Scanning the mempool for ghosts in the machine — and finding a familiar ghost: the narrative that cloud services are the only path to monetization. American Bank Securities just dropped a report claiming that AI in China will be 'dominated' by cloud service models — compute-as-a-service, model-as-a-service, the whole stack. I read it, and my fingers twitched. Because that same logic — the “just put it in the cloud and charge for compute” — has already infected blockchain infrastructure, and it's hiding the same structural risks.
Let's be clear: I'm a trader who audits code at 2 AM. I've seen how Infura, Alchemy, QuickNode — the “clouds” of Web3 — have become the default go-to for dApps, wallets, and even L2 sequencers. The parallels are eerie: just like AI model providers are told to rent GPU time from Alibaba Cloud, blockchain devs are told to rent RPC endpoints from centralized node providers. It's convenient. It scales. But it's also a ticking time bomb.
Context: The Mempool is Not the Cloud
When a blockchain app uses Infura, it's trusting a single company to relay its transactions to the network. In 2020, Infura went down, and Ethereum DApps froze. In 2022, Alchemy had latency spikes that cost me an arbitrage window — I missed $12,000 because my bot's RPC call timed out. The cloud model gives you uptime SLAs, sure, but it also gives you a centralized choke point. The same American Bank report that lauds the AI cloud model conveniently ignores that the entire value proposition of blockchain is decentralized trustlessness.
Meanwhile, the real alpha is in the edges: running your own light node, using decentralized RPC networks like Pocket or Lava, or even building custom relayers for specific chains. I've been running my own Solana validator since 2021 — it's not easy, but when the cloud providers throttle or censor, I'm still trading. The cloud service model is an abstraction that hides the underlying protocol's health. It's a beautiful wrapper around a brittle core.
Core: Seven Dimensions of a Broken Analogy
Let me dismantle the “blockchain cloud service is the main monetization channel” thesis using the same seven dimensions the analysts love — but with real code and market structure.
1. Technical Route (Low Relevance) Cloud RPCs don't contribute to consensus or security. They're just convenience layers. The AI cloud argument depends on ever-growing compute demand; in blockchain, the demand is for decentralized compute, not centralized relay. If L2s adopt centralized sequencers, the whole security model is a sham. Based on my audit experience, I've seen code that routes through Infura by default — trivial to exploit if Infura goes rogue or gets subpoenaed.
2. Commercialization (High Relevance, But Misleading) The report claims “cloud services capture the most value.” In blockchain, the most value is captured by protocol fees and MEV, not RPC providers. Alchemy charges $49/month for a free tier essentially — they make money from volume, but they don't own the liquidity. Compare that to Uniswap's $1B+ in fees in 2024 — the real value is in the settlement layer, not the delivery layer.
3. Industry Impact (High) If the “cloud service” narrative dominates, we get an industry where dApps are portable across chains only if they rely on the same RPC provider. This kills chain sovereignty. I saw this in 2023 when Arbitrum's Nova sequencer outage forced many apps to pause — exactly like AI firms hit by Alibaba Cloud downtime. The industry becomes fragile, consolidated around a few node operators.
4. Competitive Landscape (High) The winners are Alchemy, Infura, QuickNode — not the chains. They are the “cloud AI” of crypto. But unlike AI, where model quality is a differentiator, RPC quality is a commodity. Latency and price are the only battles. New entrants like Lava Network offer peer-to-peer RPC, cutting out middlemen. I've seeded a Lava node and tested it — it's 15% slower than Alchemy but censorship-resistant. That's the edge the analysts miss.
5. Governance & Censorship (Medium-High, Ignored) The American Bank report never mentions data sovereignty or censorship resistance. In blockchain, using a US-based RPC provider means your transactions can be censored by OFAC sanctions. Tornado Cash taint checks happen at the RPC level. I switched my bot to a self-hosted Erigon node after I got my Alchemy API key blocked during a liquidation hunt. The cloud model exposes users to regulatory risk that pure on-chain transactions don't.
6. Investment & Valuation (Medium) Analysts love to value Alchemy at billions based on RPC volume. But that volume is a function of chain usage, not Alchemy's moat. A L2 using its own sequencer doesn't need RPC. As L2s mature, many will internalize the RPC layer, just as many AI firms are now building their own inference stacks on top of open-source models. The valuation skew is dangerous.
7. Infrastructure & Compute (High) The AI cloud model depends on GPU access; blockchain cloud depends on validator node access. With the shift to L2s, the compute moves to sequencers, which are often centralized. This mirrors the GPU dependency problem. Every bug is a bounty waiting for the right eyes — and the biggest bug is believing that centralized RPC is the only way to scale.
Contrarian: The Real Cash Cow is Self-Custody of Infrastructure
Retail and VCs are piling into “node-as-a-service” projects, thinking they'll be the Alibabas of Web3. But smart money is already moving away. The 2025 trend is decentralized RPC marketplaces and self-hosted light clients. I've been testing a prototype using waku p2p for order book data — it eliminated my dependence on any single provider.
The contrarian truth: The more “cloud-like” blockchain infrastructure becomes, the less it deserves the “Web3” label. The real monetization isn't in serving RPC calls — it's in securing the network (staking), providing liquidity (DeFi), or capturing MEV. The cloud layer is a tax, not a business.
Volatility is the only friend we have — and cloud services smooth it out, taking away the alpha from those who can handle the spikes. I'd rather take the volatility and run my own node than pay a middleman to pretend it doesn't exist.
Takeaway: Build the Node, Not the RPC
The next bull run won't be won by the companies that sold picks and shovels — it'll be won by those who dug the mines themselves. Alchemy and Infura will still make money, but the real alpha is for protocols that let users self-custody their infrastructure. The American Bank report on AI cloud should be a cautionary tale for blockchain: don't let convenience centralize your trust.
Surviving the crash taught me to trade the panic — and now I'm seeing the coming panic when a major L2's cloud provider goes down during a flash crash. The question isn't if it happens, but whether you have your own node to fall back on.