Gartner Predicts AI Inference Costs Per Agentic Workflow Will Increase More Than Fivefold Through 2028
On August 17, 2026, Gartner, the world's leading information technology research and advisory firm, released an important forecast: AI inference costs per agentic workflow will increase more than fivefold through 2028. This forecast highlights a core contradiction in enterprise AI transformation: although the token cost per inference is declining, as AI evolves from simple chatbots to multi-step autonomous execution systems, total AI costs for enterprises will actually rise significantly.
Core Forecast: 5X Growth in Inference Costs
According to Gartner's forecast, AI inference costs per agentic workflow will increase more than fivefold through 2028. The logic behind this forecast is clear: compared to basic chatbot interactions, routing a task to an agentic reasoning model increases provider inference costs by at least five times, and the cost gap often widens further as task complexity grows. In other words, while AI agents let enterprises complete more complex tasks with fewer tokens, the total compute consumed per workflow far exceeds traditional conversations.
Gartner analysts point out that AI agents reason, replan, call other agents, and work continuously in the background. This multi-step, multi-round interaction execution model causes the inference cost per workflow to grow exponentially. For enterprises transitioning from chatbots to agentic AI, this is a key variable that must be incorporated into budget planning in advance.
$2.59 Trillion: The Total AI Spending Pie
Alongside the inference cost forecast, Gartner also updated its overall global AI spending projection: by 2028, global AI spending is expected to reach $2.59 trillion. This figure covers AI-related hardware, software, services and talent spending, reflecting that AI has become the core direction of global enterprise IT investment. Notably, this forecast is higher than previous estimates, mainly because the adoption of generative AI and agentic AI has exceeded expectations.
Gartner emphasizes that while AI infrastructure spending (such as GPU procurement and data center construction) takes up a large share of budgets, what truly determines the ROI of enterprise AI investment is application-layer efficiency. As agentic AI becomes more widespread, inference costs will continue to rise as a share of total AI spending, becoming a key cost item that enterprise CFOs and CIOs need to watch closely.
The Cost Decline Paradox: Why Total Costs Rise Instead
A thought-provoking phenomenon: the token prices of large language models are falling rapidly, with multiple mainstream model providers cutting prices several times over the past year, some by more than 90% per unit. However, Gartner's forecast shows enterprise AI costs are rising rather than falling. The reason lies in changing usage patterns: enterprises are no longer just occasionally asking AI questions but letting AI agents autonomously complete entire business processes.
For example, a traditional customer service chatbot may only need one model call to answer a user question, costing less than $0.01. But an AI agent completing a task like 'process a customer refund request' may require multiple reasoning steps, CRM system calls, ticket generation and notifications. The entire workflow may need dozens or even hundreds of model calls, with total cost potentially dozens of times higher. This phenomenon of 'buying tokens for less money but using more tokens' is the core logic behind Gartner's forecast.
Enterprise Strategy: From Cost Center to Value Center
Facing rapidly growing inference costs, Gartner recommends that enterprise product leaders adopt a series of countermeasures. First, establish a refined AI cost tracking system that accounts inference costs by workflow and business unit to identify the most expensive AI use cases. Second, adopt a tiered model strategy, selecting appropriate models based on task complexity: lightweight models for simple tasks, high-performance reasoning models only for complex tasks. In addition, inference efficiency can be optimized through caching, batch processing and model distillation.
Gartner also emphasizes that enterprises should focus on the actual business value of AI investment rather than simply pursuing cost minimization. If an AI agent workflow can replace higher-value business processes or create new revenue sources, even higher inference costs are worthwhile. The key is to establish a cost-value evaluation framework so that every cent of inference cost corresponds to quantifiable business output.
Sources
Frequently Asked Questions
Q1: Why do AI costs rise when token prices are falling?
A: Because usage is exploding. AI agents require multiple reasoning steps, calls to other systems, and continuous background work. The number of model calls per workflow is dozens or even hundreds of times that of traditional chat. Unit cost declines cannot keep pace with usage growth, so total costs rise instead of falling.
Q2: What is an agentic workflow?
A: An agentic workflow is the process by which an AI system autonomously completes a full business process, including understanding goals, making plans, calling tools and systems, executing multi-step tasks, and verifying results. Unlike one-shot Q&A, agent workflows are multi-round, multi-step autonomous execution processes.
Q3: How can enterprises control AI inference costs?
A: Key methods include: establishing refined cost tracking systems, adopting tiered model strategies (lightweight models for simple tasks), optimizing with caching and batch processing, reducing inference overhead through model distillation, and establishing cost-value evaluation frameworks to prioritize high-ROI AI use cases.
Q4: Where does the $2.59 trillion AI spending go?
A: It covers AI hardware (GPUs, servers, data centers), AI software (models, platforms, applications), AI services (consulting, implementation, operations) and related talent spending. Infrastructure hardware spending accounts for the largest share, but application-layer spending is growing fastest.
🔧 Related AI Tool Recommendations
Quickly extract core content from long texts, improve information processing efficiency
Detect whether text is AI-generated, assess content authenticity
Intelligently rewrite text for different business scenarios
Conclusion
Gartner's forecast of 5X growth in AI inference costs serves as a wake-up call for enterprises embracing agentic AI. It reveals a key fact: AI's cost structure is shifting from pay-per-conversation to pay-per-workflow autonomous execution, placing new demands on enterprise budget planning, technology selection and ROI evaluation.
The good news is that AI's potential is far from fully realized. Through refined cost management and value-driven investment strategies, enterprises can fully control costs while extracting greater business value from AI agents. The future competition lies not only in who uses AI, but in who uses it smarter and more economically.