A Menlo Ventures partner just dropped a bombshell: China’s top five AI startups collectively hit $2.6 billion in annualized revenue. Zhipu at $1B, DeepSeek at $500M, Kling at $500M, Moonshot at $200M, MiniMax at $400M. The industry cheered. Investors started FOMO-ing. But here’s what nobody wants to admit: these numbers are built on sand.
⚠️ Deep article forbidden
I’ve spent 22 years in this space—first as a coder auditing EOS wallets in 2017, then as a journalist navigating the Terra crash in 2022. I’ve learned one iron rule: when a single source drops a headline revenue estimate with no audited financials, you run the other way. This analysis isn’t about AI—it’s about every crypto project that’s ever claimed “we generated $X in fees” without proof. The same structural flaws apply.
Context: Why this story matters in crypto
The $2.6B figure comes from Deedy Das, a Menlo Ventures partner. He has no publicly disclosed methodology. No breakdown of revenue streams. No verification with independent auditors. Yet the crypto press ran with it. Sound familiar? It’s the same pattern as when “$XXX million in TVL” was used to pump Uniswap clones in 2020, or when Terra’s “$2B in Ozone insurance” was cited without verifying collateral.
The crypto market is currently sideways—the perfect breeding ground for unverified narratives. When prices aren’t moving, analysts manufacture catalysts. Revenue figures become FOMO bait. The Deedy Das post is a perfect case study in how a single, unsourced tweet can move markets. We need to deconstruct it, not worship it.
Core: The three lies hidden in the revenue estimate
Lie #1: Revenue is not profit. Deedy’s numbers are gross revenue, not net. For crypto protocols, this is like quoting Uniswap’s total swap fees without deducting gas costs, L1 security subsidies, and operational overhead. Zhipu’s $1B likely includes government contracts and cloud revenue—non-recurring, low-margin stuff. DeepSeek’s $500M at their API prices (2-10% of OpenAI’s) implies razor-thin unit economics. In crypto, this is the equivalent of reporting Tether’s $4.5B Q1 2023 profit without noting that 90% came from treasury bill yield—a temporary macro tailwind, not a sustainable business model.
Lie #2: Top-line growth hides concentration risk. Deedy grouped five companies, but the spread is massive: Zhipu is 38% of the total. If Zhipu loses a single enterprise client or a government contract expires, the whole narrative collapses. In crypto, this is analogous to claiming “DeFi had $50B in total value locked in Q2 2023” while knowing that 60% sat in Lido and MakerDAO—two smart contracts vulnerable to fork or regulatory action. Concentration masks fragility.
Lie #3: Revenue timing is everything. Deedy’s estimate covers a twelve-month period, but was it calendar year 2024 or trailing twelve months? Did it include pre-seed contracts that won’t renew? In crypto, revenue timing is even trickier: a single airdrop campaign can inflate a protocol’s fee generation for a quarter, then disappear. Axie Infinity once reported $1B in monthly revenue during its peak. That’s not recurring revenue. That’s a bubble.
⚠️ Deep article forbidden
During the 2020 Compound yield farming crisis, I decoded cToken interest-rate models live on Twitter Spaces to calm panicked investors. I learned that panic-proof communication starts with clear, actionable data. Deedy’s post does the opposite—it triggers FOMO without giving readers the tools to verify. As a community, we must demand the same transparency from crypto revenue claims.
Contrarian: The $26B figure is actually bearish for crypto
Here’s the counter-intuitive take: even if the estimate is accurate, it’s terrible news for crypto AI tokens. Why? Because the revenue is concentrated in centralized companies (Zhipu, DeepSeek) that don’t need blockchain. They’re using traditional cloud infrastructure. They’re raising fiat VC money. They’re proving that AI is a centralized winner-takes-most game—exactly the opposite of what crypto AI projects (Render, Akash, Bittensor) promise. Investors chasing the AI crypto narrative should realize that the real AI revenue is being captured by Web2 companies, not Web3.
This is the same trap as when people claimed “blockchain will disrupt supply chains” while Maersk and IBM built TradeLens (and later killed it). The biggest revenue in a new technology often flows to incumbents, not to new decentralized networks. If you believe the $26B figure, you should short crypto AI tokens and buy NVIDIA stock. Period.
Furthermore, Deedy’s estimate highlights the fundamental flaw in all revenue-based valuations in nascent markets: they fail to account for the cost of acquiring those dollars. In crypto, the cost is often hidden in token emissions. In AI, it’s hidden in government subsidies and investor capital. Zhipu’s $1B probably required $2B in training compute and marketing. That’s not a business—it’s a loss leader. The same goes for many DeFi protocols that pay out 80% of their fee revenue as staking rewards. Revenue is a vanity metric when you can’t subtract the cost of capital.
⚠️ Deep article forbidden
Takeaway: What to watch next
Instead of celebrating the $26B headline, ask these three questions: 1. Can the top five companies provide unit economics (revenue per user, marginal cost per API call)? 2. What percentage of their revenue is recurring (subscriptions) vs. one-time (deployments)? 3. How much of that revenue is from related-party transactions (e.g., parent company subsidies)?
For crypto AI projects, the real signal is not revenue—it's developer retention on smart contracts, node operator count, and compute utilization rates. Ignore the CEO’s tweet about “$X revenue.” Look at the on-chain data.
In a sideways market, stories are the only thing that moves prices. Don’t buy the story without checking the facts.
_This article is based on personal analysis and should not be considered investment advice. Always DYOR._