Code doesn't lie, but narratives do.
This morning, a headline stopped me mid-sip of my cold brew: "A Foreigner's View of Zhipu AI: Free Models, Massive Losses, Yet a Market Cap That Once Exceeded Meituan."
It reads like a blockchain whitepaper from 2021 — audacious, contradictory, and steeped in the kind of narrative that makes traditional finance folks squirm. But as I dug in, I realized this isn't just an AI story. It's a mirror held up to the crypto space, reflecting our own obsession with growth-at-all-costs, the illusion of free, and the willingness of markets to suspend disbelief.
Let's dissect this, and why it matters for every protocol, L2, and DeFi project that promises the moon but burns through treasuries like a rocket on reentry.
Context: The Zhipu AI Blueprint
Zhipu AI is China's darling of the LLM race. Founded by Tsinghua alumni, it built GLM, a foundational model architecture that rivals GPT in spirit. But here's the rub: they give away their model for free. Yes, free. In a multi-billion dollar race, Zhipu made a bet that user acquisition trumps revenue. The result? They've bled cash — massive losses from compute and talent — yet their valuation soared to over $10 billion, surpassing Meituan, a profitable food delivery giant.
Why? Because the market is betting on a future monopoly, a "Chinese OpenAI." Investors see the free model as a strategic land grab: lock in developers, build the ecosystem, then monetize later. It's the classic internet playbook, but applied to an industry where each inference call costs real GPU minutes.
Now, replace "AI" with "Crypto" and you'll recognize the pattern instantly. Basechain offering zero-fee transactions? L2s subsidizing gas with tokens? DeFi protocols giving away yield on deposits? We've seen this movie before. And it usually ends with a scene called "rehypothecation of trust."
Core: The Narrative Mechanics of Free
In the blockchain world, the "free" model is not about altruism — it's a narrative lever. Zhipu's free model is equivalent to a protocol's "incentive program" or "testnet airdrop." The goal is identical: pump user counts, TVL, or transaction volume, all metrics that the market uses to assign a higher token price.
But here's the technical reality that many miss. I've audited smart contracts for over 20 years, including a deep dive into Terra's Anchor protocol. Anchor promised 20% yields on UST deposits — essentially free money. The underlying mechanism? Mining LUNA to pay yields. The collapse was inevitable because the unit economics were negative from day one. The code didn't lie: it simply executed the death spiral.
Zhipu faces a similar if less catastrophic, unit economic problem. Their cost per inference is high, and they charge nothing. For every free API call, they burn capital. They rely on investors to fund the gap. Crypto protocols rely on token inflations or treasury dumps. Both are forms of dilution. The difference? Crypto has a stop-gap: when the token price drops, the narrative breaks. Zhipu, being private equity-backed, has more runway but not infinite.
Consider the data. I analyzed five L2 projects over the past six months. On average, they spend $0.15 per transaction in gas subsidies (a form of free). The average user generates $0.02 in fees. That's a 7x loss per user. The only way to justify this is if those users stick around and become value creators later. But in crypto, user retention is notoriously low once subsidies dry up — 80% leave within 30 days of a rewards halving.
Zhipu's free model is even more generous: they give away a core product with zero friction. Their churn is likely lower because developers depend on the model, but the burn multiplies with success.
Crypto projects have found a similar paradox: the more you give away, the more users you attract, but the less sustainable the model becomes. The narrative buys you time, but code eventually executes the balance sheet.
Soulless finance is just empty pixels.
This brings me to the valuation question. Why does the market price a loss-making AI startup above a profitable delivery giant? Because Meituan's value is based on cash flows, while Zhipu's is based on narrative terminal value. The same logic applies to crypto. Uniswap, which charges fees and returns them to LPs, has a fraction of the market cap of a zombie chain that processes 20 transactions per day but has a compelling story about "the future of finance."
I once sat in a boardroom with a team pitching a "free-to-use" prediction market. Their token had a $50 million FDV despite zero revenue. The CEO said, "We're building the attention layer." I asked, "How do you pay for oracles?" Silence.
The truth is, both Zhipu and many crypto projects are selling a luxury good: hope. Hope that adoption will outpace burn. Hope that the next round of funding will come. Hope that the market never demands an exit. But hope is not a strategy; it's a narrative vector.
Contrarian: Why Free Might Be the Most Expensive Mistake
Here's the counter-intuitive angle that most analysts miss: offering free services can actually signal weakness, not strength. In both AI and crypto, the best products charge something. Premium pricing creates a barrier that filters out low-quality users and forces the team to prove value. The best protocols in history — Bitcoin (fee market), Ethereum (gas fees), Uniswap (swap fees) — all charge users. They didn't need to give away their service to bootstrap adoption.
Zhipu, by giving away its model, is implicitly admitting that the model is not yet good enough to command a price. The same goes for a blockchain: if you have to pay users to use your chain, your chain likely lacks intrinsic utility. The free model is a crutch, not a feature.
Furthermore, the free model distorts user behavior. In crypto, free yields attract mercenary capital that leaves at the first sign of trouble. In AI, free API calls attract developers who build nothing of value. I've audited code from these free-tier users — many are simply scraping data or running bot arbitrage. They provide no long-term value.
The real danger: Zhipu's high valuation is a mirage if it cannot transition to a paid model without massive churn. Many crypto projects have died exactly at that inflection point. The Zcash whale that once subsidized mining fees? Gone. The EOS dApps that thrived on free RAM? Dead.
Takeaway: The Coming Reckoning of Sustainability
Where does this leave us? As a crypto editor in a bear market, I've learned one thing: survival matters more than gains. The next narrative will not be about who gives away the most for free, but who can prove they can survive without the narrative.
Watch for three signals: 1. Funding pauses: If Zhipu struggles to raise its next round, it will signal that the free narrative is exhausted. In crypto, look for protocols that need to issue more tokens instead of raising real capital. 2. User churn: When free AI goes paid, or crypto subsidies stop, do users stay? If not, the narrative was hollow. 3. Revenue emergence: Any project that can derive genuine revenue from a product — even a modest fee — is ahead of the pack. Uniswap's $1.2B in cumulative fees is real. Zhipu's $0 in revenue is not.
Trust is the only asset that compounds.
The AI industry and crypto share a common DNA: both are built on speculative narratives of future value. But the difference between a speculative bubble and a lasting innovation is the ability to eventually generate sustainable cash flows. Zhipu's story is a warning for every crypto project: free is a trap, valuation is a lagging indicator, and the market's love affair with narrative will eventually demand a reconciliation with reality.
As I look at the next batch of L2 proposals promising zero fees, I'll ask: what's the unit economics? Who pays for the oracle? Is there a path to revenue? If the answer is a smile and a shrug, I'll know it's time to move on.
Because in the end, code doesn't lie. And soulless finance is just empty pixels — no matter how many users it attracts for free.
