The code does not lie, but the balance sheets often omit.
Last week, Bloomberg reported that OpenAI, Anthropic, DeepSeek, and six other AI labs are marching toward IPOs between 2026 and 2028. The numbers are staggering: OpenAI at $852 billion pre-money, Anthropic at $965 billion, DeepSeek at $71 billion. Yet the underlying data—cumulative fundraising, revenue models, compute costs—paints a geometry that demands forensic scrutiny.
As a crypto security audit partner who has dissected over 200 token launches and DeFi protocols, I see a pattern: the same incentive misalignment that led to the 2022 L1 collapse is now being layered into AI equity narratives. The market is pricing hype cycles, not sustainable cash flows. Zero trust is not a policy; it is a geometry.
Context: The AI IPO Wave as a Mirror of Crypto Tokenomics
The report covers ten AI labs across the US and China that are actively preparing for public listings. OpenAI has raised $180 billion cumulatively, Anthropic $132 billion, DeepSeek $7 billion. Their valuations range from 4.7x to 7.3x cumulative funding, suggesting investors are paying a premium for future dominance rather than current earnings.
But here’s the catch: none of these companies have disclosed detailed financial statements. The report itself admits it lacks data on revenue, EBITDA, customer concentration, or cash burn rates. This is precisely the informational asymmetry that plagued crypto ICOs in 2017 and DeFi token mints in 2021. The code does not lie, but it often omits.
Perplexity, valued at $21 billion after raising a paltry $200 million in its latest round, exemplifies the disconnect. Its subscription model is real, but the search market is dominated by Google. The valuation implies a $15 billion revenue run rate—a multiple that defies even optimistic projections. This is not analysis; it’s speculation dressed as data.
Core: Systematic Teardown—Valuation Anomalies and Structural Risks
Let me run the numbers through a forensic lens.
First, the OpenAI vs. Anthropic paradox. OpenAI raised 36% more capital than Anthropic, yet Anthropic’s valuation is 13% higher. Why? The report speculates it’s a “safety premium” or a bet on Anthropic’s constitutional AI approach. As a security auditor, I see a different vector: structural illiquidity. Anthropic’s early backers (including Google and Salesforce) may have negotiated liquidation preferences that inflate the perceived price. In crypto, we call this a “vanity valuation”—a number set by a handful of private transactions with no secondary market validation. The same happened with Terra Luna’s $40 billion peak before the collapse.
Second, the Chinese discount. DeepSeek’s $71 billion valuation is only 10x its $7 billion raised, compared to OpenAI’s 4.7x. This suggests the market is pricing in significant regulatory and geopolitical risk. But is the discount sufficient? DeepSeek’s MoE architecture and open-source strategy could justify a higher multiple if it captures enterprise adoption in China. However, the report reveals zero revenue data. Compiling the truth from fragmented logs tells me that investors are betting on narrative, not fundamentals.
Third, the compute cost overhang. The report completely omits infrastructure expenditures. Training GPT-4 cost approximately $100 million in GPU compute alone. Inference adds ongoing operational expenses. For a company like OpenAI, if its $180 billion cumulative funding is consumed mostly by compute (which it likely is), its IPO capital raise will be used to service existing debt, not to innovate. In crypto, we call this a “burning platform”—a company that needs to sell equity to keep its lights on.
I’ve seen this playbook before. During the 2020 DeFi summer, projects like YFI reached $40 billion valuations with zero revenue and no product. When the hype faded, the price collapsed 95%. The AI IPO wave is following a similar trajectory, but with a slower fuse—a multi-year runway of private funding that postpones the reckoning.
Contrarian: What the Bulls Got Right
To be fair, the bulls see a transformative technology. AI is not a speculative token; it has real-world utility in healthcare, finance, and defense. OpenAI’s ChatGPT has over 100 million weekly active users. Anthropic’s Claude powers enterprise chatbots for major banks. These are not vaporware.
Furthermore, the IPO timelines themselves are a positive signal. Unlike most crypto projects that list tokens within months of conception, these companies are taking 5-7 years to go public. That indicates operational maturity: they have built teams, products, and customer bases. The code does not lie, but it often omits the timeline of infrastructure development.
Also, the Chinese discount may be an opportunity. If DeepSeek captures a significant share of China’s enterprise AI market (projected to be $100 billion by 2030), its $71 billion valuation would be a steal. The same applies to Moon’s Dark Side (Kimi) and Baichuan. The report notes that Chinese labs face chip export restrictions, but they are pivoting to domestic alternatives like Huawei Ascend. This is analogous to crypto projects building on Solana after Ethereum’s congestion—a forced migration that can lead to innovation.
Takeaway: Demand On-Chain Proof Before Allocating
The AI IPO wave is the most consequential capital market event since the dot-com bubble. But as a security auditor, I must remind readers: security is the absence of assumptions.
Before buying any of these IPOs, demand the same level of transparency we require in crypto audits: verifiable revenue streams, auditable compute costs, and independent security certifications for model alignment. The report provides none of this. The market is currently paying for optionality, not evidence.
If any of these companies experience a catastrophic model failure (hallucination-driven lawsuit, regulatory shutdown, or compute shortage), the valuation will reprice overnight. Compiling the truth from fragmented logs, I predict that within 3 years of IPO, at least one of these ten will trade below its private funding round valuation. The rest will survive, but only those that treat security as a geometry—not a policy—will thrive.
