Information Matters has published its first quarter 2026 research report on the agentic AI market. The central estimate for 2026 is $40 billion, with a range of $33–$48 billion. That number sits deliberately between two much louder ones: the $12–18 billion you get from application-layer analyses that exclude the foundation-model layer, and the three-digit billions you get from content-factory research firms applying an arbitrary compound growth rate to an arbitrary base. Our figures are built bottom-up from primary sources — earnings calls, SEC filings, company announcements, first-party developer surveys. No aggregator research firm is cited as primary data.
[Download the full Agentic AI Market Report and Forecast Q1-2026 report (PDF)]
The single most consequential assumption in the report is the attribution rate we apply to foundation-model revenues. Anthropic and OpenAI together now have approximately $54–55 billion in annualised run-rate revenue as of April 2026. Including that in full would produce a market figure that is largely a measure of two companies. Excluding it entirely — as many application-layer analyses do — understates what is actually happening. We apply a 25–35% attribution rate, representing the share of API calls and product usage that involves genuinely agentic behaviour (tool use, multi-step reasoning, autonomous action). We arrive at that range from four sources: Anthropic’s own product disclosures, a16z’s enterprise CIO survey, OpenAI’s product mix, and — new this quarter — Salesforce’s agentic-work-unit telemetry, which supplies a useful triangulation on the ratio of agentic to generative token consumption.
The Anthropic trajectory deserves particular scrutiny. The company’s run-rate revenue went from $9B at end-2025 to $14B in February, $19B in March, and $30B in April. That is $16B of run-rate growth in roughly eight weeks. More than 1,000 of Anthropic’s customers now spend $1M+ annually on Claude, a number that has more than doubled since February. This is not the shape of a bubble. It is the shape of a business that is compute-supply-constrained on the back of genuine, compounding enterprise adoption. The Broadcom deal and expanded Google partnership announced in April are the tell: demand is not the bottleneck.
The most interesting data point in the enterprise platform layer is Salesforce’s Q4 FY26 disclosure, which on our read is two stories, not one. On the usage side, Salesforce reported 2.4 billion agentic work units delivered (+57% quarter-on-quarter) and 19 trillion tokens processed (5× year-on-year). That is production-scale consumption inside deployed accounts. On the commercial side, Agentforce ARR of $800M across 29,000 deals implies an average contract value of roughly $28,000 — a pilot metric, not a transformational commitment. Heavy usage, light commitment. The single highest-signal disclosure of 2026 will be Salesforce’s FY27 Q3 report, which should show whether median deal size in the top quartile of accounts is growing materially. If it is, the application-layer forecast lifts toward the bull scenario. If it is not, the 2027–2028 numbers need to come down.
The coding agent segment remains the clearest evidence of production-scale adoption anywhere in the stack. Cursor at $2B ARR. GitHub Copilot at 4.7 million paid seats. Claude Code at approximately $1B ARR within six months of public launch. The JetBrains Developer Ecosystem Survey (24,534 respondents) puts regular AI-tool use at 85% of developers. The same survey’s April 2026 update shows that only 41% of employers have a formal AI-tools policy. The policy-to-practice gap is a near-term procurement headwind and a medium-term tailwind: the conversation in enterprise procurement is no longer whether, but which vendor.
The 2030 forecast is $140B in the central case, $88B in the bear case, $215B in the bull case. We deliberately do not set the range with a CAGR assumption; we set it with three named, observable triggers. Whether enterprise pilots convert to production — measurable in Salesforce FY27 Q3. Whether the first frontier-model IPO filing discloses gross margin above 70% at inference — measurable in Anthropic’s likely 2026–2027 S-1. Whether EU AI Act enforcement actions begin producing compliance-driven procurement — measurable from late 2026 onwards. Each trigger either lifts the base case toward the bull scenario or pulls it toward the bear.
New in this edition is a curated list of twelve smaller companies to watch — drawn from the 564-company Information Matters proprietary database and deliberately excluding the names already covered in the competitive-landscape section. The list clusters into evaluation and observability, voice-agent infrastructure, and workflow/vertical specialists. It will rotate every quarter: companies that hit their named signals stay on; those that miss rotate off.
We will refresh this report every quarter. The Q2-2026 edition is scheduled for July.
[Download the full Agentic AI Market Report and Forecast Q1-2026 report (PDF)]
