Speed kills, but slow kills too in this game.
Daiwa just dropped a bombshell on the Asian markets this morning. The bank slashed Tencent’s target price to 670 HKD while simultaneously jacking up AI capital expenditure forecasts by a staggering 67%—from 108 billion to 181 billion RMB by 2026. Profit forecasts were chopped by 1% to 6% across the near term. The market yawned, selling off 2% before lunch. But anyone who’s spent a decade watching liquidity cycles knows: this is not a yawn moment. This is the sound of a giant shifting its weight onto a new tectonic plate.
Behind the dry numbers lies a story every crypto native should feel in their gut. Tencent is doing what blockchain protocols have been doing for years—burning short-term earnings to build infrastructure that may or may not ever see real demand. The difference? Tencent actually generates cash. But so did Bitmain during the 2018 mining bust. We’ve seen the moon. Now we’re looking for the exit.
Context: The Old Money Moves to the New Frontier
Let’s back up for a second. Tencent is the 800-pound gorilla of Chinese internet. WeChat, QQ, gaming (League of Legends, PUBG Mobile, Honor of Kings), Tencent Cloud—it’s a cash machine. In 2023, it raked in over 600 billion RMB in revenue and 150 billion in operating profit. The core business is mature. User growth has flatlined. The next act is AI.
Tencent’s AI push is not experimental. It’s existential. If they don’t build the infrastructure to power the next generation of services—intelligent NPCs in games, AI-driven advertising algorithms, enterprise cloud AI—they risk being disrupted by AI-native startups or by rivals like ByteDance and Alibaba who are also spending billions. The board has decided: short-term pain, long-term gain.
But here’s where the story gets interesting for crypto observers. This is the same playbook we saw in the 2021 Layer2 infrastructure boom. Projects raised hundreds of millions to build rollups, data availability layers, and validator networks. The pitch was always: "Build it, and the users will come." In most cases, they didn’t. The infrastructure was overbuilt relative to actual demand. Token prices collapsed. Investors got rugged by their own optimism.
Tencent is different—it has 1.3 billion monthly active WeChat users and a sticky ecosystem of advertisers and developers. If AI demand emerges, Tencent can flip the switch and monetize fast. But that "if" is a big one. And the scale of capex is so large that even a 10% overestimation of demand would result in 18 billion RMB in wasted depreciation. That’s a coin toss that could sink a year’s worth of gaming profits.
Core: Breaking Down the Numbers and the Hidden Assumptions
The Capex Surge
Daiwa’s revised capex forecast for 2026 is 181 billion RMB. To put that in perspective, that’s roughly $25 billion USD. That’s more than what Amazon Web Services (AWS) spent on infrastructure in 2014. Tencent is signaling that it intends to build one of the largest AI compute clusters in the world, likely powered by a mix of NVIDIA H100/B200 chips and domestic alternatives like Huawei’s Ascend series.
The Profit Squeeze
The report cuts 2024-2026 EPS by 1-6%. That might sound small, but it’s a massive absolute number—potentially 3-4 billion RMB in lost earnings per year. The mechanism is simple: higher depreciation from all those GPUs and data centers. Every server has a 3-5 year life. The more you buy, the more you write off each quarter.
The Unspoken Assumption: Chip Supply
The report notes "improving chip supply" as a reason for the higher capex. This is code for: Tencent has found a stable workaround for US export controls. Either it’s stockpiling NVIDIA chips through grey-market channels, or it’s placing massive orders for Huawei’s Ascend 910B. Given the geopolitical climate, the smart money is on domestic alternatives. But domestic chips have lower performance and higher power consumption. That means Tencent’s cost per teraflop may be higher than global peers like Microsoft or Google. Competitive disadvantage, baked in.
The Monetization Window
Daiwa expects AI monetization to begin in H2 2026. That’s 18 months from now. In crypto time, that’s an eternity. In enterprise time, it’s fast. But here’s the rub: Tencent is building for inference workloads, not just training. That means they anticipate a wave of real-time AI applications—chatbots, content generation, recommendations—that will consume compute continuously. That’s a bet on the marginal cost of inference dropping enough to spur mass adoption. If instead the market remains dominated by training (which is more concentrated and lower volume), the inference capacity will sit idle.
First-hand observation from the trenches:
I’ve worked on the exchange side during the ICO frenzy of 2017. I’ve seen projects raise millions for "infrastructure" that turned out to be a single server and a Medium article. Tencent is not a scam. But the psychological pattern is identical: euphoria about a new technology, followed by a collective decision to overbuild, followed by a painful correction when reality doesn’t match the timeline. The only question is the magnitude of the correction. "We bought the dip, but the floor kept dropping."
Contrarian: Why this Infrastructure Might Be a Trap
Most crypto analysts will tell you that Tencent’s AI capex is a no-brainer. "They have the cash, they have the users, they’ll capture the value." I disagree. The contrarian view here is that Tencent is overpaying for a race that doesn’t yet have a finish line.
First, the data availability (DA) layer hype in crypto taught us a lesson: 99% of rollups don’t generate enough data to need dedicated DA. Similarly, 99% of enterprise AI workloads may not need the kind of massive inference infrastructure Tencent is building. Most companies will use cloud APIs or fine-tune small models. The real demand for heavy compute may be concentrated in a few hyperscale players. Tencent might be building a highway to a ghost town.
Second, the organizational drag. Integrating AI into every product line—from QQ to WeChat Pay to advertising—requires rewriting software stacks that are decades old. That’s not a capital problem; it’s a talent and execution problem. I’ve seen crypto projects with billions of dollars fail because they couldn’t coordinate their engineering teams. Tencent has better managers, but the complexity is orders of magnitude higher. The legacy code is a debt that no amount of GPU can pay off.
Third, the regulatory risk. China’s AI governance is tightening. New rules on algorithm registration, content safety, and data localization are being drafted. Any delay in compliance or forced changes could push the monetization timeline past 2027. That would leave Tencent holding the bag on billions in depreciating hardware.
"Hype is the fuel, but fundamentals are the engine."
If the fuel (AI hype) evaporates, the engine (core gaming and ad revenue) better be strong enough to keep the plane flying. But Daiwa is already cutting profit forecasts. The engine is losing power just as the pilot floors the throttle.
Takeaway: Watch the Cash Flow, Not the Headlines
The next 18 months will be decisive. I’ll be tracking three signals:
- Capex-to-revenue ratio: If Tencent’s capex exceeds 30% of total revenue, alarm bells will ring. Currently it’s around 20%. The Daiwa forecast pushes it to 28%. That’s dangerous territory.
- Cloud AI revenue disclosure: Tencent Cloud needs to start disclosing AI-related revenue. If by Q1 2026 there’s no meaningful line item, the thesis is broken.
- Core business health: Gaming and advertising must maintain at least 5% growth to fund the AI bet. Any slowdown there will trigger a downward spiral.
For the crypto community, Tencent’s gamble is a mirror. Many protocols are running the same playbook—spending treasury on infrastructure (validators, bridges, Layer2 sequencers) without proven demand. The same laws of capital allocation apply. "The crowd moves fast, but the ledger moves faster."
Tencent’s ledger is clear: billions in, uncertain returns. Smart money will wait for the first real revenue numbers before calling this a win. Until then, I’m keeping my short-term powder dry and watching the depreciation line.