AWS recorded its quickest development in additional than 4 years throughout the second quarter, however Amazon mentioned it nonetheless lacked sufficient computing capability to satisfy buyer demand regardless of elevating its 2026 capital spending forecast to $220 billion.
AWS income rose 37% 12 months over 12 months to $42.2 billion within the quarter ended June 30. The outcome exceeded the 31.21% development anticipated by analysts surveyed by LSEG and marked the cloud unit’s strongest efficiency in 18 quarters.
“AWS is booming proper now,” Amazon CEO Andy Jassy mentioned throughout the firm’s earnings name. AWS added greater than $4.6 billion in income from the earlier quarter, about 80% greater than its earlier largest quarterly improve.
Amazon elevated its forecast for 2026 money capital expenditure from roughly $200 billion to $220 billion. Jassy mentioned the extra spending would nonetheless not present sufficient infrastructure to satisfy all of the demand Amazon anticipated throughout the 12 months.
“Even at that quantity, we are going to nonetheless not have sufficient capability to satisfy all the demand we’ve got in 2026,” Jassy mentioned. He anticipated the capability constraint to proceed in 2027.
The “lion’s share” of AWS computing capability deliberate for 2027 had already been reserved by clients, Jassy mentioned, whereas some capability scheduled for 2028 had additionally been dedicated. AWS backlog reached $496 billion on the finish of the quarter, up from $364 billion three months earlier.
The reservations present Amazon with advance details about buyer necessities because it plans information centres and computing infrastructure. The corporate didn’t establish the purchasers concerned or disclose how a lot capability would stay for organisations procuring assets nearer to deployment.
Buyer commitments prolong past present capability
AWS gives merchandise that permit clients to order specified computing assets prematurely. EC2 Capability Blocks present entry to some GPU-based situations for outlined durations, whereas On-Demand Capability Reservations safe chosen EC2 capability inside an Availability Zone.
These merchandise are separate from the longer-term buyer commitments that Amazon mentioned for 2027 and 2028. The corporate didn’t say whether or not these commitments would have an effect on on-demand availability, pricing, or allocation for smaller enterprise clients.
Most of Amazon’s present AI capability is contracted for phrases of at the least 5 years, in accordance with Jassy. These commitments give the corporate customer-demand visibility earlier than it purchases gear, though Amazon didn’t disclose how the contracts are distributed throughout buyer teams.
Amazon’s disclosures verify that infrastructure availability is limiting the demand AWS can presently serve. They don’t set up that capability has change into a extra vital aggressive issue than pricing throughout the broader cloud market.
Jassy mentioned Amazon begins spending on information centres about two years earlier than the services open. The corporate due to this fact incurs building and gear prices earlier than clients can use the brand new capability and earlier than the services start producing income.
Increasing cloud capability requires websites, electrical energy connections, buildings, servers, and networking gear. The event timeline means a better capital expenditure finances doesn’t instantly translate into further AWS capability.
A US Division of Power advisory report mentioned grid-connection requests for hyperscale services requiring between 300 MW and 1,000 MW or extra can contain lead occasions of 1 to a few years. It additionally recognized constraints involving transformers, switchgear, transmission infrastructure, and different gear required to provide giant information centres.
The Worldwide Power Company expects world electrical energy consumption from information centres to double between 2025 and 2030, with electrical energy use by AI-focused services anticipated to triple. It additionally recognized energy gear, superior chips, and different part constraints as elements limiting how shortly new capability can enter service.
AI workloads additionally require greater than entry to accelerators. Giant coaching and inference clusters depend upon reminiscence, storage, cooling, energy distribution, and high-bandwidth networking.
Amazon mentioned software program and course of enhancements, customized chips, and internally developed community gear have been serving to it use server and community capability extra effectively. The corporate recorded $53.1 billion in money capital expenditure throughout the second quarter, primarily associated to AWS and generative AI.
Reminiscence and infrastructure prices improve spending
Jassy mentioned increased reminiscence prices have been the principle purpose Amazon elevated its 2026 capital expenditure forecast by roughly $20 billion. Amazon didn’t disclose how the upper costs would have an effect on the quantity of computing capability delivered by that spending.
Reminiscence is used all through AI servers. Excessive-bandwidth reminiscence provides information to accelerators throughout coaching and inference, whereas standard server reminiscence helps the broader computing system.
Amazon didn’t specify whether or not the rise associated primarily to high-bandwidth reminiscence, standard server reminiscence, storage, or a mix of these classes. It additionally didn’t present a breakdown of spending on processors, networking, storage, cooling, and electrical infrastructure.
Amazon mentioned its AI and semiconductor companies had every exceeded an annualised income run charge of $25 billion. AWS reached an annualised income run charge of $169 billion throughout the quarter.
Jassy mentioned AI adoption was additionally growing clients’ consumption of core AWS companies. These embrace computing, storage, databases, and networking used to help AI purposes and their related information.
AWS working earnings rose to $16.6 billion from $10.2 billion a 12 months earlier. The division recorded a 39% working margin throughout the quarter, alongside the 36.7% improve in income.
Capital expenditure weighs on free money movement
Amazon recorded a trailing 12-month free money movement outflow of $7.6 billion, in contrast with an influx of $18.2 billion a 12 months earlier. Purchases of property and gear elevated by $66.1 billion over the identical interval, primarily reflecting AI funding.
Amazon’s trailing 12-month working money movement rose 33% to $161.4 billion. The rise in property and gear purchases was nonetheless giant sufficient to maneuver free money movement into unfavourable territory.
The hole displays the timing of Amazon’s infrastructure programme. Knowledge centre building begins about two years earlier than opening, whereas servers and networking gear are typically bought a number of months earlier than they enter service.
Jassy mentioned Amazon has stronger visibility into buyer necessities earlier than buying servers and networking gear as a result of these belongings are put in nearer to deployment. He mentioned Amazon wouldn’t make these gear purchases if the corresponding buyer demand was absent.
Jassy mentioned information centres can generate income for greater than 30 years with out requiring Amazon to repeat the preliminary building expenditure. Servers and networking gear function on shorter funding cycles.
On common, servers and networking gear take barely lower than three years to achieve break-even, Jassy mentioned. He added that the gear has a helpful lifetime of at the least 5 to 6 years, leaving one other two to a few years to generate free money movement after the preliminary funding has been recovered.
(Photograph by Martin Woortman)
See additionally: Microsoft expands Azure AI with Databricks and Mistral
Wish to be taught extra about Cloud Computing from business leaders? Take a look at Cyber Safety & Cloud Expo going down in Amsterdam, California, and London. The great occasion is a part of TechEx and is co-located with different main know-how occasions, click on right here for extra data.
Cloud Computing Information is powered by TechForge Media. Discover different upcoming enterprise know-how occasions and webinars right here.
