Watching the silence between the candlesticks.
Late last week, a single data point rippled through the analyst circuit: five Chinese AI startups—智谱, DeepSeek, 可灵, Moonshot, and MiniMax—collectively generated an estimated $2.6 billion in revenue during 2024, according to a Menlo Ventures partner's back-of-the-envelope calculation. The number is unverified, unaudited, and almost certainly contested by the companies themselves. Yet it was seized upon by bullish headlines as proof that the Chinese AI ecosystem has finally crossed the chasm from research curiosity to commercial viability.
For those of us who parse liquidity flows for a living, the report read less like a victory lap and more like a diagnostic scan. Because when you apply the same forensic lens that we use on crypto protocols—unit economics, revenue quality, competitive moats—the $2.6B figure begins to reveal fractures that the optimists are missing. And those fractures hold uncomfortable parallels for the crypto industry's own ongoing struggle to translate engineering brilliance into sustainable, high-margin revenue.
Context: The Global Liquidity Map and the Crypto Analogy
To understand why a Chinese AI revenue report matters in a blockchain newsletter, you first have to accept that the two industries share a deeper structural DNA. Both AI and crypto emerged from academic roots, both were fueled by a decade of near-zero interest rates, and both now face the same existential question: Can you build a defensible business model that doesn’t rely on perpetual capital infusion?
In crypto, the question manifests as a handful of L1s and L2s generating fee income while thousands of others burn through treasury. In AI, it manifests as a handful of model providers racing to convert raw compute into recurring revenue. The Menlo report offers a rare snapshot of that conversion rate—or at least, an estimate of it. The five companies span different niches: 智谱 (claims ~$1.0B) leans on Chinese government and SOE contracts; DeepSeek (~$0.5B) runs an aggressive open-source, low-price API strategy; 可灵 (~$0.5B) is tied to Kuaishou’s video ecosystem; MiniMax (~$0.4B) focuses on multimodal consumer apps; Moonshot (~$0.2B) differentiates on long-context reasoning via Kimi.
On the surface, the diversity is a strength. Beneath the surface, it’s a fragmented battlefield with no clear winner—eerily reminiscent of the L2 landscape in 2023, where dozens of rollups competed for the same thin liquidity. The core tension is identical: user adoption is real, but revenue concentration is low, and the cost of acquiring each dollar of revenue may exceed the dollar itself.
Core: Dissecting the $2.6B—A Forensic Structural Analysis
Let’s take each company’s revenue claim and stress-test it through the lens of unit economics, using public data where available.
1. 智谱 ($1.0B) – The Government Contractor
智谱’s revenue is disproportionately tied to bespoke deployments for Chinese ministries and state-owned enterprises. While high-margin in isolation, these contracts are non-recurring, politically contingent, and often bundled with hardware or cloud commitments. In crypto terms, think of them as proof-of-stake delegations from a single whale that could vanish overnight. The real question isn’t whether 智谱 made $1B in 2024—it’s whether it can repeat that without a similar level of government spending. Based on my experience auditing tokenomics for 40+ ICOs in 2017, I’ve seen how easily “government adoption” can be mistaken for organic demand.
2. DeepSeek ($0.5B) – The Cost Leader’s Paradox
DeepSeek’s strategy of near-zero-cost API pricing is the most direct analog to crypto’s “we’ll capture market share first, monetize later” mantra. Its $0.5B revenue implies an astronomical number of token completions—likely in the trillions. But at its published prices (roughly 1/30th of OpenAI’s GPT-4o), the gross margin on API sales is razor-thin or negative when factoring in electricity, hardware depreciation, and personnel. DeepSeek is effectively subsidizing inference to build a data flywheel. This is not sustainable without continuous external capital. In crypto, we saw the same dynamic with DeFi liquidity mining in 2020—high volumes, low retention, and eventual collapse when rewards dried up.
3. 可灵 ($0.5B) – The Platform Trap
可灵 is embedded inside Kuaishou, a publicly traded short-video platform. Its $0.5B may partly represent internal transfer pricing (e.g., Kuaishou paying its own AI division for video generation tools used in advertising). This inflates the headline number without reflecting true external market demand. It’s comparable to a blockchain ecosystem’s foundation reporting fee revenue from its own validator operations—valid as a metric, but not reflective of third-party adoption.
4. MiniMax ($0.4B) & Moonshot ($0.2B) – The Niche Players
MiniMax has built a product (Halo) that leans into social and entertainment, Moonshot into long-context recall. Both have strong product-market fit in narrow verticals, but neither has demonstrated the ability to scale beyond their initial core user base. Their combined $0.6B is less than the annual revenue of a single mid-tier crypto exchange—and their growth rates are far less certain.
When you sum these four caveats, the $2.6B figure becomes less a proof of health and more a conglomerate of fragile, non-repeatable, and low-margin revenue streams. The structural insight here is that revenue diversification can mask revenue fragility. Crypto protocols should take note: high TVL does not equal high fee income, and high fee income does not equal sustainable profit.
Contrarian: The Decoupling Thesis—Why Crypto Doesn’t Need Revenue to Be Valuable
Now for the counter-intuitive angle. Perhaps the obsession with revenue is itself a trap. In crypto, the most valuable assets—Bitcoin, Ethereum—are not judged by revenue per user. They are valued as monetary networks and settlement layers, where security, decentralization, and liquidity are the primary metrics. The same logic applies to AI. DeepSeek’s open-source model, for example, accrues value through ecosystem adoption, not direct billing. Its $0.5B revenue may understate its true network effect, just as Bitcoin’s $1.2T market cap far exceeds its on-chain transaction fees.
This is the decoupling thesis: revenue is not synonymous with value creation in protocol-based economies. The Menlo report implicitly assumes that commercial revenue is the only legitimate proxy for success, but that ignores the possibility that open-source models or infrastructure layers can generate outsized value without capturing it directly. In crypto, we’ve seen this with L1s that charge minimal fees but host billions in value—the value is distributed to token holders, not to the protocol itself.
However, this argument cuts both ways. For a VC-backed startup, revenue is the only path to eventual profitability without continuous dilution. And for most crypto projects that aren’t Bitcoin, the same is true. The decoupling works for pure monetary assets, but for application and infrastructure tokens—like the ones these AI startups would issue if they went the crypto route—revenue remains the single most important indicator of long-term viability.
Takeaway: Positioning for the Next Cycle
What does this mean for a blockchain native in 2026? Three things.
First, the Chinese AI revenue report is a cautionary tale against mistaking top-line growth for bottom-line health. Every DeFi protocol that quotes its TVL or daily volume should be similarly scrutinized: how much of that is organic? How much is subsidized? How much is fake?
Second, the companies that survive this AI commercialization cycle will be those that either achieve genuine high-margin revenue (like 智谱’s government contracts, if recurring) or those that pivot to a platform model where value accrues to an asset, not to an income statement. For crypto, that means doubling down on native value accrual mechanisms—buyback-and-burn, staking yields, fee switching—rather than chasing vanity metrics.
Third, the fragmentation we see in AI mirrors the fragmentation in L2s and cross-chain infrastructure. Too many teams chasing the same marginal user, slicing thin liquidity into ever-smaller pieces. The winners will be the ones who consolidate user attention and capital into a single, defensible point of truth—whether that’s a model, a chain, or a community.
Flow follows the path of least resistance. The $2.6B reported by Menlo Ventures is not resistance—it’s a headline. The real resistance lies in unit economics, customer retention, and the painful grind of turning engineering output into recurring cash flow. That’s a lesson the AI industry is learning now, and one that crypto, despite its advantages, has not yet fully internalized.
Harvesting the liquidity that others overlook. The silence between the candlesticks is where the real data lives. And what it tells me is that $2.6B is a mirage—but the desert beneath it is real, and it is hungry for sustainable oases.
