AI firms may be judged less by how quickly they grow than by whether each sale leaves enough money to pay for the computing behind it. In a Forbes Business Council essay, investor Mala Ramakrishnan argues that customer mix and inference costs are becoming decisive business questions, not tidy details for the finance team.
Watch Desk analysis
What happened
Ramakrishnan says AI queries carry ongoing compute costs that can squeeze gross margins, unlike the relatively low cost of serving an additional customer in traditional software. She cites reported figures that put OpenAI’s blended gross margin at about 33% and Anthropic’s in the mid-60% range, with its API business reportedly above 80%. Those figures are claims in her essay, not independently established comparisons here.
Her argument is that enterprise and developer customers can bring more predictable, higher-value revenue than a large free consumer audience. She also points to practical ways operators can control costs: route simpler tasks to smaller models, cache repeated context, batch requests and use more efficient hardware. The catch is that agentic workflows can use many more tokens per task, so lower prices per token do not automatically mean lower total bills.
Why it matters
If serving each customer remains expensive, growth alone may leave an AI company needing more capital rather than moving towards profitability. That makes pricing, customer mix and engineering choices part of the same business problem. It also complicates the familiar promise that cheaper computing will eventually fix the economics: usage can expand to fill the savings.
For operators, the useful takeaway is to track the cost of delivering a task alongside what customers will pay for its result. A model that is cheaper per token may still be the wrong choice if it needs more tokens, while an efficient system does not help much if prices fail to capture the value it creates.
Our read
This is a useful lens on the AI business race, provided we treat the margin comparisons as reported estimates rather than settled scorekeeping. The strongest point is practical: unit costs, usage and pricing have to be considered together. “We’ll fix the economics at scale” is less a strategy than a very expensive hope.
What to watch
- Whether AI firms publish comparable, clearly defined margin figures.
- Whether customer revenue grows faster than the cost of serving heavier workloads.
- How much model routing, caching and batching improve total costs in practice.
- Whether pricing changes as AI products move from chat towards agentic tasks.
Discussion spark: Should AI companies prioritise enterprise customers and tighter usage economics, or keep pursuing broad consumer adoption even if it delays profitability?
Sources and evidence
- The Gross Margin Divide That Now Separates AI's Winners – Forbes (30 October 2026, 13:45 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.