The 5% surge in Tencent’s stock on 12 June 2025 was not a reaction to better-than-expected earnings or a new game licence. It was a narrative flip — pure and simple. Goldman Sachs warned that AI inference costs could wipe out 5-17% of operating profit. JPMorgan countered with a USD 126 billion incremental revenue forecast by 2030. The market chose the latter. But as a Quantitative Strategist who has spent years dissecting on-chain flows and institutional liquidity, I see something else beneath the surface: a structural battle between centralised efficiency and the permissionless ideals that underpin blockchain’s value proposition.
Hook
Tencent’s WorkBuddy now boasts a DAU/MAU ratio of 65-75% — a stickiness metric that rivals Slack at its peak. Its WeChat AI assistant "Xiaowei" is being tested on 1.43 billion monthly active users. The Hunyuan 3 model has been integrated into 131 products, and token consumption has grown 10x in six months. Yet the entire AI stack runs on a single, opaque, centrally governed infrastructure. The data that feeds these models flows through WeChat’s closed ecosystem — every message, every mini-programme interaction, every payment request. For a blockchain analyst, this is both fascinating and alarming. It represents the ultimate counterpoint to the decentralised, trust-minimised architecture that crypto advocates champion.
Context
To understand the significance, we need to strip away the marketing narratives and look at the technical mechanics. Tencent’s AI strategy is not about model supremacy — Hunyuan 3 likely trails GPT-4o and Claude 3.5 on standard benchmarks like MMLU or HumanEval. Instead, it is about integration density. WorkBuddy automates workflows across Tencent Docs, Tencent Meeting, WeCom, and 30+ external tools via a SkillHub library of 790,000 skills. "Xiaowei" embeds an agent directly into the world’s largest super-app, allowing users to perform tasks — pulling sales data, generating PPTs, booking mini-programmes — through natural language commands.
This is the exact opposite of how DeFi protocols operate. In DeFi, every action is a transaction on a public ledger; you can audit the smart contract, trace the liquidity flow, and verify the outcome. Tencent’s AI agent, by contrast, executes code on proprietary servers, with no public record of what actually happened. The user trusts the platform blind. "Code is law until the block confirms the error" — but here, there is no block to confirm. The error simply disappears into the black box of corporate AI.
Core: The On-Chain Evidence Chain That Doesn’t Exist
Let me apply the same methodological framework I used during my 2017 ICO due diligence audits, when I traced 14,000 ETH across 300 wallets to verify a token’s compliance claims. For Tencent’s AI, we have to construct an evidence chain from the few public data points available.
- Token consumption growth of 10x: This is the only "on-chain" analogue. But unlike Ethereum’s gas consumption, which is transparent and auditable, Tencent’s token usage is self-reported. We have no way to verify whether the growth came from genuinely useful inference or from low-value system prompts and template responses.
- DAU/MAU ratio of 65-75%: High, but not necessarily indicative of AI value. WorkBuddy is essentially a workplace messaging tool with AI features bolted on. The daily logins could be driven by compliance obligations (e.g., checking company announcements) rather than active AI agent usage. In crypto terms, this is like confusing a high transaction count with organic demand — a classic fallacy I dismantled in my 2020 DeFi yield backtests, where 80% of high-volume pools were actually wash-trading or subsidy mining.
- Cost base – Goldman Sachs’ worry: The 5-17% profit erosion estimate assumes that Tencent will offer free AI inference at scale. But that is a choice, not a law. Tencent can throttle usage, introduce tiered pricing, or optimise inference with custom ASICs (its "Purple Xia" chip). The real risk is not cost — it is liquidity illusion. Just as a DeFi protocol with high TVL but no organic borrowers is a ticking bomb, a super-app with high AI activity but no clear monetisation path is a valuation mirage. "Efficiency without liquidity is just an illusion."
- The JPMorgan USD 126 billion forecast: This assumes a per-user ARPU of roughly USD 10 per month by 2030 — plausible only if "Xiaowei" unlocks payments, advertising, and commerce at a massive scale. But China’s regulatory environment remains hostile to unsupervised AI agents handling financial transactions. The PBOC and CAC have not issued guidelines for agent-initiated transfers. Without regulatory clarity, that revenue line is as speculative as an ICO whitepaper promising "disruption."
Contrarian: Centralised AI Might Be More Aligned with Crypto’s Real Value
Here is the counterintuitive angle — and it is one that most blockchain purists will reject. Tencent’s walled-garden approach may actually accelerate crypto adoption in the long run. Why? Because every failure of centralised AI — every data leak, every misaligned agent, every regulatory crackdown — creates a demand signal for decentralised alternatives. "Data demands respect, not reverence." The more Tencent entrenches its AI into daily life, the more visible the trade-offs become.
Consider the mini-programme generation feature in "Xiaowei". Users can create lightweight apps via natural language prompts. This lowers the barrier for non-programmers, but it also centralises control over what apps exist, what data they collect, and how they are monetised. The developer never owns the code — Tencent does. In contrast, a blockchain-based agent platform would allow developers to own their smart contracts and retain sovereignty over user interactions. The trade-off is complexity and latency; but for sensitive use cases — medical records, legal contracts, financial advice — that trade-off is acceptable.
The second contrarian point: Goldman Sachs’ cost concern might be overblown by design. As an analyst, I know that worst-case scenario models often assume zero optimisation. Tencent has decades of experience running large-scale web services. Its inference engine likely uses FP8 quantization, speculative decoding, and continuous batching. The real cost per token is probably 50-70% lower than what Goldman’s model assumes. Moreover, "Gravity always wins when leverage exceeds logic" — applied here, the leverage is the market’s willingness to pay for an AI narrative. If Tencent can demonstrate even modest revenue from WorkBuddy subscriptions (e.g., USD 5 per user per month for 50 million MAU), that is USD 3 billion annualised — enough to justify the current valuation uplift.
Takeaway: The Next Signal Is Not a Price Jump – It’s a Regulatory Filing
The market has priced in a successful narrative flip. But blockchain investors know that narratives without structural data do not sustain. The next critical signal for Tencent’s AI will not come from a user growth announcement or an investment bank upgrade. It will come from a regulatory filing — specifically, the People's Bank of China granting WeChat AI permission to execute payment transactions on behalf of users. If that happens, the USD 126 billion forecast becomes a conservative floor. If it is delayed or denied, the stock will re-rate downward, and WorkBuddy’s high DAU/MAU will be exposed as a vanity metric.
For the blockchain community, the lesson is clear: centralised AI is not the enemy — it is the stress test. Every friction Tencent encounters validates the need for permissionless, auditable, user-owned AI infrastructure. The question is whether the crypto ecosystem can deliver that infrastructure before the walled garden becomes too comfortable for users to leave.
"Volatility is the tax you pay for uncertainty." Tencent’s AI journey will be volatile. But the underlying asset — the ability to embed intelligence into the world’s largest social and financial super-app — is real. Watch the regulatory flows, not the hype. Data demands respect.