Rob

It’s already happening.

The issue with most prediction market startups is that they’re building features, not apps

do agents get sunday scaries checking their micropayment balance?

Compute isn’t venture. It’s infra with venture demand curves and utility contract structures The right comp is 90s era independent power producers, except the offtakers have stronger balance sheets and steeper demand than anything in energy history Most investors will miss it

They're continuing to generate anti-data center propaganda with AI and I can't help but laugh

tokenised stocks are the next stablecoins

The internet’s business model was built around attention: free content in exchange for ads. That worked because humans were doing the browsing, and humans can be marketed to. But agents don’t look at ads. As agents start acting on our behalf, the economics of the open web start to break. Why publish for free if your audience is mostly bots? One answer is micropayments. An agent fetching one article or hitting an API once shouldn't have to subscribe at $10 a month. The natural unit of pricing is a fraction of a cent per request. Another path is that advertising mutates. Websites can slip sponsored information directly into their content to influence the agents.

After almost six years, I think this is it. The Chain Wars have concluded.

we keep blaming 'not enough compute' when the real bottleneck is moving data in / out of memory. for agent style products (everything were getting high on rn), requests come one at a time and in bursts so there is not really a 'great' way to batch them efficiently. so for those buying more gpus w/o looking at the algo efficiency, the result is just lighting $$$ on fire. IMO the opportunity for a startup is making single request inference cheap and predictable. think ai lab for performance engineering where highest spend is on inference and performance like latency. the unlock comes from squeezing memory use and cutting waste, things like cache compression, smarter decoding, quanitzaiton that keeps quality/routing each request to the cheapest acceptable path. the opp here is to figure out cost for these large companies because they got customers to keep happy.

Now live: $XYZ100, $BRENTOIL, $GOLD, and $DOGE leveraged tokens