Hook
Bill Ackman just did something that should make every crypto founder sit up. On March 10, Pershing Square’s 13F filing revealed a massive rebalance: Amazon became the fund’s fourth-largest holding, while Alphabet was effectively dumped. The move was framed as a bet on AI monetization, but the real story is about infrastructure neutrality versus model lock-in. In crypto, we’ve been fighting the same battle for years—and most of us are betting on the wrong horse.
I’ve been tracking this space since 2017, when I started running BlockNaija in Lagos. Back then, everyone was pitching “the next Ethereum killer.” Now, with AI agents and decentralized compute flooding the narrative, the same pattern is repeating. The Amazon–Alphabet trade is a signal. We need to decode it before the next bull cycle wipes out the naive projects.
Context
For those who don’t follow traditional markets, here’s the quick context: Pershing Square is a hedge fund with a long track record of concentrated bets. By boosting Amazon and slashing Alphabet, Ackman is implicitly saying that Amazon’s AI revenue path is clearer and more defensible. The logic? AWS sells compute and model-hosting services on a pay-per-token basis. Every AI startup, whether it uses Anthropic, OpenAI, or a fine-tuned Llama, pays AWS. Alphabet, meanwhile, owns Gemini and Google Cloud, but its core search business is threatened by the very AI it champions. Chatbots reduce ad clicks. The structural conflict is real.
In crypto, we have an analogous split. On one side, you have infrastructure protocols that are “model-agnostic”—Render Network, Akash, or Chainlink’s oracle network. They don’t care which AI model wins; they just provide the rails. On the other side, you have projects that build their own AI models and then try to monetize via a token—like Bittensor’s subnet structure or various “AI agent” platforms. The question is: which camp will attract the equivalent of Pershing Square capital?
Core: Why Infrastructure Neutrality Wins in Crypto’s AI Era
Let’s dive into the numbers. AWS’s AI revenue is now estimated at over $100 billion annualized, growing at 40%+. That’s not because AWS has the best model—it’s because they have the best platform. They offer Bedrock, which lets customers choose from multiple models. They invest in Anthropic, but they also host OpenAI’s GPT and Meta’s Llama. The customer is not locked in. This neutrality creates trust.
In crypto, we saw the same dynamic with oracles. Chainlink’s dominance is not because its data feeds are technically superior to every competitor—it’s because it’s decentralized and neutral. No single entity controls the price feed. Projects that tried to build proprietary oracles (like Maker’s old Oracle Security Module) eventually migrated to Chainlink because the market demanded neutrality.
Now apply this to AI compute. The decentralized compute market is still nascent, but the early leaders are the ones that are model-agnostic. Render Network handles rendering for any GPU workload, not just a specific AI. Akash provides a marketplace for any compute. Bittensor, on the other hand, is a network of specialized subnets that each run a specific model. That’s closer to Alphabet’s model: you own the model, you own the network. But the risk is that if a better model emerges outside Bittensor, the subnet’s value collapses.
Based on my experience auditing DeFi protocols in Lagos, I’ve seen this pattern before. In 2020, during DeFi Summer, the projects that survived the bear market were the ones that focused on composability—Uniswap, Aave, Maker. They didn’t try to own the entire user experience. They provided lego blocks. The same will happen in crypto AI. The projects that provide neutral infrastructure for any model to run on will accumulate network effects, while the “model-first” projects will face existential risk from regulatory headwinds and technological disruption.
Let’s talk regulatory risk. Alphabet is facing a U.S. Department of Justice antitrust case that could force it to break up its search business. The remedy is still being debated, but the risk is real. In crypto, the equivalent is the SEC’s crackdown on tokens that are too centralized. A project like Bittensor, where the core team controls a significant portion of the subnet structure, could be deemed a security. A neutral infrastructure protocol like Akash, where the network just matches buyers and sellers of compute, has a much stronger argument for being a commodity.
Contrarian: The Case for Model Lock-In (And Why It’s Still a Trap)
Now, the counter-intuitive angle. Some argue that owning the model is actually the higher-upside bet. If Gemini becomes the dominant AI, Alphabet’s bet pays off massively. Similarly, if Bittensor’s subnets produce the best open-source models, the TAO token could moon. The contrarian view is that infrastructure neutrality leads to commoditization—you become the “picks and shovels” in a gold rush, while the miners (the model owners) get the real wealth.
But here’s the problem: in crypto, “owning the model” is almost impossible to sustain. The pace of AI innovation is too fast. A model that is state-of-the-art today could be obsolete in six months. The team behind the token would need to constantly update the model, which creates centralization and governance hell. I’ve seen this play out with algorithmic stablecoins like Terra. The “model” (the algorithm) seemed perfect until it wasn’t. The infrastructure (the blockchain) survived, but the model collapsed.
Moreover, the regulatory environment for crypto AI is even more uncertain than for traditional tech. The EU AI Act imposes strict transparency requirements on models. If a crypto AI project’s model is closed-source, it risks regulatory pushback. If it’s open-source, the token’s value driver becomes fuzzy. The neutral infrastructure projects, on the other hand, can just say “we don’t make the models, we just provide the compute.” That’s a much easier sell to regulators.
Takeaway
Trust the process, but verify the code. Ackman’s move is a canary in the coal mine for crypto’s AI narrative. The winning projects will be those that are model-agnostic, regulatory resilient, and focused on building the rails rather than the trains. As we approach the next bull cycle, I’ll be looking at protocols that can host any AI without being tied to a single model. The ones that try to be the Google of crypto AI will likely end up as the Alphabet—brilliant technology, but structurally vulnerable to disruption.