the great ai lie
when more users costs more money
you see the headlines. another ai company hits a billion-dollar valuation. you almost believe the hype.
but you stop. you run the numbers.
and the math doesn’t work.
this is the dirty secret of the ai boom. we broke the number one rule of scalable software.
the problem isn’t what you think
for decades, software has run on a simple, beautiful principle: zero marginal cost.
companies like salesforce can add another customer for almost nothing. that’s how they hit 80% gross margins. it’s a license to print money.
ai is different.
every query, every image, every line of text generated burns actual money on gpus. your costs scale linearly with your success. on a good day, you’re looking at 50-75% margins.
this isn’t a software business. it’s a high-tech manufacturing line with worse economics.
why your old playbook is obsolete
your mba textbook is useless here.
that classic 3:1 lifetime value to customer acquisition cost ratio? it’s meaningless when your costs aren’t fixed.
your best, most engaged users can become your biggest liability. the moment they scale up, they can flip from profitable to loss-making. you’re not acquiring customers. you’re acquiring cost centers.
we need a new set of rules to value these businesses.
the real fix: playing a different game
some figured this out early. they didn’t try to win by out-spending the competition. they won by refusing to play their game.
the playbook is counterintuitive but effective.
- build on existing platforms to avoid infrastructure costs.
- engineer viral growth for zero-cost acquisition.
- shape demand to balance compute loads, turning a bug into a feature.
the result is a lean, profitable machine. contrast that with the giants burning billions to chase revenue, hoping they’ll find a path to profitability before the money runs out.
two kinds of ai companies
you can split the market in two.
“ai supernovas” achieve massive scale at negative margins. they’re betting they can optimize their way to profitability before the vc checks stop clearing. it’s a high-stakes race against time.
“ai shooting stars” grow steadily while protecting their margins from day one. they prioritize sustainable growth over vanity scale.
the market is sending a signal
look at the valuation gap.
private markets are still paying 20x+ revenue for the hottest ai companies. the public markets are pricing companies with decent growth at 4-5x.
that’s not just a pricing difference. it’s a prophecy. the public markets are demanding proof of a real business model, and the private markets are still running on hype. that gap will close, and it won’t be pretty.
the infrastructure layer is the only one immune. they’re selling the picks and shovels. they benefit from the gold rush without bearing the direct cost of model development. they have the margins everyone else is dreaming of.
what actually matters
if you want to know what’s real, ignore the press releases and track these numbers.
- compute cost as a percentage of revenue.
- customer concentration risk.
- a clear path to positive unit economics.
enterprise customers paying thousands per seat will always beat consumer subscriptions at $20 a month. production deployments integrated into critical workflows will always beat experimental budgets.
your technical moat is a mirage
that performance gap you’re so proud of? open source compressed the sota gap from 8% to 1.7% in one year. your technical moat will evaporate overnight.
lasting advantages are built elsewhere.
- data network effects: your product gets better with every user.
- workflow integration: it becomes too painful to replace you.
- regulatory compliance: you build a wall of lawyers that no startup can scale.
build something that gets stickier with use, not just marginally faster.
the coming consolidation
too many companies raised money at valuations that require perfect execution to justify. the market will consolidate, and it will be quick.
those without a clear path to profitability will get acquired for their talent or shut down.
in the end, unit economics trump everything. gravity always wins.