Product1 publisher3 min readPublished
Agentic AI's 2030 market more than doubles depending on who draws the category line
Grand View Research and MarketsandMarkets both compound the agent market at about 46% a year and finish $28.12bn apart in 2030. The gap is in what each firm counts, so a plan quoting either figure is betting on a category boundary.
The Product Desk · Product desk
What happened
- Grand View Research tracks enterprise agentic AI deployment specifically and projects it from $2.6bn in 2024 to $24.5bn by 2030, a compound annual growth rate of 46.2%.
- MarketsandMarkets draws its boundary around agent software in general and puts the market at $52.62bn, up from a $5.26bn base in 2024, at a 46.3% compound rate over the same horizon.
- Three agent payment protocols landed in four months: Stripe and Tempo's Machine Payments Protocol in March 2026 with more than 100 integrated services, Coinbase's x402 to the Linux Foundation in April, and Mastercard's Agent Pay for Machines in June.
- Stacklok's 2026 survey of senior technical leaders found 29% of software organizations running MCP in limited production and 12% in broad production, with security ranked the leading barrier ahead of cost.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint Because both forecasts compound at nearly the same rate, the size a business case inherits is set by where the analyst drew the category line, and demand growth never closes a 2.15x boundary gap.
- exposure If the money agent sellers are chasing is application software budget being displaced, their revenue depends on incumbent vendors losing renewals, and the buyer's procurement cycle sets the pace.
- decision A seller in the payments layer has to commit to one of four competing standards before it can bill a single agent transaction, and its integrated customers inherit that choice.
- contradiction Binance's position is that shared infrastructure spares developers from rebuilding plumbing, while TNW says enterprise evidence shows integration's budget share is not falling, so the saving cannot be assumed in a plan.
Somebody has to put one of these numbers on slide four of a budget request, and nobody in that room is going to read the methodology note. Grand View Research tracks enterprise deployment; MarketsandMarkets draws a wider boundary around agent software in general [1][2]. TNW's own reading is that a forecast range is a decent proxy for how well a category has been defined, and that agentic AI is currently defined differently by everyone measuring it [3].
The two 2024 bases sit 2.02x apart and the 2030 destinations 2.15x apart, a spread of $28.12bn [6][4][5]. Grow each base out over the six years and Grand View's multiplies by 9.4, MarketsandMarkets' by 10.0, which is its 2024 figure with a zero on the end [7]. The write-up describes that horizon as "four years" in one sentence and the "same six years" in the next [8].
What matters to a seller is which layer of the budget it bills for. TNW splits agent budgets into data, execution, identity and payments, and says only one of the four has a settled standard [9]. Of payments it writes: "Four standards, four sets of incumbents, no consolidation" [11].
Binance went the other way on August 20, 2026, opening market data and trading to compliant agents through an MCP endpoint instead of building a proprietary connector [12]. "That makes it easier to create AI-driven financial applications without having to recreate the underlying infrastructure each time," said Jeff Li, VP of Product at Binance [13]. TNW treats that as testable: if shared infrastructure removes rebuild cost, integration's share of agent budgets should be falling, and it says most evidence on enterprise deployment points the other way [14].
The most current data the category has is protocol counts, which TNW says are easy to misread [28]. Anthropic counted more than 10,000 active public MCP servers in December 2025, and monthly SDK downloads reached roughly 97 million by March 2026 against about 100,000 in the month of launch, a factor of about 970 [15][16][17]. A May census found 15,926 GitHub repositories carrying the mcp-server topic against 9,652 latest-version records in the official registry, 61% of the repository count [18][19]. About 59% of the organizations Stacklok surveyed were not running MCP in production of any kind [21]. Among servers in the wild, 8.5% implement the OAuth 2.1 that the specification makes mandatory for remote deployments [22].
The demand-side figures come in a different unit. Gartner projects machine customers controlling roughly $30tn of purchases by 2030 and agents commanding $15tn of business-to-business purchases as soon as 2028; McKinsey's narrower estimate for agent-mediated consumer commerce is $3tn to $5tn by 2030 [25][26]. Gartner's 2028 B2B figure is about 285 times the larger of the two 2030 supply-side forecasts [27].
Two things sit next to whichever number goes in the plan. One is the layer being billed and whether its standard is settled or contested. The other is where the money comes from, a new budget line or an application renewal that someone else is cancelling. Gartner puts $234bn of enterprise application software spend at risk from agentic AI. That is displacement of existing budget, and about 9.6 times Grand View's entire 2030 market [23][24].
What to watch
- Whether either firm restates its 2024 base in its next update, since the base moves the 2030 destination more than the growth rate does.
- Consolidation among the agent payment protocols. That would settle the layer a seller has to choose in today.
- A rise in the share of MCP servers implementing the mandatory OAuth 2.1, the barrier Stacklok's respondents ranked first.