AI’s next phase to drive higher computing demand, not less, warns Deloitte

The artificial intelligence industry is entering a phase of even greater demand for computing power as companies shift from building models to deploying them at scale, Deloitte said in a report released on March 18.

The consulting firm’s annual Technology, Media and Telecommunications outlook challenges the view that the rise of “inference,” running AI models to answer queries, would reduce the need for expensive chips and large data centres.

Instead, Deloitte said inference will account for about two-thirds of AI computing by 2026, up from one-third in 2023, without easing pressure on infrastructure spending.

“The world likely needs all the data centres and enterprise on-premises AI factories that are currently being planned and all the electricity that these facilities will need,” Deloitte said in its TMT Predictions 2026 report.

The firm estimates global spending on AI data centres could reach about $400 billion in 2026 and rise to as much as $1 trillion annually by 2028.

Deloitte attributed the sustained demand to two emerging techniques, post-training scaling and test-time scaling, that significantly increase computing needs.

Post-training methods, including fine-tuning and reinforcement learning, can consume roughly 30 times the computing resources needed to train the original model, while test-time scaling, or long thinking, can require more than 100 times the computing power of a basic inference task.

“Most AI companies now use these techniques to make AI models better in various ways,” the report said, adding that both approaches are “AI compute hogs”.

Chip demand to remain strong

Deloitte said the market for inference-optimised chips could exceed $50 billion in 2026, but this will not replace demand for high-end graphics processing units (GPUs) used in AI training.

“Instead of the chip market being an ‘either-or,’ it looks like it will be a ‘both-and,’” the report said.

High-performance chips, which can cost more than $30,000 each, are expected to account for around $200 billion in spending in 2026, with the overall AI chip market projected to cross $400 billion by 2028.

Despite improvements in chip efficiency, Deloitte said demand for AI computing is growing four to five times each year and is expected to continue at that pace through 2030, with implications for global energy consumption.

Edge computing limited

The report said edge devices such as smartphones and personal computers will play a limited role in advanced AI workloads, even as many now come with dedicated neural processing units.

“Almost all AI computing performed in 2026 will be done mainly in the kind of giant AI data centres being planned or on relatively expensive high-end AI servers,” Deloitte said.

India data centre push 

In India, Deloitte expects data centre capacity to expand sharply from about 1.5 gigawatts in 2025 to around 10 gigawatts by 2030, driven by rising AI workloads, hyperscaler expansion and 5G adoption.

Power demand from data centres could reach about 57 terawatt-hours by the financial year ending March 2030, increasing the sector’s share of national electricity consumption to between 2.5% and 3%.

Deloitte said challenges including power availability, land approvals, water access and execution capacity will determine whether India can scale its infrastructure build-out.

While the firm noted that future breakthroughs could reduce computing needs, it said such a shift is unlikely in the near term.

“At some point, it is possible that new techniques could see a breakthrough... But that won’t be in 2026,” the report said.