who used this structure to avoid ownership thresholds that would trigger consolidated financial reporting. Consequently, the P&L shows an $80 billion net loss for the first six months of 2026, 100% of which is driven by the non-cash appreciation of these converts rather than operational burn.

Market reaction to the IPO has been mixed, with secondary markets pricing the company at $1.6 trillion, while institutional investors argue the valuation should be closer to $1 trillion. Prediction market odds for a pre-Thanksgiving IPO fell from 85% to 60% following rumors of delay. The IPO is viewed by analysts as a critical stress test for the sector, highlighting a potential maturity mismatch where long-term compute commitments, totaling nearly $800 billion by 2036, are funded by short-term model revenues. Anthropic has $855 billion in committed compute deals, which annualizes to roughly $100 billion per year, and 80% of its revenue for the three months ending September came from models six months old or younger. This reliance on frontier status underscores the urgency of maintaining technological leadership, a challenge exacerbated by Meter’s recent safety report, which concluded that the pace of model improvement has roughly doubled over the past year due to AI-assisted research.

Midterm Elections and Data Center Political Risk

The 2026 US midterm elections, resulting in a Democratic takeover of the House and Senate, triggered significant volatility in equity markets, particularly in the semiconductor and AI sectors. The sell-off was driven by fears that Democratic control of key states like Texas and Ohio would halt data center construction, increasing AI costs and slowing adoption. Approximately 40 ballot initiatives across 10 states voted against datacenter expansion, potentially halting up to $250 billion in projects, or 25-30% of planned investment. This political shift threatened to reduce chip demand, directly impacting Nvidia’s 75% margins and causing a sharp decline in the Philadelphia Semiconductor Index, which fell 12% for the week. The Nasdaq 100 dropped over 6 percent, and the S&P 500 fell 4.4%, marking the largest single-day decline since March 2020.

Meta President Dina Powell-McCormick warned that the industry has failed to adequately explain the benefits of data center investments to local communities, describing them as the visible manifestation of AI that residents may reject. To mitigate this political risk, Meta employs a strategy of paying for its own electricity, directing tax surpluses to local services, and securing bipartisan support. However, analysts argue the market reaction was overblown, noting that data centers approved for 2027 and early 2028 remain secure due to two-year approval lead times. The correction was attributed to the liquidation of leveraged positions rather than a fundamental bubble burst, with recovery expected ahead of Nvidia's November 17 earnings. The election results also signaled a voter rejection of the Iran conflict, boosting de-escalation hopes and lowering oil prices, while the shift to a Democratic-controlled Congress implies increased US debt to negotiate with President Trump, a factor viewed as positive for gold.

Enterprise AI Adoption and Market Valuations

Despite the market volatility, enterprise AI adoption is accelerating faster than expected, driven by agentic capabilities rather than seat-based licensing. Menlo Ventures' State of Enterprise AI report, based on 400 U.S. decision-makers responsible for $4 billion in spend, revealed that enterprise AI spending grew from $33 billion in 2025 to $177 billion in 2026, a 5.4x growth rate. Average spend per enterprise rose 4.7x, and enterprises spent 2.5 times their budgeted amounts. Anthropic captured 55% of business API spend, up from 40%, while OpenAI recovered to 31% and Google fell to 9%. Frontier model spend share rose from 35% to 63%, and coding agent revenue for Codex and Claude Code jumped from $1 billion to $24 billion.

This surge in adoption contrasts with skepticism regarding private market valuations. Scott Wilson, CIO of the Washington University in St. Louis endowment, expressed doubt that companies like Anthropic and OpenAI will achieve their private valuations in public markets. Wilson’s strategy targets a 50/50 split between public and private assets, with a bias against mega-funds that do not need endowment capital. He argues that concentrated direct investing and co-investments are superior to traditional fund-of-funds models, noting that without these direct positions, the fund would have merely matched peer group averages over the last decade. The endowment has grown from just under $8 billion to nearly $20 billion over nine years, generating $17 billion in profits. Wilson identifies his biggest mistake as failing to add capital to winning positions due to frame-of-reference bias, a decision he estimates cost the fund billions.

AI Security and Architectural Risks

The rapid deployment of AI agents has raised significant concerns regarding security vulnerabilities and data privacy. Meredith Whitaker, president of Signal, argues that consumer AI agents are currently untrustworthy due to "perilous architectures" that grant root permissions and break down security sandboxes. She cites specific incidents, including a zero-click exploit that pulled corporate data from Microsoft Co-pilot and a prompt injection via a Reddit comment that hacked Perplexity’s agentic browser. Whitaker highlights that in 11% of Anthropic’s Claude Chrome experiments, the agent failed to resist prompt injection despite guardrails. She contends that agents like Meta’s Muse are logging contact lists and creating dossiers, effectively hollowing out application-layer privacy.

Neil Chilson, head of AI policy at the Abundance Institute, counters that agents flip the traditional internet bargain by using user data for the user’s benefit rather than for ad sales. He argues that agents provide "agency" by automating tasks previously reserved for the wealthy, such as managing finances or scheduling. Chilson demonstrates that agents can enhance privacy; he used Muse to opt out of five major U.S. data brokers with a six-word prompt. He acknowledges security risks but compares them to historical issues like floppy disk viruses, arguing that liability laws and market incentives will drive companies to build safer systems. Both agree that current systems are flawed but disagree on whether the convenience and potential for improved privacy outweigh the immediate risks of prompt injection and data exfiltration. OpenAI’s identification of 943 vulnerabilities in the Linux kernel further underscores the transition period of technical debt, where AI-generated patches are increasingly accepted after manual fixes fail.

Also Notable

In the longevity sector, Mark Hyman of Function Health argues that muscle is the "currency of longevity," recommending 45-minute strength training and 120 to 180 grams of protein daily for those over 60. Function Health tests over 110 biomarkers, identifying that 70% of users are deficient in at least one nutrient, and offers add-ons like the Gallery multi-cancer detection test and the Azure Scan whole-body MRI. In the crypto space, a $6.34 billion Binance hack was attributed to North Korea via a compromised developer laptop and Ethereum EIP-770; $2.85 billion was frozen, and 98% of user funds were reimbursed. Ethereum has accelerated its quantum-resistant roadmap to 2028, while D-Wave stock fell from $38 to $10.

In corporate news, The New York Times company has reportedly entered exclusive negotiations to acquire Letterboxd for $330 million, a platform that grew from 3 million users in January 2021 to 28 million by April of this year. Amazon is launching a feature allowing customers to chat with its Alexa AI agent via text messages to facilitate shopping, a rollout that precedes the holiday season. IONOS CFO Patrick Heider outlined the company's strategy to leverage AI agents for its seven million SMB customers, reporting 320,000 net new customers in 2024 and guiding for an EBITDA margin of 37-38% for the year. Tesla discontinued the Cybertruck, which sold only 11,500 units, to focus on robotics and AI, while Apple reportedly reduced employee AI spending caps from $300 to $200 daily.