The chart whispers; the ledger screams the truth.
In the first half of 2025, a bizarre divergence emerged from the East. Tencent, the $400 billion Chinese tech mammoth, reported a 12% earnings beat — yet its AI lab, the crown jewel of its next-gen strategy, is facing a cash squeeze. Simultaneously, DeepSeek, the independent research lab that stunned the world with its R1 model, is struggling to keep the lights on.
This is not a story of technology failure. It is a story of capital structure failure. And for anyone watching the intersection of macro liquidity, AI capex, and crypto’s DePIN narrative, it’s a warning signal.
Context: The Two Roads to Intelligence
To understand the funding gap, you must first understand the two distinct technical routes behind these names.
Tencent’s AI lab — actually a collection of at least three entities: the 2016-founded AI Lab (basic research), the Hunyuan model team (product), and various embedded teams inside WeChat, gaming, and cloud — follows an ecosystem-driven, multi-modal approach. It builds models that can be immediately plugged into WeChat’s 1.3 billion monthly active users, into gaming NPCs, into ad algorithms. The feedback loop is data-rich, but the upfront cost is monstrous: compute, talent, and infrastructure spread across multiple verticals.
DeepSeek, by contrast, is a research-driven, efficiency-obsessed lab. Its MoE (Mixture of Experts) architecture achieved near-frontier performance on a fraction of the compute budget of GPT-4 or Llama 3. The R1 series, released open-source in early 2025, proved that a Chinese team could out-innovate with less. But “less” is relative. DeepSeek’s operating costs are still in the tens of millions of dollars per year, and its revenue is essentially zero. Open-source does not pay the electricity bill.
Both entities face the same structural headwind: China’s restricted access to advanced AI chips (NVIDIA H100/H800, B200) due to US export controls. But their ability to navigate that headwind is wildly different.
Core: The Macro-First Liquidity Lens
Let me layer my traditional macro lens onto this picture.
Capital flows where intelligence meets speed. That’s a signature line I use often, and it applies directly here. The intelligence is the AI model performance; the speed is the ability to deploy capital efficiently. Tencent has speed through its existing revenue streams. DeepSeek does not.
Tencent’s gaming division alone generated $8 billion in Q1 2025. Its fintech and cloud unit contributed another $5 billion. Every dollar spent on AI compute can be cross-subsidized by a dollar earned from a gamer buying a skin in Honor of Kings or a brand paying for a WeChat mini-program ad. The AI lab does not need to be profitable today. It just needs to prove that the cross-subsidy logic holds.
DeepSeek has no such luxury. Its parent company, High-Flyer Quant (a quantitative hedge fund), is profitable, but not at the scale of Tencent. And the AI lab’s cash burn rate — estimated at $30–50 million annually for compute alone, based on my analysis of similar-sized labs — is a significant draw on High-Flyer’s resources. The recent funding rumors suggest DeepSeek is seeking external capital, but the valuation expectations clash with the reality of zero revenue. The market is telling them: “Your technology is amazing, but show me the business model.”
This is a classic situation I call structural fragility. The lab’s technical moat is real, but its financial moat is nonexistent. In the crypto world, we’ve seen this before: Luna’s algorithmic stability was mathematically elegant, but its capital base was a single point of failure. When the external liquidity dried up, the elegance meant nothing.
The ledger screams the truth. Tencent’s ledger shows a diversified income stream funding a capex-heavy AI unit. DeepSeek’s ledger shows a single revenue source (High-Flyer’s trading profits) funding a pure research lab. The structural difference is stark.
Contrarian: The Decoupling That Isn’t Happening
The conventional wisdom in crypto and AI circles is that “technology democratizes access.” The narrative says that open-source models like DeepSeek’s R1 will level the playing field, allowing smaller players to compete with giants. The contrarian view I’m putting forward is: the opposite is happening.
We are witnessing a re-concentration of AI compute power around institutional moats. Tencent, Alibaba, ByteDance, and Baidu — the Big Four of Chinese tech — are building massive GPU clusters (10,000+ H100-equivalent cards each) and amortizing the cost across their ecosystems. DeepSeek, despite its engineering brilliance, cannot afford a cluster of that scale. The chip export controls make it even harder to buy the latest hardware at any price.
The result? The gap between the “best model” and the “best capitalized model” is widening. Tencent’s Hunyuan may not be as technically advanced as DeepSeek’s R1 on a per-parameter basis, but Tencent can run 10x more inference queries, collect 100x more user feedback, and iterate twice as fast. Over 12 months, that speed advantage translates into a model quality advantage.
History does not repeat, but it rhymes in code. This rhymes with the 2010s cloud computing wars: Amazon Web Services had the first-mover advantage and the capital, so it crushed smaller competitors not by technology alone, but by the ability to invest in infrastructure over a longer horizon.
In the crypto context, this is a direct parallel to the Bitcoin mining centralization debate. ASICs were supposed to be available to anyone, but the capital requirements for a modern mining farm (tens of millions for machines, cheap power, and cooling) have concentrated hashrate among a few large pools. The same is happening in AI compute.
Takeaway: The DePIN Signal for Crypto Investors
So why should a crypto investor care about an AI funding gap in China?
Because the answer to DeepSeek’s cash problem may involve a blockchain solution. I’ve been tracking the rise of DePIN (Decentralized Physical Infrastructure Networks) projects that aim to tokenize compute resources — GPU time, storage, bandwidth. Projects like io.net, Akash, and Render are building marketplaces for idle compute. If DeepSeek or a similar independent lab were to issue a token that represents access to future compute time or a share of model revenue, it could unlock a new source of capital.
But more immediately, the Tencent/DeepSeek divide signals a broader trend: AI is becoming a capital-intensive commodity, not a technology differentiator. The winners will be those with the deepest pockets and the most diversified revenue streams. For crypto, this means the next leg of the AI bull run will be about infrastructure tokens that enable cost-effective compute access, not about AI model tokens that promise breakthrough intelligence.
Capital flows where intelligence meets speed. But also where capital meets patience. Tencent has patience. DeepSeek does not. The question is: can crypto’s DePIN thesis provide the missing patience?