Every major technology build-out requires two phases, according to Carlota Perez: the infrastructure to power it, & then the applications to pay for it.
Compared to the dotcom, AI is achieving in it half the time.
There are two markets in AI : applications & infrastructure.
Bardia Shahali has built sales teams both : first at Fireworks AI from roughly $3m to roughly $400m, & now VP of Sales at Granola.
Overall, they demonstrate that inference businesses can be quite healthy even at current prices & efficiencies.
I identify two distinct AI markets: applications and infrastructure.
I cite @BardiaShahali as a case study: he built sales teams first at Fireworks AI and now serves as VP of Sales at Granola.
I note that DeepSeek can generate 8× the tokens per hour compared with closed‑model competitors.
I track Microsoft’s AI segment growth at 40% year‑over‑year, with other firms modestly higher.
These observations shape my sales‑strategy framework: treat applications and infrastructure as separate value propositions, use Bardia Shahali’s career path as a scaling blueprint, and prioritize inference efficiency (e.g., token throughput) because DeepSeek’s 8× advantage signals a competitive moat. I also benchmark market momentum against Microsoft’s 40% AI growth.
In early October 2026 I highlighted that the AI inference market could grow from $160 B to $350 B within a year, making it the largest software market and overtaking databases.
The projected market size for AI inference in 2026 is $160 B and is expected to reach $350 B in 2027.
I project software‑company profit margins will shift from the historical ~72 % to a range of 30 %–50 % as inference cost structures dominate.
Berkeley’s study shows that the right harness can cut the cost of an answer by 71 % with no loss of accuracy.
AI inference could grow from $160B to $350B next year, overtaking databases as the largest market in software.
For software companies, that could mean margins closer to 30–50% than the usual 72%.
In early October 2026 I posted a thread on X analysing the emerging inference market and drawing parallels to the historical database market.
I estimate the inference market size at $350 B, based on a bottoms‑up calculation.
The market is growing at 200 % year‑over‑year.
I see inference as a stack—chips, memory, interconnect, power, orchestration, apps—each layer capable of hosting multiple large players.
- video (compute‑bound)
Just as databases split into relational, document, graph, time‑series and vector categories, inference will fragment into these slices, each with its own IPO‑scale potential.
I will measure market size by dollars spent rather than raw usage metrics.
While inference lock‑in is weaker than database lock‑in—more like a CDN than a tightly coupled engine—I still consider it a strategic factor.
These observations shape my decision to treat inference as a distinct market category and to focus on owning a single slice rather than the whole stack.
For further context see my LinkedIn market observations where I discussed the database‑to‑inference transition.
- long‑context (KV‑cache memory‑bound)
- edge (power‑bound)
- voice
- local