The comfort of old mental models
Every new technology cycle starts the same way: “we explain the new thing using the language of the old one”.
AI startups today are being framed as SaaS, but better. Recurring revenue, expansion, enterprise contracts, gross margins that will “improve with scale”.
It sounds familiar. Comforting, even. And that’s exactly the problem.
Why SaaS logic breaks in AI
SaaS worked because marginal cost was effectively zero. Once the software was built, selling it again was almost free. The entire venture model of CAC payback, LTV expansion, land-and-expand, was designed around that assumption.
AI breaks it.
In AI, every interaction has a cost. Inference is not free. Latency matters. Compute prices fluctuate. Usage is unpredictable.
Yet many AI startups still pitch like nothing has changed.
ARR without a cost curve
They show ARR curves without addressing cost curves. They talk about usage growth without explaining who absorbs the variable expense when usage explodes. They assume scale fixes margins, when in AI, scale often exposes them.
And this is not a theoretical concern, it’s already happening.
When usage becomes the enemy
We’re seeing companies with incredible demos and real demand struggle as soon as customers actually use the product the way it was intended. Heavy users become margin destroyers. Enterprise contracts look attractive until you realize pricing was designed for adoption, not sustainability. What worked at 10 customers quietly collapses at 1,000.
The uncomfortable truth is that AI businesses reintroduce something SaaS largely eliminated: cost discipline as a core strategic function.
Pricing is no longer a growth lever
In SaaS, pricing was a growth lever. In AI, pricing is survival.
Token-based pricing, usage-based pricing, and flat fees have trade-offs, and none are neutral. Underprice usage, and margins evaporate. Overprice it, and adoption stalls.
The challenge isn’t just finding the right model, it’s aligning pricing with value, not compute.
Infrastructure is strategy
This is why infrastructure decisions matter far earlier than founders expect. Model choice, hosting strategy, fine-tuning vs APIs, caching, orchestration, these aren’t technical footnotes. They are the business model.
In SaaS, infrastructure was an optimization problem. In AI, infrastructure is strategy.
What winning AI companies will do differently
The companies that win won’t be only the ones with the most impressive models.
They’ll be the ones who understand their cost structure better than their customers understand the product.
They’ll design experiences that guide usage, not encourage abuse. They’ll be honest about margins early, even when it makes the story less sexy. They’ll treat gross margin not as a future outcome, but as a design constraint.
A different category, different rules
The market will eventually catch up to this reality. When it does, the AI startups that survive won’t look like traditional SaaS companies. They’ll look closer to infrastructure businesses with product-level UX, or platforms where economics are engineered, not assumed.
AI is not SaaS 2.0. It’s a different beast, with different rules.
And the faster founders and investors stop pretending otherwise, the better the outcomes will be.
Portfolio Companies
News about our Portfolio Companies
Kala raised US $55 million in funding to promote financial inclusion.
Cashea was adopted by the informal economy to boost toy sales this past December 24th.
Portfolio companies currently raising investment rounds
Qurable: An AI-powered platform that unifies customer data, integrates payments, and enables scalable personalized experienceshelping brands, creators, and audiences build relationships beyond transactions.
Wisecricket: A platform that automates financial audits using AI: monitoring invoices, contracts, and fiscal compliance in real time.
Please don’t hesitate to contact us in case you are interested in connecting with any of these amazing companies.
Sapiens: A Brief History of Humankind - Yuval Noah Harari. In Sapiens, Yuval Noah Harari walks through 70,000 years of human history to show how myths, not biology, shaped our societies; and why the stories we believe may matter more than the technologies we invent