Gartner’s $239B AI forecast: Agentic workflows take half of GenAI model revenue

Click any chart or figure to open the full-resolution image. Original analytical charts include detailed tables; dollar values are shown to the nearest million where the source supports it. Published forecasts, calculated metrics, and illustrative allocations are identified separately.

Agentic workflows will drive 50% of generative AI model revenue by 2030, up from 5% in 2025. Gartner published that projection on September 17, 2026 inside its forecast analysis of the generative AI models market (G00861842). The number rewrites how every enterprise buyer should read the firm’s $239 billion AI platforms and models forecast.

Multistep reasoning, tool integration, and repeated validation steps multiply inference events per completed business process. Gartner quantifies the impact at $38 billion in additional spending by 2030 from scaling agentic workflows alone. Declining inference prices will not offset it. Lower unit costs unlock deeper automation, which drives up token volume per task faster than prices fall. Gartner calls this a structural tailwind for GenAI model spending.

That consumption flywheel sits inside a market growing from $39 billion in 2025 to $239 billion by 2030 at a 42.8% compound rate, per Gartner’s June 25 forecast (G00855897). I built this analysis from the two Gartner reports and the country-level dataset, which contains 1,316 rows across 9 regions and 7 years. Gartner supplies the underlying forecasts and forecast analysis; the charts identified as original and the enterprise-buying implications are my independent analysis. For the agentic spending crossover that sets the stage for this post, see Gartner’s $5.95 trillion AI forecast puts the chatbot era on a 2027 deadline (August 5, 2026).

$239 Billion by 2030 and Where It Breaks Down

How to read the numbers: Dollar spending levels are current U.S. dollars. Gartner’s reported segment growth rates and CAGRs use constant currency, with 2024 exchange rates; they therefore need not match growth calculated from the current-dollar totals. Regional CAGRs in this article are independent calculations from Gartner’s current-dollar data for 2025–2030. Figures are rounded, so components may not sum exactly. Forecasts are projections, not observed outcomes.

Gartner segments the market into two halves. AI platforms grow from $26.3 billion in 2025 to $100 billion by 2030, a 30.0% CAGR. GenAI models grow from $13 billion to $139.2 billion, a 59.9% CAGR.

GenAI models overtake AI platforms in 2027. By 2030, models command 58% of total spending.

• Total market 2026. $64.3 billion, up 60.7% year-over-year. Adding $25 billion in net new spending.
• AI platforms 2026. $36 billion, growing 34.7%. Data science and ML platforms account for $26.4 billion. App development platforms account for $9.5 billion.
• GenAI models 2026. $28.3 billion, growing 113.5%. Foundation models at $23.4 billion. DSLMs and specialized models at $4.9 billion.
• 2030 total. $239.2 billion. GenAI models at $139.2 billion. AI platforms at $100 billion.

Where platform and model spending goes

Gartner maps all four segments by 2030 market size and CAGR in its opportunities framework. Foundation GenAI models sit at $105.4 billion and 55.2% CAGR. DSLMs reach $33.9 billion at 83.4% CAGR, the fastest growth rate in the forecast.

Data science and ML platforms carry roughly three times the spending of app development platforms. Their spending ratio rises from about 2.8 to 1 in 2025 to 3.3 to 1 in 2030.

DSLMs Are the Fastest-Growing Segment in This Forecast

Domain-specific language models and specialized GenAI models grow at an 83.4% CAGR from 2025 to 2030. That is faster than foundation models at 55.2% and faster than AI platforms at 30.0%. The segment goes from $1.6 billion in 2025 to $33.9 billion by 2030.

The share shift tells the structural story. DSLMs represented 5.5% of GenAI model spending in 2024. By 2030 they will command 24.3%.

Slower percentage growth, larger dollar additions

Gartner’s constant-currency growth rates decelerate from 452.0% in 2025 to 34.5% in 2030. Current-dollar spending additions do the opposite. Each period adds more net new spending than the last, climbing from $1.3 billion to $8.6 billion per period.

The 2Q26 vs. 1Q26 outlook changes confirm the direction. Gartner revised its 2030 DSLM forecast upward by $14.3 billion and cut its 2030 foundation model forecast by $32.1 billion. The year-by-year revisions tell the same story at every point in the window. DSLM revisions climb from $189 million for 2025 to $1.3 billion for 2026 to $14.3 billion for 2030. Foundation model revisions run negative in every year, starting at $2.7 billion for 2025 and widening to $32.1 billion for 2030.

Agentic AI Rewrites Inference Economics

Gartner’s September forecast analysis identifies agentic workflows as the single largest driver of GenAI model spending growth through 2030. The mechanism runs on volume, not price. Agentic workflows consume more tokens per completed task than conversational AI. Each autonomous process generates multiple inference events as task complexity rises, demanding advanced reasoning, larger context windows, and repeated validation. I tracked the agentic spending crossover in Gartner forecasts agentic AI will overtake chatbot spending by 2027 (February 16, 2026). The September data confirms the trajectory is accelerating.

Gartner’s framework identifies three forces operating simultaneously. Agentic workflows add $38 billion. Multimodal expansion adds $32 billion. Open-weight substitution removes $25 billion. The net effect grows the addressable revenue pool while shifting where the money lands.

• Agentic share of GenAI revenue. 5% in 2025, growing to 50% by 2030.
• Spending impact. $38 billion in additional spending by 2030 from scaling agentic workflows.
• Token economics. Consumption shifts from model calls per user interaction to model calls per completed business process.
• Production commitments. 75% of foundation model monetization locked into multiyear commitments by 2030, up from 20% in 2025.

Gartner Forecasts Frontier Revenue Will Concentrate by 2030

Gartner expects two or three suppliers to dominate the GenAI LLM marketplace in North America and Asia-Pacific by 2030. Its more specific monetization forecast puts 85% of the foundation-model revenue pool in those regions with the top two vendors. That is a concentration forecast, not a prediction that all other vendors disappear. The firm’s 2025 market share data shows Anthropic, OpenAI, and Google collectively accounted for 63% of enterprise spending.

The concentration thesis extends beyond model quality. Sustained investment in model development, inference capacity, reliability, security, integration, and global distribution creates a cost structure that favors vendors able to keep reinvesting at scale.

With GenAI in the Trough of Disillusionment in 2026 per Gartner, enterprises are gravitating toward frontier models supplied by incumbent SaaS providers. That buying preference narrows opportunities for standalone frontier model companies and creates a direct winner-take-all competition to become the model provider of choice for software vendors. For context on how every credible forecast sizes this market, see my roundup of agentic AI forecasts and market estimates, 2026.

North America Commands 54% of the Market Through 2030

Gartner’s country-level dataset shows North America remaining near 54% of worldwide spending throughout 2024–2030. Spending rises from $13.2 billion in 2024 to $130 billion in 2030. The separately calculated 2025–2030 current-dollar CAGR is 43.4%, using a 2025 base of $21.5 billion.

The United States alone hits $33.1 billion in 2026 and $123 billion by 2030. Gartner’s China region grows fastest among these three markets at a 58.3% current-dollar CAGR from 2025 to 2030, expanding from $4.1 billion to $40.9 billion. This regional grouping includes China, Hong Kong, and Taiwan; it is not mainland China alone. Europe grows at a 40.8% current-dollar CAGR over the same period to $45.1 billion, with the United Kingdom at $4.8 billion in 2026, France at $2.2 billion, and Germany at $1.9 billion.

• North America 2030. $130 billion (54.4% share), 43.4% current-dollar CAGR (2025–2030).
• Europe 2030. $45.1 billion (18.8% share), 40.8% current-dollar CAGR (2025–2030).
• China region 2030. $40.9 billion (17.1% share), 58.3% current-dollar CAGR (2025–2030).

Open-Weight Models Are Eroding the Paid Revenue Pool

Gartner projects 30% of routine, high-volume enterprise GenAI inference will run on open models or enterprise-controlled models by 2030, up from 5% in 2025. That substitution reduces the GenAI model revenue pool by $25 billion by 2030.

Production traffic shows the economic split taking shape, but it is not a direct measure of Gartner’s worldwide forecast. Vercel’s AI Gateway routes tens of trillions of tokens between production applications and AI labs each month. In August, open-weight models processed 56% of gateway tokens but accounted for 14% of estimated spending, according to Vercel’s September production index.

The same report puts open-weight token share at 7% in December 2025. A later snapshot shows how quickly the mix can move: in a post amplified by Chamath Palihapitiya on September 19, Guillermo Rauch reported 78.4% open-weight and 21.6% closed-weight token volume. That reported snapshot is separate from August’s monthly share. Vercel’s leaderboard documentation says the percentages beside models default to the most recent day, or the day selected by hovering, rather than the aggregate for the full chart window.

A fixed daily comparison of tokens and spending

For a fixed comparison, Vercel’s September 18 daily export shows DeepSeek V4.1 Flash at 59.3% of all token volume. GLM 5.3 Flash accounted for 7.5%, DeepSeek V4 Flash 0731 for 2.7%, and Kimi K3 for 2.5%. These are shares for one day, not six-month averages.

The spending picture is different. Anthropic accounted for 64% of estimated gateway spend in August. The roughly 30% token-volume figure belongs to July: Vercel’s August index reported 65.1% of July spend on 30% of July token volume. That report also put the average price per Anthropic token at 4.4 times the average across every other lab, not 4.4 times the gateway-wide average.

The September 18 export makes the volume-spend gap concrete. Claude Opus 4.8 accounted for 13.7% of estimated spend, Claude Opus 5 for 9.0%, Claude Sonnet 5 for 5.5%, and Claude Sonnet 4.6 for 4.3%. DeepSeek V4.1 Flash, with 59.3% of tokens, accounted for 5.1% of estimated spend.

What gateway usage can and cannot establish

For enterprise buyers, the implication is to budget for workload mix rather than token volume alone. Vercel’s production data shows that the models processing the most tokens are not necessarily the models capturing the most spending. That supports evaluating lower-cost models for suitable workloads while reserving more expensive models for cases where their performance justifies the cost. The aggregate shares do not, by themselves, establish which models handled the most complex tasks.

This volume-spend split is consistent with the pricing pressure behind Gartner’s projected $25 billion reduction in the paid model revenue pool. It does not independently validate that worldwide estimate. Gateway spending and model-provider revenue are different measures, and a single gateway is not the whole market.

Palihapitiya’s September 19 post made a separate prediction: within 12 months, the top three models would be open source. He named Nebius, Iren, Baseten, Together, and Fireworks as the clouds he expected to benefit. That is his forecast, not a conclusion established by Vercel’s usage data.

A caveat on the Vercel data: it covers a specific population routing traffic through AI Gateway, not market share for the entire AI industry. Vercel estimates spending using published list prices; actual bills may differ. Token volume includes input, output, reasoning, cached-input, and cache-creation tokens. The September report also broadens the open-weight classification relative to earlier reports and notes that prior months may be revised. Monthly findings, daily exports, and social snapshots should therefore be read as distinct observations, not interchangeable measures.

What This Means for Enterprise Buyers

Enterprise AI budgets are coming under greater scrutiny, with spending shifting toward providers who can demonstrate clear value across cost, latency, performance, and reliability. Five takeaways for enterprise leaders making AI platform and model decisions over the next 12 months. For context on how security spending fits this picture, see Gartner’s $248.9B security forecast makes securing AI the only segment accelerating through 2030 (July 6, 2026).

1. Build for model routing, not model loyalty. The market is fragmenting by workload. Enterprises that lock into a single model provider risk overpaying for tasks a smaller, specialized model can handle at the required quality. The DSLM forecast supports evaluating specialized models rather than assuming every task needs a frontier model.
2. Budget for agentic inference volumes. Agentic workflows consume significantly more inference tokens per completed task than conversational AI. If your AI budget is sized for chatbot-level consumption, it will not survive contact with production agentic workloads.
3. Evaluate open-weight alternatives for routine workloads. 30% of high-volume enterprise inference shifts to open or enterprise-controlled models by 2030 per Gartner. The Vercel production data shows that shift is already underway. Start identifying which production workloads can move now.
4. Watch the consolidation timeline. If frontier revenue concentrates as Gartner forecasts in North America and Asia-Pacific by 2030, enterprises should assess provider resilience and migration options. Concentration does not imply that every smaller provider exits.
5. Demand platform-level governance and cost attribution. The AI platform market grows to $100 billion by 2030 because enterprises need orchestration, evaluation, cost visibility, and policy enforcement. For a deeper look at the governance gap, see Gartner’s $244.2B security forecast shows enterprises spend 17x more on AI tools than securing AI itself (March 24, 2026).

Bottom line

Gartner’s forecast puts a number on the structural tension running through every enterprise AI budget. The total market nearly quadruples from $64 billion in 2026 to $239 billion by 2030. Agentic workflows drive half of GenAI model revenue by the end of that window. Open-weight models absorb a growing share of token volume while frontier providers retain the lion’s share of spend. The $25 billion revenue pool reduction and the $38 billion agentic spending addition operate simultaneously, pulling the market in two directions at once.

Enterprises that build for model routing, budget for agentic consumption growth, and negotiate multiyear commitments with the surviving frontier providers will be better positioned for where this market lands in 2030. Enterprises that treat AI spending as a single-vendor procurement decision will not.

Related on Software Strategies Blog

• Gartner’s $5.95 trillion AI forecast puts the chatbot era on a 2027 deadline (August 5, 2026)
• Gartner’s $248.9B security forecast makes securing AI the only segment accelerating through 2030 (July 6, 2026)
• Gartner’s $244.2B security forecast shows enterprises spend 17x more on AI tools than securing AI itself (March 24, 2026)
• Roundup of agentic AI forecasts and market estimates, 2026 (February 26, 2026)
• Gartner forecasts agentic AI will overtake chatbot spending by 2027 (February 16, 2026)

This post is my personal reflection on Gartner’s AI platforms, models, and generative AI research from an industry analyst perspective. It does not represent my employer.

Sources and methodology

• Gartner, Forecast: AI Platforms and Models, Worldwide, 2024-2030, 2Q26, Arunasree Cheparthi, Amarendra, Radu Miclaus, John Lovelock, 25 June 2026, G00855897. Press release.
• Gartner, Forecast Analysis: Generative AI Models, Worldwide, 2026, Arunasree Cheparthi, Radu Miclaus, John Lovelock, 17 September 2026, G00861842.
• Vercel AI Gateway, dated model export, September 18–19, 2026; the model-share comparisons and chart use September 18 rows, retrieved September 20, 2026. Leaderboard methodology. Data licensed under CC BY 4.0.
• Vercel AI Gateway Production Index, September 2026, vercel.com/blog/ai-gateway-production-index-september-2026, covering data through August 2026.
• Vercel AI Gateway Production Index, August 2026, vercel.com/blog/deepseek-overtakes-google-on-volume-cost-per-token-falls, covering data through July 2026.
• Guillermo Rauch, quoted in Chamath Palihapitiya’s September 19, 2026 post; open-weight share cited as a reported snapshot, separate from the monthly index.
• Chamath Palihapitiya (@chamath), post on X, September 19, 2026, x.com/chamath/status/2101406709231337710.