Amazon Raises 2026 Capital Spending Forecast to $220 Billion on AI Memory Costs

Amazon Raises 2026 Outlook on Surging Memory Prices

Amazon on Wednesday lifted its projected 2026 capital spending to a record $220 billion, a sharp upward revision driven by soaring memory chip prices that have dramatically increased the cost of building and equipping AI data centers. The company disclosed the new figure in an investor update, saying that higher prices for DRAM and high-bandwidth memory (HBM) — essential components in AI servers — are forcing the technology giant to spend more than originally budgeted to maintain its infrastructure expansion pace. The announcement sent ripples through supply-chain stocks and reignited broader questions about how long the AI infrastructure boom can sustain its current trajectory.

The $220 billion target represents a meaningful jump from the roughly $180 billion the company had signaled just months earlier for 2026, with Amazon citing memory costs as the primary culprit. According to the company, memory components now account for a substantially larger share of total server build costs than they did in 2024, and procurement prices have continued to rise quarter after quarter. The company said the revised outlook accounts for both higher unit prices and the need to secure supply volumes well in advance through long-term agreements with memory manufacturers. Amazon executives emphasized that the spending increase is not a change in strategic direction but rather a necessary adjustment to prevailing market conditions.

Why Memory Costs Are Soaring

The price surge traces directly to the same force fueling Amazon's spending: the explosive growth of generative AI. AI accelerator chips, such as Nvidia's data-center GPUs, rely heavily on HBM — an advanced type of memory that stacks chips vertically to deliver ultra-high bandwidth. Demand for HBM has outstripped supply for several consecutive quarters, and the few manufacturers capable of producing it — primarily SK Hynix, Samsung, and Micron — have been operating at full capacity. The shortage has pushed contract prices for both HBM and conventional DRAM upward at rates not seen in decades, with some analysts estimating that memory now represents as much as 60% of the bill of materials for a high-end AI server.

The supply crunch is not expected to ease quickly. Building new fabrication capacity for advanced memory takes years, and manufacturers have been reluctant to lock in capacity expansion at historically cyclical prices. This creates a structural tension: hyperscalers like Amazon, Microsoft, and Google all need guaranteed memory supply to meet their own AI deployment timelines, but memory makers are cautious about betting on sustained demand. The result is a market where buyers lock in long-term contracts at elevated prices rather than risk having no supply at all. Amazon's revised capex guidance suggests the company has chosen the former path — securing volumes aggressively despite the premium.

A Broader Industry Shift

Amazon is not alone in feeling the pinch. Its hyperscale peers have all raised capital spending forecasts in recent quarters, and memory costs have been a recurring theme in earnings calls across the technology sector. Microsoft and Alphabet have each committed hundreds of billions of dollars to AI infrastructure through 2026, and both have warned shareholders that rising component costs could compress margins even as revenue grows. The collective effect has been a demand shock that has rippled through the entire semiconductor supply chain, benefiting memory makers but pressuring cloud providers' profitability.

For decades, cloud providers benefited from relentless declines in computing and storage costs. That long-standing assumption is now being tested. Memory prices have historically followed boom-and-bust cycles, but the AI-driven surge has been unusually persistent, and industry analysts increasingly argue that the era of cheap, abundant memory may be over for the foreseeable future. This has profound implications for how cloud computing is priced. AWS, Amazon's cloud unit, has not yet announced broad price increases, but analysts widely expect the company to pass on at least some of its higher infrastructure costs to customers over time.

What the Spending Means for Amazon's Business

The capex increase puts renewed pressure on Amazon's free cash flow and near-term profitability, even as the company's core retail business continues to generate steady earnings. Amazon has historically been willing to sacrifice short-term margins for long-term infrastructure dominance, a strategy that has served it well in the past but that is now being tested by the sheer scale of AI investment. In its shareholder communications, the company has framed the spending as a defensive necessity: failing to invest adequately today could mean losing ground to rivals in the AI cloud race — a race that is still in its early innings.

The investments are concentrated in AWS's data-center footprint, with significant spending on new regions, networking equipment, and AI-optimized server racks. Amazon has also moved to vertically integrate parts of its chip supply chain, including its custom Trainium and Inferentia processors, which are designed to reduce reliance on Nvidia GPUs. While those chips still require HBM, Amazon's custom designs give it slightly more flexibility in how memory is configured, potentially mitigating some of the cost pressure over time. Analysts noted, however, that the design advantage would not meaningfully offset the current memory price cycle, at least in the near term.

Looking Ahead

Amazon's revised guidance all but guarantees that 2026 will be a historic year for global data-center investment. If memory prices remain elevated through the end of next year, other hyperscalers will likely be forced to follow with similar upward revisions. For enterprise customers, the outlook suggests that AI cloud services will not get cheaper — at least not yet. Companies that provision AI capabilities from AWS and its rivals should expect pricing pressure to continue, particularly for compute-heavy training workloads.

The longer-term question is whether today's aggressive investment will ultimately be rewarded. Amazon is effectively betting that AI demand will continue to grow for years, justifying the massive upfront costs. History offers some reassurance: the buildout of the internet in the late 1990s and the prior cloud infrastructure expansion both involved periods of heavy spending that eventually produced outsized returns for the companies that persisted. The less comfortable possibility is that the industry overshoots — building far more capacity than demand can support — echoing the fiber-optic overbuild of the early 2000s. For now, Amazon's leadership is clearly betting that caution costs more than ambition.