The ledger never sleeps, only updates. This morning, Zhipu Protocol's token is trading at a $12B fully diluted valuation. Last quarter, it burned through $210M in operational costs. Free AI inference. No revenue. Yet the market prices it like a monopoly-in-waiting.
I've seen this pattern before. In May 2022, I spent three weeks tracing the Anchor Protocol's yield mechanics. The same disconnect: high hopes, zero cash flow, algorithmic promises. That didn't end well.
Context: The Zhipu Model
Zhipu Protocol is a decentralized AI inference network. Users submit prompts; a distributed set of GPUs run the GLM-4 model for free. No gas fees. No token burn for queries. The protocol pays compute providers in its native token, ZPT, which it mints from a foundation reserve.
This is not Bittensor. Not Render. Zhipu's model isn't for decentralized training or rendering—it's for inference at scale. Think of it as a public goods AI layer. The team calls it "democratizing intelligence." The critics call it a money furnace.
Why free? The thesis: acquire users aggressively, build developer ecosystem, then flip a switch later. Classic Silicon Valley playbook. But in crypto, the unit economics are worse. Cloud compute costs real fiat; tokens don't pay AWS bills. The foundation must sell ZPT to fund operations, creating constant sell pressure.
Core: Tracing the Flows
I pulled the on-chain data from the foundation's multi-sig wallet (0xZP...FOUND). Over the past 90 days:
- $210M worth of ZPT minted and sent to compute provider addresses.
- Only $8M worth of ZPT burned via governance buybacks (a mechanism to reduce supply).
- Net token inflation: ~$202M per quarter.
If it isn't on-chain, it didn't happen. The supply is growing at 15% per quarter. Yet the price has held—even increased—fueled by narrative and spot ETF whispers.
User growth: 2.3M unique wallets queried the model in April. Average query cost: ~0.02 ZPT in compute subsidies. That's down from 0.05 in January due to efficiency gains, but still deeply negative. The protocol's "revenue" is zero. Its "cost of goods sold" is the entire operational budget.
Compare to traditional SaaS: a user costs $0.10 per query, and they pay nothing. Zhipu is losing money on every interaction. This is a textbook operationally unsustainable model—unless the endgame is a monopoly.
Based on my experience auditing Uniswap V2's factory contract in 2020, I saw a similar narrative-vs-reality gap. The market believed Uniswap would capture all DEX liquidity. It did. But Uniswap also had a fee switch optionality from day one. Zhipu has no such mechanism yet.
Contrarian: The Free-to-Paid Trap
The contrarian angle is not that Zhipu fails. It's that the free strategy is actually the most expensive move possible. Chaos is just data waiting to be indexed. Right now, the market is indexing Zhipu's user growth as a positive signal. It's not. Without a path to monetization, user growth is just cost growth.
Here's the unreported angle: Zhipu's free model is bleeding liquidity out of the broader AI token ecosystem. Capital that could flow to Bittensor or Akash is piling into ZPT because of the narrative. But Bittensor has staking yields and subnet fees. Akash has compute marketplace revenue. Zhipu has nothing but a promise.
Speed is the only moat in a borderless war. Zhipu moved fast to be the default free inference layer. But speed without a monetization plan is just fast failure. The Terra/Luna cascade taught me that algorithmic models need real anchor yields. Zhipu's anchor is its burn rate. When the foundation runs low on seed capital—estimated at 18 months of runway—they must either raise at a lower valuation or flip the fee switch.
The Flippening Event
A proposal went live yesterday: ZIP-12. It suggests a $0.001 per query fee, collected in ZPT and burned. If passed, this is the litmus test. Will users stay?
My prediction: 60% of retail users leave. But institutional users—those building on the model—might stay if the model is truly superior. The question is whether the remaining 40% generates enough fees to cover costs. At $0.001 per query, 2.3M queries per day yields $2,300 daily revenue. Against $2.3M daily operational cost. That's a 0.1% recovery rate.
Adapt or get front-run by your own assumptions. If Zhipu is truly the best model, it could charge more—$0.01 per query—and still retain high-value users. But that kills the free narrative. The token might repriced on "subscription model" hype, then crash when the math doesn't work.
Takeaway
The next signal is the vote on ZIP-12. If it passes, watch the user retention data on-chain. If the active wallets drop by less than 30%, ZPT could hold. If they drop by more than 50%, the protocol's valuation narrative collapses. The truth is hidden in the block height—and in the quarterly foundation wallet movements. I'll be watching.
Zhipu is a bet on future monetization. But future is not here. And in crypto, future cash flow is priced at a discount to current. Until that flips, this is a speculative narrative play, not a value investment. The ledger never sleeps. I'll keep updating the analysis.