Record Capital In, Almost Nothing Measurable Out
The money and the results are hard to reconcile. Last quarter, AI startups pulled roughly 80% of global venture funding, with a handful of names taking close to two-thirds of every venture dollar. At the same time, MIT’s research found that 95% of AI pilots did not move the P&L, and surveys put the share of companies seeing AI in their earnings at well under half.
Record capital flows in one end. Almost nothing measurable comes out the other. That gap has appeared with every platform shift, and it always closes the same way: not with a better label, but with a company that points a suddenly cheap capability at one specific, expensive problem and builds something worth paying for.
Labels Rotate. The Queue Doesn’t.
Labels do not die. They pass the premium to whatever comes next. Client-server gave way to the web, then cloud, then mobile-first, then generative AI, and now “agentic.” The rush to stamp the newest label on old work is not a sign the term is fading. It is a sign it is at the height of its value.
What does not rotate is the underlying business problem. A real-world problem is not a category like “AI for customer service.” It is a queue with a name, a volume, a cost per unit, and a person whose calendar it eats. Those problems were there before anyone said “agentic,” and they will outlast the word. The companies pulling ahead pick one pain point, execute well, and partner smartly. It is scope discipline and delivery, not model size.
ElevenX Capital’s Perspective
At ElevenX Capital, we treat the model as a rented input that gets cheaper every few months. That frees us to focus on the assets that are genuinely hard to replicate, the unglamorous ones software has fought over for thirty years.
Distribution is one. An incumbent can ship an AI feature to fifty thousand seats on a Tuesday while a better-engineered startup spends eighteen months getting past procurement. Workflow depth is another: the approvals, exceptions, integrations, and compliance that nobody wrote down, earned slowly and hard to transfer. AI makes a good product a little better and a bad one faster at being bad. The moat was always somewhere else.
In Five Years
In five years, “AI company” will mean about as much as “SaaS company” does today. A few will be extraordinary, and nearly all of them will have AI inside, the way they all run on a cloud backend now.
Trace the ones worth a fortune back to their origin, and they will not land on a model or a funding round. They will land on a founder who got hold of one specific, profoundly boring problem and refused to let go while everyone around them changed vocabulary four times.
The hype is always loud. The compounding stays quiet.
