Introduction
Three quarterly earnings calls. Three upward revisions. One shared reality: AWS, Google Cloud, and Microsoft Azure have each posted record-breaking Q2 2026 results — while simultaneously acknowledging that their combined investment firepower still falls short of demand. The total infrastructure spend planned by the three hyperscalers this year is approaching $595 billion, and all three have publicly stated they cannot serve every customer who wants capacity.
Triple-Digit Growth, Persistent Shortfall
The numbers speak for themselves. Google Cloud posted 82% year-over-year revenue growth in Q2, reaching $24.8 billion. AWS grew 37% to $42.2 billion. Azure recorded a 43% increase. Taken together, global spending on cloud infrastructure services exceeded $143 billion in the April–June 2026 quarter alone — an annualised run rate of more than $500 billion.
To fund this expansion, Amazon has raised its annual capex guidance to $220 billion; Google is now targeting $195–205 billion; and Microsoft is holding course at approximately $175 billion. That adds up to roughly $595 billion being deployed in 2026 to build out data centres, extend networks, and rack the compute that the global AI economy requires.
The Capacity Paradox
What makes this situation genuinely unusual is that despite these staggering sums, all three providers have spoken openly about their inability to fulfil total demand. AWS leadership has stated it will not have sufficient capacity to serve all of its customers through 2026. Google's CFO has confirmed that demand continues to outpace the investments currently underway. These are not marketing disclaimers — they reflect a real physical constraint.
The shortage is hitting multiple layers of the supply chain at once. GPUs, high-bandwidth memory, and storage remain under acute pressure, with procurement lead times for high-end hardware now stretching to nine to twelve months. Because the hyperscalers have absorbed the bulk of available chip allocations from manufacturers, enterprises looking to build their own on-premises AI infrastructure are facing delivery windows pushed back to early 2027.
What This Means for CIOs
The operational implications are often underestimated. When organisations cannot procure hardware on their own timelines, they turn to the cloud — only to run into quota restrictions and waitlists for GPU instances. Multi-month delays to provision AI compute capacity on major platforms have become routine, even for large, well-resourced enterprises.
This environment is reshaping how technology leaders must approach infrastructure planning. Reserving cloud capacity in advance — the same way you would secure budget allocations or software licences — is fast becoming a core operational competency in its own right. Long-term commitment vehicles — reserved instances, committed use contracts — are regaining strategic appeal, not primarily for the cost savings, but simply to guarantee access to a resource that has become genuinely scarce.
Compute Access as a Competitive Differentiator
A fundamental shift in the nature of the constraint is underway. Not long ago, the barriers to cloud adoption were regulatory, cultural, or financial. Today, for AI workloads in particular, it is the physical availability of compute resources that dictates project timelines. The organisations best positioned for the next phase will be those that treated capacity planning with the rigour of industrial or logistics forecasting — anticipating their needs several quarters ahead, not several weeks.
At $595 billion in 2026 commitments, the hyperscalers are building as fast as any organisation in history has ever built anything. But when demand breaks records quarter after quarter, even that figure does not guarantee a place in the queue.