The AI IPO Wave Is The Dot-Com Bubble’s Evil Twin. Anthropic’s Delay Just Proved It.

By Anurag Gurtu

When a corporation postpones its public offering to present better figures to investors, that represents a tactical move. When the New York City pension fund publicly requests a delay over AI safety apprehensions, your primary competitor’s chief executive labels his own public offering “ill-advised” and defers it to at least next year, technology stocks decline due to worries surrounding open-weight models, and your own prospectus warns of “existential risks to humanity” stemming from the merchandise you are vending, that no longer qualifies as a strategy.

That is an entire marketplace transmitting a distinct signal.

Anthropic’s financial figures remain extraordinary. Its third-quarter 2026 revenue reportedly reached $16.8 billion—reflecting a 45% increase quarter over quarter. This indicates an approximate annualized revenue run rate of $67 billion for the enterprise.

What the S-1 actually says

The following figures derive from Anthropic’s leaked prospectus:

• $4.6 billion: Anthropic’s full-year 2025 revenue.
• $7.33 billion: The amount Anthropic expended on compute and infrastructure exclusively in 2025—exceeding its overall revenue.
• $42 billion: Anthropic’s net loss for 2025. The upcoming roadshow will promptly clarify that $34 billion of this total constituted a noncash accounting charge—specifically an adjustment regarding the estimated valuation of financing instruments capable of eventually converting into shares.
• $518 billion: The minimum financial outlay Anthropic projects over a decade toward AI infrastructure across six partners, which include Google, Microsoft, and Amazon, based on the confidential IPO prospectus evaluated by Reuters. Roughly 80% of this aggregate sum remains noncancelable or payable regardless of active utilization.
• $2 trillion: The public offering valuation Anthropic continues to target for November, establishing a metric equivalent to 435 times its 2025 revenue.

An additional statistic from the S-1 that the roadshow will likely minimize: A full quarter of Anthropic’s 2025 revenue originated from merely two unnamed customers.

Three signals the market is sending

While the IPO postponement serves as the primary headline, it remains far from the sole indicator requiring evaluation.

Signal one: NYC Comptroller Mark Levine. Last week, the administrator overseeing New York City’s pension funds—which currently hold roughly $7 million in Anthropic exposure, though potential holdings could expand significantly should Anthropic enter the Russell 3000 index post-IPO—dispatched a formal letter to Anthropic CEO Dario Amodei urging a complete halt to the public offering.

The cited hazards mirror those explicitly highlighted within Anthropic’s own prospectus: progressively autonomous artificial intelligence systems capable of functioning unpredictably, generating security vulnerabilities, and, quoting the document directly, presenting “existential risks to humanity.”

Signal two: Sam Altman delays OpenAI’s IPO. OpenAI has confirmed it will bypass public markets throughout 2026. Chief Executive Altman deemed an IPO under current circumstances “ill-advised” in light of artificial intelligence safety worries. Although this framing involves strategic positioning—casting OpenAI as the more prudent participant—it nonetheless delivers a genuine signal when the leader of the world’s second-most valuable AI enterprise evaluates prevailing conditions and opts to wait.

Signal three: The chip selloff. The previous week’s $2 trillion semiconductor stock selloff stemmed partially from anxiety surrounding the expenditures required for AI infrastructure buildouts. Forecasts from Gartner help elucidate this apprehension: Through 2030, the research firm projects a maximum of $3.6 trillion in revenue generated by AI data centers, compared against the $8.5 trillion necessary to cover planned investments, financing expenses, and a 12% return.

The open-weight problem

The underlying structural obstacle remains immune to any level of Q3 revenue expansion: Anthropic’s $2 trillion valuation necessitates that closed-model artificial intelligence retain a durable pricing premium over open-weight alternatives. That premium is currently experiencing systematic dismantling.

Open-weight artificial intelligence continues to challenge the economic foundations supporting closed-model valuations. Enterprise adoption climbed from 23% to 67% in 2026, with corporations documenting cost reductions ranging between 70% and 90%. DeepSeek R1 delivers frontier-tier performance across the majority of enterprise benchmarks at a rate of 7 cents per million input tokens; independent assessments place Anthropic’s comparable input and output pricing at approximately 18 times and 23 times higher, respectively.

Because DeepSeek’s weights remain freely downloadable, enterprises gain the ability to self-host without executing model-provider API agreements or incurring per-token billing structures, particularly as the capability disparity with closed models reportedly narrows within six-month intervals following a frontier release.

Faced with this environment, New Constructs—the research organization that categorized Anthropic’s proposed offering as “an unprecedented test of investor gullibility”—computed that substantiating a $2 trillion valuation would demand an immediate 10% after-tax operating margin maintained over a decade, extraordinary revenue acceleration, and $358 billion in annual after-tax operating profits by 2035.

Why this is worse than 1999

The dot-com bubble possessed a legitimate justification, as nobody understood what the internet would ultimately evolve into. The artificial intelligence public offering wave of 2026 lacks that rationale entirely.

Global expenditures directed toward artificial intelligence are projected to achieve $2.59 trillion in 2026, marking a 47% year-over-year increase according to Gartner figures. Nevertheless, a July 2025 MIT NANDA study estimated that 95% of the examined organizations derived zero return from generative AI. Concurrently, KPMG released findings earlier this year indicating that a mere 7% of surveyed executives reported establishing a proven return on investment from AI initiatives.

This provides the framework wherein Anthropic encourages investors to pay 435 times 2025 revenue for an entity burdened by $518 billion in locked infrastructure commitments, a 25% revenue concentration tied to two unnamed clients, and a pricing moat that open-weight models consistently close within six-month cycles.

Shifting the timeline from October to November alters none of these underlying metrics. It merely appends a single additional quarter of revenue performance.

What actually survives this correction

The reality that no artificial intelligence IPO roadshow will articulate publicly is this: The most prominent enterprise AI victors will not necessarily control the most costly model. Instead, they will govern the architecture capable of translating any model into regulated, auditable, and quantifiable results—leveraging proprietary enterprise information, domain-specific fine-tuning, multiagent orchestration, and operational controls that render autonomous execution dependable.

This intrinsic value remains independent of maintaining scarce model capabilities or achieving a $2 trillion valuation. It compounds directly within production environments, where customers recognize that the underlying model does not constitute the product; the deployment itself does.

The dot-com crash did not materialize because the internet was inherently flawed. It occurred because the pricing was incorrect. Artificial intelligence investors confront an identical distinction: Genuine technology does not automatically validate its market valuation. A robust financial quarter alongside a one-month schedule adjustment fails to bridge that chasm. November simply affords Anthropic additional time to rationalize the arithmetic.

---

Anurag Gurtu, the co-founder and CEO of Airrived, is an entrepreneur and product leader with more than 20 years of experience across cybersecurity and AI. He has built three generative AI companies, including one that reached unicorn status. Gurtu is known for turning advanced AI into scalable, real-world platforms, and previously contributed to successful exits, including acquisitions by Splunk, Tripwire and Sun Microsystems, and a public company, FireEye. He holds a master’s degree from the University of Southern California and CISSP certification.

?Frequently Asked Questions

01Why did Anthropic delay its IPO?

Anthropic faced mounting market pressures, including public requests from the New York City pension fund to pause the offering over AI safety concerns, alongside broader industry caution highlighted by competitors and sliding tech stocks.

02What are the primary financial concerns raised in Anthropic’s prospectus?

Key concerns include heavy net losses, massive long-term commitments for compute and infrastructure, significant revenue concentration among a small number of customers, and high valuations relative to projected returns.

03How do open-weight models impact closed-model valuations?

Open-weight models offer enterprise-grade performance at a fraction of the cost and allow self-hosting, threatening the pricing power and market dominance required to justify multi-trillion-dollar valuations for closed-model providers.

---

Related Crunchbase queries:

• Global IPOs For Venture-Backed Companies In 2026
• Global Venture Funding To AI Startups In 2026

Illustration: Dom Guzman