Private AI

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…

The Onchain Inference Market Map On June 7 I wrote about base:0xc52aedec3374422d7510e294cfaa90799595cba3 a token with a real product and zero revenue. I thought it was one weird story. Turns out it's a whole market of weird stories. And they're tangled into each other way more than it looks from the outside. Spent the week going through the data. Four players on the map: three crypto projects and one giant with no token that could roll over all of them. I'm not a developer. I just look at what's visible onchain and in public dashboards, and I keep asking one question: if people actually start buying AI inference with crypto at scale who makes money on it? First, the market itself, for scale Inference is every single request to a model. You type a prompt, the model answers, somebody sends the bill. OpenRouter, the biggest model aggregator, routes roughly $1B of those bills a year. That's the top of this market. Now the bottom: AntSeed, the youngest player on Base, has done $65K in total volume. Ever. Surplus $0. Everything on this map lives between that billion and that zero. SURPLUS: usage, no monetization An open inference market on Base sellers compete on price for every request, payments run on crypto rails. The product's transaction metrics are at all time highs. I covered this in my last two posts. But: 100B tokens circulating since day one, around a thousand holders, thin liquidity, ~$5M market cap last time I checked. The protocol charges no fee. Revenue is $0 on purpose, the team is growing usage first. The honest version of the risk: product growth isn't wired to the token at all yet. The founder publicly asked the community how to fix that connection. The wires are sitting right there. Nobody's plugged them in. AntSeed: monetization, no token trading Launched May 15. Two people and seven AI agents, the whole idea is three months old. Also an open inference market, also Base, also USDC per request. A mirror image of Surplus assembled in reverse order. Onchain numbers: $64.96K gross volume, 194,000 micro payments, average settle 33 cents. 578 paying buyers, 140 sellers. The protocol already takes a 4% fee $2.5K collected so far. The ANTS token exists, but in a locked state: emissions accrue only for real network activity (epoch nine is running now, ~40M accrued out of a 1.04B hard cap), and transfers are blocked. You can't buy it, you can't sell it, there is no price. Build the economy first, open the market later. And the detail that impressed me most. The team caught farmers inflating volume for future rewards and froze ALL provider emissions until a validation system ships. Network revenue dropped roughly in half from the peak after that. They publicly cut their own best looking chart to avoid paying for fake volume. Young projects rarely have the stomach for that inflated numbers usually get nursed all the way to a listing. Venice: competitor and supplier at the same time base:0xacfe6019ed1a7dc6f7b508c02d1b04ec88cc21bf is the most mature token in this niche. Different model: stake the token, get a share of the platform's compute, plus buybacks and burns funded by actual revenue. Over 40% of supply burned already. Market cap in the hundreds of millions orders of magnitude above everyone else here. But that's not the interesting part. Venice has a second token base:0xf4d97f2da56e8c3098f3a8d538db630a2606a024 , tokenized compute. And the DIEM pool runs inside AntSeed. 112 people staked ~$439K worth of DIEM, the pool serves requests on the AntSeed network, and buyer USDC streams to stakers onchain. Current pool yield: 6.9% APY in stables. The operator fee is already accruing into a treasury earmarked to buy and burn ANTS. Read that again: a Venice token earns revenue inside a competing marketplace. Yes, the numbers are tiny $1.5K paid to stakers, total. But structurally it's the most interesting thing I've seen in this niche. These projects aren't fighting over the market. They're snapping into each other like lego. OpenRouter x402: the elephant The biggest threat to all three has no token at all. OpenRouter is migrating to x402, Coinbase's payment protocol stablecoin payments for inference straight over HTTP, no accounts, no API keys. x402 has already processed over 100M transactions, most of them on Base. If OpenRouter's $1B of volume moves onto crypto rails, that's the same open inference market at a thousand times the scale. And the value lands in USDC and Coinbase infrastructure. Holders of niche tokens get nothing from that. Unless their tokens grow a value capture mechanism before the giant arrives. What the map actually says They're all building the same thing: a market where AI agents buy intelligence for themselves and pay in stables. The only real difference is where the value flows. OpenRouter: value flows to the company and to USDC. Venice: to the token, through buybacks from real revenue. AntSeed: the mechanism is built and already dripping pennies, the token is still in its cage. Surplus: usage grows, value flows nowhere. Yet. And the whole niche, minus OpenRouter, is at hundreds of users. Not thousands. The dashboards look great, the charts go vertical, but the absolute numbers are one decent food truck's revenue. Any of these projects can die before "who captures the value" becomes a question worth real money. What I'm watching next: whether Surplus turns on a fee, whether AntSeed's revenue survives the farmer purge, and how much of OpenRouter's volume actually arrives on x402. I don't know which of them makes it to the big market. But the question "does the holder get anything when the product grows" doesn't run on trust here you can check it onchain, project by project, epoch by epoch. I'll keep checking. And writing down what I see. All numbers from onchain data and public dashboards at the time of writing. Of the tokens mentioned, I hold base:0xc52aedec3374422d7510e294cfaa90799595cba3 NFA.