The data suggests an anomaly. Anthropic, the AI safety-first lab, carries a post-money valuation of 9650 billion RMB. OpenAI, the household name, sits at 8520 billion. The difference is 1130 billion—roughly the market cap of a mid-tier blockchain protocol. This inversion defies logic unless you dig into the capital markets subtext.
Let me be precise. These numbers come from a recent analysis of Chinese and US AI large model companies preparing for IPO between 2026 and 2028. The source is a seven-dimension framework analysis—I stripped it to its raw signals. The result is a picture of a market segmenting along fault lines most coverage ignores.
Code does not lie, but it rarely speaks plainly. Neither does valuation. The numbers tell one story; the missing data tells another.
Context first. The AI industry is transitioning from 'funding frenzy' to 'capital exit.' OpenAI has raised 1800 billion RMB cumulative. Anthropic 1320 billion. Yet the latter’s valuation is higher. That alone should raise flags. Perplexity, a search product, is valued at 210 billion. China’s DeepSeek—open-source, cost-efficient—at 710 billion. The timelines are tight: OpenAI and Anthropic likely late 2026, Chinese firms 2027-2028.
I have spent years auditing Layer 2 protocols. The patterns are eerily similar. Hype inflates base layer tokens. The real friction is underneath.
Core analysis. Let me force-rank the critical technical gaps.
First: Valuation multiples without revenue. The analysis explicitly flags that no company disclosed income, EBITDA, or user numbers. OpenAI’s valuation-to-funding ratio is 4.7x. Anthropic’s is 7.3x. DeepSeek’s is 10x. But this is not a bullish signal—it reflects market enthusiasm for future cash flows, not current earnings. In crypto terms, this is the equivalent of a project at 100x TVL with zero fees. The S-1 documents will be the first real stress test. Based on my experience with zero-knowledge rollup audits, I know that a missing transaction cost can break an entire economic model. Here, missing revenue data is the canary.
Second: Infrastructure dependency is invisible. The analysis notes zero mention of GPU clusters, cloud contracts, or chip supply. Yet this is the single largest cost driver. OpenAI and Anthropic rely on NVIDIA H100/B100. Chinese firms face export controls. The analysis gives a medium-low confidence on this dimension, but I believe it is the highest unquantified risk. In blockchain, we call this the 'rollup sequencer bottleneck'—if the hardware fails, the L2 stops. Here, if compute costs rise 20% year-over-year, valuation multiples compress.
Third: The China discount is mispriced. DeepSeek at 710 billion versus Anthropic at 9650 billion—a 13x gap. But DeepSeek has raised only 70 billion, so its multiple is higher. The market is pricing in a risk premium for regulatory opacity, chip access, and lower brand recognition. Yet DeepSeek’s open-source strategy could be a moat if enterprise adoption scales. The analysis misses a key point: tokenization of compute credits could bridge this gap. A blockchain-based AI compute market would make infrastructure transparent.
Contrarian angle. The analysis flags ethical and security concerns as a 'major omission' with low confidence. I disagree. The real blind spot is the missing integration protocol between AI and existing financial infrastructure. These companies are not just going public—they are being forced into a traditional equity framework that cannot account for model decay, alignment shifts, or training data sources. In crypto, we handle this with on-chain governance and slashing. In traditional finance, it’s a fiduciary failure waiting to happen. The contrarian play is not to short the IPOs, but to long the decentralized AI infrastructure tokens that solve these verification problems.
Furthermore, the analysis identifies a potential bubble risk but does not quantify it. Let me do that. If OpenAI’s revenue is, say, 10 billion RMB (a guess, but plausible), the P/S ratio at 8520 billion is 852x. For reference, NVIDIA’s P/S in 2024 was ~35x. The AI IPO wave is priced for perfection. Any delay in GPT-5 or a regulatory clampdown will trigger a 30-50% correction. The infrastructure IPOs (compute, data center REITs) will be more resilient.
Takeaway. Beneath the friction lies the integration protocol. The AI IPOs will either force a convergence with blockchain-based verification (proof-of-inference, decentralized training) or expose the gap. My forecast: by 2028, at least one major AI company will issue a security token for compute credits. The market will demand transparency that only cryptography can provide. The clock is ticking.
