Hook
A single tweet from Menlo Ventures partner Deedy Das has sent ripples through both the AI and crypto communities: China’s top five generative AI startups—Zhipu, DeepSeek, Kling, Moonshot, and MiniMax—are collectively generating an estimated $2.6 billion in annual revenue. For a sector often dismissed as a playground for venture-backed hype, this figure demands a hard look. But here’s the twist: these aren’t just AI companies. They are the vanguard of a new paradigm where decentralized inference, tokenized compute, and verifiable provenance converge. And the blockchain industry should be paying close attention.
Context
The narrative of Chinese AI has long been overshadowed by US dominance. OpenAI, Google, and Anthropic command the spotlight. Yet behind the Great Firewall, a different story is unfolding. Zhipu (estimated $1B revenue) has cemented itself as the state-backed champion, securing government and enterprise contracts. DeepSeek ($500M) shocked the market by proving that an open-weight, extreme-cost-efficiency model can generate real revenue—a thesis that resonates deeply with the blockchain ethos of permissionless access and sovereignty. Kling ($500M), the video generation arm of Kuaishou, Moonshot ($200M, known for Kimi’s long-context capabilities), and MiniMax ($400M, focused on multimodal social apps) round out the list. But why should a crypto reader care? Because these startups are quietly laying the infrastructure for a verifiable, decentralized AI layer.
Core Insight
The revenue numbers themselves are less important than the business model dynamics they reveal—dynamics that mirror the tokenomics of decentralized protocols. Take DeepSeek: its API pricing is up to 90% cheaper than OpenAI’s, achieving scale by subsidizing inference through aggressive optimization and, likely, negative unit economics. This is exactly the playbook of high-L1 chains—selling cheap blockspace to capture mindshare and developer mindshare, hoping network effects will drive future monetization. Moonshot’s Kimi, with its 200K+ token windows, has become the go-to long-context assistant for Chinese developers, similar to how Polygon became the go-to sidechain for Ethereum users seeking low fees. Zhipu’s government contracts resemble private consortium chains—secure, compliant, but lacking the permissionless innovation of public networks.
Importantly, Deedy Das’s estimate lacks granularity: we don’t know the split between API calls, on-premise deployments, cloud partnerships, and subscription revenue. But even assuming a conservative 70% cost of goods sold, the implied gross profit of ~$780 million validates that Chinese AI has crossed the chasm from science project to commercial viability. For blockchain builders, this is a critical signal. The same forces driving these AI revenues—cloud commoditization, open-source momentum, and government adoption—will shape the next wave of on-chain AI services.
Contrarian Angle
The conventional wisdom is that these numbers signal a healthy, thriving ecosystem. I see a different risk: the race to zero on API pricing is unsustainable unless accompanied by a tokenized incentive layer that aligns usage with value accrual. DeepSeek’s low pricing is a classic loss-leader, but without a native token to capture future upside from developer lock-in, it remains vulnerable to margin compression. Compare this to protocols like Bittensor (TAO) or Akash Network (AKT), where compute providers are paid in tokens, creating a feedback loop between demand, staking, and income. China’s AI startups operate in a fiat-only world—they burn investor capital to subsidize users, with no token to absorb the volatility or reward early adopters. If the bear market in venture funding deepens, these models may collapse under their own weight. The contrarian truth: the revenue numbers are impressive, but the unit economics are fragile. A blockchain-native AI competitor could undercut them further by using token emissions as a subsidy, then capture network effects.
Takeaway
Code has conscience. And if the Chinese AI giants want to sustain their growth into 2027, they will need to embrace decentralized verification—not just for ethical reasons, but for economic survival. The question is not whether Web3 can learn from Web2 AI; it’s whether Web2 AI can afford to ignore Web3’s token-based incentive design. Liquidity flows where belief resides. My belief is that the next billion-dollar AI startup will be built on a blockchain, and the seeds are already being sown in Beijing, Hangzhou, and Shenzhen.
