If you need cloud compute capacity on demand, the window is closing. Amazon CEO Andy Jassy has confirmed that AWS's large committed capacity is largely sold out through 2027, with part of 2028 already allocated, and practitioners are already reporting difficulty securing On Demand instances for ordinary EC2 workloads. This is not a theoretical pinch, it is happening now, and similar constraints are appearing at other providers. The era of infinite, instantly available cloud resources is giving way to a world where planning and commitment are prerequisites, not options.
This shift forces a hard question: how do you build for the future when the infrastructure you depend on is already booked years ahead? The answer lies in rethinking how you consume compute. Locking into long-term reservations may feel like a safe bet, but it ties your hands as workloads evolve. A more flexible path is to explore tools that let you run the same work across multiple environments without rewriting your stack. For teams feeling the squeeze, the real opportunity is not in fighting for scarce On Demand slots, it is in designing systems that treat capacity as a variable, not a given. Amazon's own investment strategy, detailed in Amazon's path to a trillion in revenue runs through massive AI investment, shows they are betting big on AI infrastructure, but that bet concentrates capacity into specialized hardware like Bedrock, leaving general-purpose EC2 to compete for scraps. Meanwhile, startups like the one building prefabricated AI data centers on assembly lines, covered in This startup bets an assembly line can build AI data centers at scale, point to a future where capacity itself becomes modular and faster to deploy, but that is still years from easing today's crunch.
For most organizations, the concrete takeaway is this: treat On Demand as a luxury, not a right. If your workflows depend on spinning up instances at will, you are exposed. The smarter move is to pre-provision baseline capacity with reservations, then layer in spot instances or alternative providers for burst needs. This is not about abandoning the cloud, it is about using it with eyes open. The constraints Jassy described are not a hiccup; they are the new normal. The question worth watching is whether AWS will prioritize On Demand headroom for smaller customers or reserve it for the largest commitments that fill those 2027 contracts. The answer will tell you everything about who the cloud is really built for.