Amazon raises its 2026 AI and AWS spending outlook to $220 billion
Amazon just told Wall Street it will spend more on AI infrastructure this year and still not have enough. On the Q2 earnings call, CEO Andy Jassy raised 2026 capital spending guidance to about $220 billion, up from the ~$200 billion outlook the company held earlier in the year, per CNBC’s earnings report.
The consequence for buyers is blunt: even at that number, Jassy said Amazon will not have enough capacity to meet all demand in 2026, and he expects the same squeeze in 2027. Demand already booked into 2028, he said, is “striking.”
Why CapEx jumped another $20B
Amazon had guided to $200 billion in February and held that line in April. The new ~$220 billion cash CapEx figure is a $20 billion step-up in a single quarter’s outlook. Jassy pinned the increase on rising memory prices and surging AI/AWS demand.
Quarterly capital expenditures hit $54.2 billion in the June quarter, versus $32.1 billion a year earlier. Trailing-twelve-month free cash flow flipped to a $7.6 billion outflow, compared with an $18.2 billion inflow a year prior. That is the investor tension in one pair of numbers: AI demand is real enough to force more spend, and the cash return story is getting uglier while the company builds.
Wall Street had already expected a hike after Alphabet raised its own spending plans as high as $205 billion. Amazon cleared that bar and then some.
AWS growth and the $496B backlog
AWS is the demand engine behind the CapEx. Cloud revenue was $42.2 billion for the quarter, beating StreetAccount’s $40.54 billion estimate. Sales grew 37% year over year, against a 31% consensus, the unit’s fastest growth since 2021.
Jassy called AWS “booming” and said AI and custom-chip units each exceeded a $25 billion annual revenue run rate. The contracted backlog of AWS work not yet live reached $496 billion. That backlog is the tell: customers are signing for capacity Amazon has not finished building.
Companywide, Amazon beat on the top and bottom lines: revenue of $200.61 billion versus a $196.47 billion estimate, and adjusted EPS of $1.97 versus $1.82. Net income was $62.6 billion, or $5.75 per share, including about $53.4 billion pre-tax from investments in Anthropic. The stock jumped more than 10% in extended trading on the print.
Trainium and Bedrock in the spend mix
Amazon is not only buying NVIDIA-class GPUs and renting the narrative. Jassy highlighted homegrown silicon (Trainium and Graviton) and AI products such as the Bedrock model marketplace as growth pillars aimed at enterprises. Custom chips and AI services sitting above a $25 billion run rate each is Amazon’s way of saying the CapEx is not pure commodity GPU shopping.
Still, memory cost inflation and capacity shortfalls cut across vendors. Trainium and Bedrock do not erase a shortage of power, racks, and high-bandwidth memory. They change the mix of what Amazon can sell when it does bring capacity online.
What capacity shortfalls mean for buyers
If Jassy is right through 2027, enterprise AI roadmaps that assume “AWS will have a rack when we need it” are wrong. Expect longer lead times, tighter reserved-capacity fights, and pricing power that favors the cloud seller over the buyer who waited.
Near-term guidance already shows friction elsewhere: Amazon guided Q3 revenue to $197–$202 billion, below the $204.1 billion Street number, citing Prime Day moving into June and tough year-ago comps. Operating income guidance for the quarter is $22.5–$26.5 billion.
The strategic read is not subtle. A $220 billion CapEx plan that still leaves demand unmet is both a growth signal and a rationing warning. Microsoft and Google are spending hard too. Capacity is the scarce input. Teams that lock multi-year AWS commitments early get the GPUs and Trainium slices.
