I just published the second part of my essay about the current state of web3-AI :
jrodthoughts.medium.com/cath…. This one reflects on core physics of AI and the AI market:
Most web3-AI projects sooner or later need to face a hard truth: AI is a centralization force and the entire market is web3.
Some key points:
1. The Physics of Friction: Intelligence craves density. Training frontier models requires thousands of GPUs screaming at each other over 900 GB/s interconnects (NVLink). When you decentralize training, you replace that ultra-low latency with the public internet. You introduce friction. In a race where OpenAI is spending billions to shave milliseconds, a "decentralized training network" that adds latency for the sake of ideology isn't a feature—it's a bug.
2. There Are No "Web3" Customers: If you are building for DAOs or "crypto-natives," you are fighting over crumbs. The real money—the acquirers and high-value contracts—is in Web2. It’s Fortune 500 companies optimizing supply chains and law firms automating diligence. They don't care about your node topology. They care about uptime, compliance, and cost. If your user experience requires a wallet or understanding "slashing," you have lost 99% of the market.
3. The Token is a Poison Pill: This is the most dangerous trap. We launch tokens to bootstrap liquidity, but for a deep-tech startup, a token can make you unacquirable. To a public company like Microsoft or Cisco, a volatile utility token is a regulatory nightmare and an integration disaster. By tokenizing too early, you kill your M&A optionality.
The Solution? The "Mullet" Strategy.Web2 in the front (SaaS contracts, fiat payments, suits). Web3 in the back (decentralized supply chains, verification).
Don't let the friction of the blockchain grind your performance to a halt. Build bridges to the mainland, not islands for pirates. Full essay is here:
jrodthoughts.medium.com/cath…