AI model and platform spending to reach US$64 billion as buyers tighten scrutiny

AI model and platform spending to reach US$64 billion as buyers tighten scrutiny

Gartner expects the market to grow 63.4% this year, with the sharpest growth coming from smaller, specialized models rather than the frontier systems that dominate headlines.

Worldwide end-user spending on AI models and platforms will total US$64.25 billion in 2026, up 63.4% from US$39.31 billion last year, according to a forecast published by Gartner on July 20. The growth rate is high by any measure. The detail underneath it describes a market where buyers have started asking harder questions about what they are getting in return.

"Enterprise AI budgets are coming under greater scrutiny, with increased focus on usage efficiency, cost control and measurable outcomes," said Arunasree Cheparthi, senior principal research analyst at Gartner. Spending, she said, is moving toward providers that can demonstrate value on cost, latency, performance and reliability.

Gartner divides the market into two halves. Spending on the models themselves will grow 117% this year, while spending on the platforms used to build and run them rises 36.9%. Models will account for roughly 44% of the total in 2026, up from a third in 2025.

Within the model half, foundation generative AI systems remain the bulk of the spending, growing 104.2% to US$23.36 billion from US$11.44 billion in 2025. The platform half is split between AI application development platforms, forecast to reach US$9.54 billion at 38.6% growth, and platforms for data science and machine learning, which grow 36.3% to US$26.44 billion.

That last figure makes data science and machine learning platforms the largest single segment in the market, ahead of foundation models. The category covers the software organisations use to prepare data, train models and put them into production, and it predates the current generative AI cycle by years. Its growth rate is the slowest of the four segments, but it remains the layer the rest of the market is built on.

The fastest-growing segment is also the smallest

Domain-specific language models, or DSLMs, are forecast to grow 210% this year, rising to US$4.91 billion from US$1.58 billion. The percentage is striking mainly because the starting point is low. Even after tripling, the segment will account for less than 8% of total spending.

DSLMs are trained or tuned for a narrow purpose rather than for general capability, covering work such as reading clinical notes, processing insurance claims or answering questions about a specific product catalogue. They are smaller than frontier models, which makes them cheaper to run and faster to respond, and they can often be hosted on hardware an organisation already operates rather than accessed through a third-party interface.

That matters in markets where data residency rules restrict what can be sent outside an organisation, and where AI budgets are measured more tightly than they are at the largest US and Chinese cloud providers. A model that costs less per query and stays inside the firewall answers two constraints at once, which is a large part of why Gartner expects the segment to triple.

The US$64 billion covers models and the software used to build and manage them. It does not cover the hardware underneath. Gartner has separately forecast total worldwide AI spending of more than US$2.5 trillion in 2026, and expects spending on data center systems alone to pass US$788 billion this year.

On that comparison, the model and platform layer accounts for roughly 2.5% of what organizations will spend on AI in 2026. Most of the money is still going into buildings, power and chips.

Cheparthi's argument is that commercial advantage is shifting toward suppliers that make AI usage legible to the people paying for it. She pointed to providers that build evaluation, cost transparency and usage tracking into customer workflows, an area that has grown into a category of its own as organizations run several models side by side and try to route each task to the cheapest one that can handle it.

The same shift creates a problem for the sellers. As pricing moves from fixed licences to consumption, revenue depends on continued use rather than on renewal. "As more models enter the market and usage-based pricing becomes harder to predict, buyers will turn to platforms that help them choose the right tools, monitor performance, enforce policy and keep costs under control," Cheparthi said.

Gartner's own research offers a caution on how much of the forecast spending turns into working systems. The firm has predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. A spending forecast counts what buyers commit, not what survives contact with production.