Microsoft AI CEO says compute scarcity defines 2026 winners

Mustafa Suleyman, the CEO of Microsoft AI, has asserted that the next era of artificial intelligence will be defined by the economics of "inference compute" rather than the pursuit of the smartest possible models.

Talking to X on Monday, Suleyman argued that the ability to afford and run models at scale—rather than just building them—will determine the industry's winners over the next two to three years.

This shift marks a significant departure from previous years, where the primary emphasis was on training increasingly larger foundation models.

For 2026, the central challenge has become serving these models to millions of users in real-time, a task that now consumes approximately two-thirds of all AI compute spending globally.

The industry faces a severe infrastructure bottleneck, with GPU lead times stretching to nearly a year and high-bandwidth memory remaining out of stock through 2026.

Furthermore, while 16 GW of data centre capacity is planned globally for this year, only 5 GW is currently under construction.

Suleyman’s "flywheel" strategy suggests that high-margin products, such as Microsoft 365 Copilot and specialised healthcare SaaS, are best positioned to thrive. These products can absorb premium inference costs, providing lower latency and better user experiences.

This creates a compounding effect: better performance leads to higher retention and more proprietary data, which in turn allows for superior model tuning and further adoption.

Microsoft's financial commitment to this vision is immense, with the company investing over $80 billion annually in AI infrastructure.

This scale has already yielded results; as of Q2 FY2026, paid Copilot seats have reached 15 million, representing a 160% year-on-year growth.

However, Suleyman warns that cash-constrained startups and consumer applications may struggle to maintain this pace, as they lack the margins to pay for necessary tokens.

While some advocates suggest that open-source models or "intelligence per dollar" may bridge the gap, Microsoft remains focused on using its financial prowess to secure a dominant position in the global AI landscape.