QWE AI Academy How big is the AI market really – and can I forecast my own slice without a paid report?
You’ve seen the headlines. One firm says the AI market hits a couple trillion soon. Another triples that. For your product or pitch, those macros are almost useless. You need a number you can defend for your niche.
Build a bottom-up AI market forecast yourself in under an hour. Use any free LLM to structure the model and math, then feed it only numbers you can source. Top-down reports give order-of-magnitude context. Bottom-up gives the figure investors – and your own hiring and budget calls – actually trust.
$2.67T vs $602B for “2026.” Same year, wildly different scopes. As of late 2026, Gartner’s spending forecast puts worldwide AI near $1.79T in 2025 and $2.67T in 2026 (49.5% growth), with infrastructure about $1.48T – mostly compute and data centers. MarketsandMarkets lands elsewhere: roughly $602B in 2026 to $3.64T by 2033 at 29.3% CAGR (hardware near half in 2026), per their June 2026 release. Grand View starts lower still – about $391B in 2025 to $3.5T by 2033 (industry analysis). Statista’s series sits in another band again. The spreads aren’t “who’s wrong.” They count different baskets.
Think of published forecasts like weather apps for the whole continent. Fine for packing a jacket. Useless for deciding if you need an umbrella on your exact street.
Top-down reports vs bottom-up DIY
Top-down: grab a published total, multiply by segment %, geography %, hoped-for share. Fast. Easy to inflate.
Bottom-up: count real buyers who match your ICP, realistic annual contract value (ACV), penetration you can actually hit. Multiply. That’s the market you can touch.
| Aspect | Top-down | Bottom-up |
|---|---|---|
| Speed | Minutes with a report | 30-60 min + verification |
| Defensibility | Low (borrowed numbers) | High (your assumptions) |
| Use when | Early go/no-go scan | Pitch, budget, hiring plan |
| Common gap | N/A | Often 20-40% vs top-down |
Operators and investors lean bottom-up for a reason: unit math you can argue. Turns out the 20-40% gap between methods isn’t failure – it’s a diagnostic. Something in company count, ACV, or penetration is off. Fix the assumption, not the spreadsheet font.
So if macros disagree by trillions and your niche isn’t “all of AI,” what number should you put on the slide? Only one you built from countable buyers.
Build the winner: bottom-up AI market forecast walkthrough
Pick one tight niche. Example: AI contract-review for mid-market legal teams in the US and UK (50-500 person firms).
- Define the ICP hard: company size, industry, buyer title, geography, tech stack must-haves.
- Count the universe. Free sources first – LinkedIn Sales Navigator filters, government business registries, trade association directories, Crunchbase free tier. Write the exact number and source date.
- Estimate reachable share and ACV. From your calls or competitor pricing pages: what do they pay now? Realistic conversion in years 1-3?
- Multiply: addressable companies × penetration × ACV = annual market you can touch.
- Add growth carefully: a conservative CAGR from a named report only for outer years – or keep it flat if unsure.
Hand the skeleton to an LLM. Never ask it for the size. Ask it to build the empty model.
You are a market analyst. Help me build a bottom-up TAM/SAM/SOM for [exact product one-liner] targeting [ICP details].
List every input I must supply (company count, ACV, penetration, sources). Do not invent any numbers. Output a simple table I can fill, plus the formulas. Flag where top-down cross-check would help.Paste the filled table back. Ask for arithmetic check and the three riskiest assumptions. Then verify every cell yourself against primary sources. Debatly’s market-research guidance matches what practitioners see: AI is strong at structure and math checks – and weak at verified current market numbers or trustworthy citations if you let it freestyle.
Pro tip: Run the same prompt in two models (ChatGPT and Claude). If the frameworks disagree on steps, your ICP is still too vague.
Cross-check once with a filtered top-down slice from Statista’s AI outlook or the MarketsandMarkets summary. Within ~30% of that filtered view? Credible zone. Bigger gap? Revisit company count or ACV before you touch CAGR.
Where DIY AI market forecasts break
Definition mismatch kills more forecasts than bad math. One shop’s “AI market” is chips and cloud. Another is pure software. Gartner’s spend lens includes the infrastructure wave; software-only totals look tiny beside it. Borrow a figure? Label year and scope in the same cell.
LLMs still invent plausible sources. “According to a 2025 IDC report…” often means a dead link or a title that never existed. Users paste those into decks. Force a verify column on every external claim – no exceptions.
Penetration is the quiet failure mode. 5% of a huge TAM sounds humble until it equals thousands of enterprise deals a two-person team cannot close. Cap SOM at what sales capacity can touch in 24 months.
FAQ: AI market forecast questions
Do I need paid reports to make a credible forecast?
No. Free filings, pricing pages, LinkedIn counts, and 8-10 of your own interviews beat a generic paid PDF for niche work. Paid macros are only the outer cross-check.
What if my bottom-up comes out tiny?
That’s useful data. A ~$3M serviceable market for solo dog-groomer booking software is a lifestyle business, not a venture bet. Better in week one than after six months of build. Expand geography or adjacent ICPs only after the core math holds – not to decorate a slide.
How often should I refresh an AI market forecast?
People treat the first model as permanent. It isn’t. Quarterly is enough for fast niches; sooner if pricing, regulation, or a hyperscaler capex shock hits. Keep the same framework. Swap only changed inputs – earnings transcripts, competitor pages, new registry counts. The mistake isn’t a stale CAGR. It’s rebuilding the whole story every time one cell moves.
Open your notes app. One sentence: exact ICP + the single number you need (year-1 revenue potential or three-year SOM). Drop the prompt above into chat and fill the first empty table before lunch. That’s your AI market forecast – owned by you, not a press release.