Amazon raised its 2026 capital expenditure forecast from roughly $200 billion to approximately $220 billion, with CEO Andy Jassy attributing the increase directly to higher memory costs on the company's second-quarter earnings call. That confirmation strengthens the case for Micron Technology (NASDAQ: MU), one of the few memory manufacturers capable of supplying high bandwidth memory and advanced DRAM at scale. The risk is that memory pricing is inherently cyclical, and a real-time supply squeeze is not a structural guarantee.
What Jassy said and what he didn't hedge
Jassy's disclosure was specific. He told investors that the higher cost of memory was pushing the 2026 capex number above the prior estimate of roughly $200 billion, and he did not frame it as a temporary line item. He added that Amazon will not have enough capacity to meet all demand in 2026 or 2027, and that demand already visible for 2028 is "striking." That last point extends the demand horizon well past the quarters analysts typically model.
The capex backdrop
Amazon spent roughly $132 billion on capital projects in 2025. The revised $220 billion forecast for 2026 is a substantial step up, with Amazon Web Services driving the bulk of the spending. Data center construction, custom silicon development, the Project Kuiper satellite constellation, and warehouse robotics all draw from the same budget. Memory costs are specifically what pushed the figure past $200 billion, per Jassy. Separately, Alphabet, Microsoft, Amazon, Meta Platforms, and Oracle are collectively expected to spend more than $700 billion in capital expenditures across 2026.
The read-through to Micron
High bandwidth memory and advanced DRAM are the components that AI accelerators consume at scale, and supply is not keeping pace with demand. When a hyperscaler revises its own budget higher because memory is expensive, pricing power belongs to the producers. Micron Technology is among the few manufacturers with the scale to supply both. That is a supply-chain fact, not a forward earnings estimate.
The counterargument
Memory markets have compressed before. Prior technology buildouts generated demand signals that looked durable until new manufacturing capacity came online and deflated pricing faster than consensus expected. Nothing in Jassy's remarks establishes when the current constraint resolves, only that it exists and is costing Amazon billions in 2026. Investors who treat the shortage as permanent are reading more certainty into the data than the data provides.
On balance, the $20 billion revision is the most direct public quantification of memory scarcity to emerge from a hyperscaler earnings call. When other major cloud providers report their own quarters, the line to watch is whether they name the same constraint.