How big is the AI inference market compared to databases?

Tom Tunguz writes that AI inference reaches roughly $350b by 2027, passing databases at $190b by nearly 2x. This year the inference market reaches about $130b, near Gartner's $161b database forecast.

Why would AI inference change software margins?

Tunguz says infrastructure becomes the dominant part of COGS, potentially more than half of revenue, and the blended 72% gross margin of classic software will decay. He adds that BYOK deals trade revenue for margin.

What is happening to AI token prices?

Capability-adjusted token costs fall about 47% each quarter, but frontier list prices keep resetting upward, rising from $0.96 to $11.25 in 18 months before GPT-6 Sol reset the tier to $4.00.

Large Language ModelsAI Business & IndustryTomasz Tunguz (Theory Ventures)12h agoOct 6, 2026, 01:00 JST

AI inference to pass databases at $350b by 2027

Send on LINE

3 Key Points

What happened
In 2025 companies paid about $25b to run AI models; this year the market reaches roughly $130b, near Gartner’s $161b database forecast, and by 2027 inference reaches ~$350b versus databases’ $190b.
Why it matters
That crossover would make inference the largest software category, turning applications into resellers of inference and reshaping their growth and infrastructure costs, according to Tom Tunguz.
What to watch
The shift hinges on whether falling capability-adjusted token costs and model competition offset rising frontier prices; watch GPT-6 Sol’s $4.00 tier as the reference point.

WHO IT HITSThis lands hardest on application founders and their finance teams: inference usage bills may soon exceed seat licenses, forcing changes to pricing, sales compensation and cost forecasting.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Tom Tunguz frames inference not as a feature of AI applications but as the market that will define software economics for the next few years. The comparison to databases is deliberate: in the mid-2010s, a startup needed only 2% of the database market to IPO. Inference, he writes, is about to surpass that category.

The numbers he cites show how quickly the transition is happening. In 2025 companies paid about $25b to run AI models. This year the market reaches roughly $130b, close to Gartner's $161b database forecast, and by 2027 inference reaches ~$350b versus databases' $190b. That would make every application a reseller of inference.

The stakes hinge on whether falling capability-adjusted token costs and model competition can offset rising frontier prices. If they do, application margins may hold; if not, infrastructure could consume more than half of revenue. Founders, CFOs and sales leaders will feel this first in their pricing and payback math.

FAQ

How big is the AI inference market compared to databases?: Tom Tunguz writes that AI inference reaches roughly $350b by 2027, passing databases at $190b by nearly 2x. This year the inference market reaches about $130b, near Gartner's $161b database forecast.
Why would AI inference change software margins?: Tunguz says infrastructure becomes the dominant part of COGS, potentially more than half of revenue, and the blended 72% gross margin of classic software will decay. He adds that BYOK deals trade revenue for margin.
What is happening to AI token prices?: Capability-adjusted token costs fall about 47% each quarter, but frontier list prices keep resetting upward, rising from $0.96 to $11.25 in 18 months before GPT-6 Sol reset the tier to $4.00.

Tomasz Tunguz (Theory Ventures)Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.