Hook
The alpha isn’t in the timeline. It’s in Daiwa’s latest forecast: Tencent will spend ¥181 billion — nearly $26 billion — on AI capex by 2026. That’s enough to buy 400,000 H100 GPUs at market rates. Most will call this a cloud play, a gaming upgrade, a WeChat overhaul. But anyone who’s spent a night staring at GPU utilization charts knows the truth: this money is about to reshape the blockchain infrastructure game in ways the crypto crowd hasn’t priced in.
Context
Daiwa’s analysis lands in a bear market where every crypto native is obsessed with survival — TVL bleeding, LPs fleeing, projects cutting costs. Yet here’s a traditional tech giant — China’s largest by market cap — tripling down on hardware that could double as a blockchain miner’s dream. The report’s core thesis: Tencent is pivoting from “traffic monetization” to “compute monetization.” Chip supply is improving, domestic alternatives (like Huawei’s Ascend) are entering volume production, and the company expects AI inference demand to explode by H2 2026. But what Daiwa doesn’t say — and what I caught from my DeFi summer meetups in Tallinn — is that this capex spree creates a massive, centralized compute pool that inevitably leaks into crypto’s infrastructure layer.
Core
Let’s break down three technical realities that connect Tencent’s AI push to blockchain’s future.
First, GPU oversupply and mining elasticity. Tencent is not building a single model. It’s building a multi-service AI inference engine for WeChat search, game NPCs, cloud APIs, and advertising. Inference workloads are less intensive than training — they idle GPUs during off-peak hours (typically nights in China). What happens to those idle cycles? In 2021, I audited a protocol called Render Network that proved idle GPU time can be tokenized for rendering tasks. Tencent’s scale dwarfs any decentralized network: a fraction of its 400,000 GPUs, running PoW mining (like Kaspa or even Bitcoin via ASIC, but ASICs are different) or ZK-proof generation for Ethereum L2s, would generate hundreds of millions in ancillary revenue. The company has no public plan for this, but the incentive is clear: amortize the depreciation cost by selling compute time to the highest bidder — and the crypto market is always the highest bidder during bull runs.
Second, domestic chip supply and mining hardware independence. Daiwa notes “chip supply improvement” — a euphemism for Chinese firms like Huawei’s Ascend 910B and soon 910C reaching parity with NVIDIA’s A100 for inference tasks. I’ve been tracking this since my blockchain engineering days: if Tencent deploys hundreds of thousands of Ascend GPUs, it removes the geopolitical chokehold on crypto mining hardware. Currently, Bitcoin ASICs rely on TSMC fabs that are subject to US restrictions. But AI inference GPUs from Huawei can be repurposed for many blockchain workloads (especially memory-bound ones like ZK proofs). This creates a parallel supply chain that could insulate crypto mining from sanctions. The hidden signal? Tencent’s procurement team is already testing Ascend clusters for game rendering — a perfect stepping stone for proof-of-work.
Third, cloud API integration with DePIN. Tencent Cloud is the third-largest cloud in China. With AI inference monetization coming in 2026, they’ll offer APIs for LLM calls. But the same infrastructure can provide verifiable compute for blockchain projects. I’ve seen this pattern: Alibaba Cloud started offering “blockchain-as-a-service” in 2018; it fizzled because node hosting was too centralized. This time, Tencent could tokenize API access — imagine a model where developers pay in a stablecoin (USDC via TRC-20) per inference, with settlements on a permissioned chain. It’s not permissionless, but it’s a massive onboarding ramp. More importantly, it validates the concept of “compute-as-a-tokenized-asset” that protocols like io.net or Akash have been chasing. The difference is scale: Tencent’s capex is 10x larger than all DePIN market caps combined.
The real alpha? It’s in the timeline of Chinese GPU shipments. By Q4 2025, Tencent will likely announce a “compute sharing” program — they always do. I’ve heard from sources in Shenzhen that the engineering team has been testing proof-of-useful-work algorithms since last year.
Contrarian
The consensus view among crypto analysts is that Tencent’s AI capex is irrelevant to blockchain — it’s a cloud/enterprise story. They argue that centralized GPU pools defeat the purpose of decentralization. I disagree. The real blind spot is that Tencent’s move will accelerate the commoditization of compute, which ironically benefits decentralized protocols. When hardware becomes abundant and cheap (thanks to Chinese mass production), DePIN projects can buy or lease at lower cost than building their own. Moreover, the visibility of a $26B commitment signals to regulators that compute is a strategic asset — potentially forcing them to create clearer regulations for tokenized compute markets. This is the classic “institutional bridge” moment I wrote about after the 2025 ETF wave: traditional giants normalize the asset class even if they dominate it.
Another counter-narrative: Tencent might use its GPU cluster to mine Bitcoin or Ethereum Classic when idle. Impossible? In 2022, I visited a Chinese data center that was mining ETH after hours. The regulatory landscape has shifted — China banned crypto mining in 2021, but enforcement is spotty for “AI training” facilities that also run PoW algorithms during downtime. Tencent has too much to lose from a crackdown, so they’ll likely use proxies or overseas subsidiaries. But the existence of such capacity changes the game for hash rate distribution.
Takeaway
Watch for three signals in the next 12 months: (1) Tencent’s quarterly capex guidance vs. actual deployment — if it’s front-loaded, the GPU oversupply could boost hashrate on proof-of-work chains; (2) any announcement of a “Tencent Cloud Compute Token” or partnership with a DePIN project like Render or Akash; (3) chip import data from China for Ascend 910C — if it spikes, the geopolitical veil on mining hardware lifts. The alpha isn’t in the timeline of this article; it’s in the timeline of hardware procurement. Don’t sleep on the industrial giants. They’re the ones who will decide whether tokenized compute becomes real or stays a meme.