Chasing shadows in the liquidity fog of 2017. That’s the sensation that hit me when I read the Financial Times report: DeepSeek, the Chinese AI startup known for its cut-throat pricing and open-source models, is raising capital at a pre-money valuation of $71 billion. Seventy-one billion. That number places it just behind OpenAI and Anthropic—a club where membership supposedly requires a GPT-4-level model and a clear path to profitability. But as someone who spent 2017 scraping ICO whitepapers and 2020 coding DeFi arbitrage bots, I can’t shake the feeling that the market is once again pricing hope ahead of substance.
Context: The DeepSeek Phenomenon
DeepSeek isn’t a household name outside of AI circles, but its impact on the model economy has been seismic. Launched in 2023 by a Chinese quantitative hedge fund, the company pioneered a mixed-expert (MoE) architecture that delivered near-GPT-4 performance at a fraction of the cost. Its API pricing—roughly 1/100th of OpenAI’s—forced a wave of price cuts across the industry. Competitors like Alibaba’s Qwen and Baidu’s Ernie had to slash prices by 90% or more to retain developers. DeepSeek also released open-source weights for its V2 model, rallying a global community of builders around a “low-cost, high-performance” ethos.
The $71B figure comes from a pre-money valuation in a new funding round, reportedly led by a mix of sovereign wealth funds and major US venture firms. The round is expected to close at a post-money valuation north of $80B. To put that in perspective: at $71B pre-money, DeepSeek is worth more than the entire market cap of SoundHound AI or half of Palantir. It’s a bet on the thesis that low-cost inference will eat the world—a thesis that sounds eerily familiar to the “decentralization will eat the world” narrative that fueled the 2017 ICO boom.
Core: Dissecting the Incentive Structure
Yields are just risk wearing a disguise, and so are valuations. To understand why $71B might be more hype than reality, I applied the same forensic lens I used in 2017 to analyze ICO tokenomics. Back then, I discovered that 80% of projects had presale allocations designed to dump on retail within six months. Today, I look at DeepSeek’s capital structure and ask: What are the real incentives behind this round?
First, the lack of audited financials. The report offers zero revenue data, no active user numbers, no API call volume. In the crypto world, we demand on-chain verification. For a company worth $71B, the opacity is a red flag. Compare this to Tether, which dominates 70% of the stablecoin market but has never released a full, independent audit. The industry pretends this problem doesn’t exist, and investors pretend they don’t need to see the books. DeepSeek’s valuation might be backed by sophisticated models of future cash flows, but without a quarterly filing, it’s a belief-based asset.
Second, the unit economics of the price war. If DeepSeek’s API pricing is 1/100th of GPT-4, then to achieve even a modest $1B in revenue, it needs 100 times the volume of OpenAI. That’s possible in a market expanding rapidly, but it also means razor-thin margins. My work on cross-border payment corridors taught me that high-volume, low-margin businesses are vulnerable to liquidity shocks. If competitor prices fall further (Alibaba has already matched DeepSeek’s pricing), the path to profitability narrows even more.
Third, the moat question. DeepSeek’s MoE architecture is innovative, but it’s not patented. Google has open-sourced its own MoE research. Meta’s Llama 4 is rumored to use a similar design. The true moat is not the model but the infrastructure optimization—the ability to run inference at 1/100th the cost of competitors. That advantage is real but fragile. It depends on access to cheap hardware (likely Huawei’s Ascend or NVIDIA’s H800) and on continuous software optimizations. In a bull market for AI, venture capital flows to the most aggressive price-cutters; in a down cycle, those same investors demand margin improvement.
Contrarian: The Decoupling Thesis
Correlation is the siren song of fools. Most analysts treat DeepSeek’s valuation as a signal of AI’s inevitable dominance. I see it as a sign of capital decoupling from fundamentals. The macro environment is flooded with liquidity—global central banks are easing, and sovereign wealth funds are desperate for yield. AI startups are the new “digital gold” for asset allocators. This dynamic creates a feedback loop: high valuations attract more capital, which fuels more spending on compute, which drives up costs, which requires even higher valuations to justify future rounds. It’s the same pattern we saw in crypto in 2021: “number go up” becomes the only metric that matters.
My contrarian take: DeepSeek’s $71B valuation is not a vote of confidence in its technology; it’s a vote of confidence in the liquidity cycle. If the macro tide turns—if inflation reignites or a geopolitical shock freezes cross-border capital flows—the valuation could halve overnight. The company has no on-chain revenue, no token to absorb volatility (unlike crypto-native projects that can adjust supply via burns), and no guaranteed path to profitability.
Furthermore, the geopolitical angle is a ticking bomb. DeepSeek is a Chinese company under CCP oversight. Its models must pass the Cyberspace Administration’s security reviews, which may restrict its ability to serve Western customers. Meanwhile, the US is tightening export controls on advanced GPUs. If DeepSeek can’t access the latest chips (B100, B200), its cost advantage may erode against competitors that can. This is the “systemic rot hidden in the fine print” of international trade law.
Takeaway: Position Yourself for the Cycle
History doesn’t repeat, but it rhymes in code. The same forces that inflated crypto valuations in 2017 and 2021 are now inflating AI valuations: abundant liquidity, fear of missing out, and a belief that this time the technology is real. It is real, but the price is not. DeepSeek’s $71B might be the peak of the current AI funding cycle.
For those of us watching from the crypto side, the lesson is clear: look for projects that offer transparency via on-chain mechanisms. AI model usage can be verified through verifiable inference protocols (like those from Ritual or Gensyn). AI compute can be tokenized, allowing investors to participate in the upside without exposure to opaque private valuations. The real breakthrough won’t be a centralized startup at $80B; it will be a decentralized protocol where the incentive structure is visible, auditable, and liquid.
So, as the liquidity fog rolls in again, ask yourself: Are you chasing a reflection, or are you building something that can survive the fog? The answer, as always, is in the fine print.