Google Cloud parent company Alphabet reported $51.4 billion in capital expenditures for the three months ended Sept. 30, outpacing Microsoft for the third time this year as both companies raced to add cloud capacity.
Microsoft, which remained ahead of Google Cloud in market share, spent $45.2 billion on capex in the same three-month period, the company’s EVP and CFO Amy Hood said during a Wednesday earnings call.
With the three largest hyperscalers — AWS, Microsoft and Google Cloud — pouring roughly $2 billion per day into capital investments, big numbers have become a point of pride. Amazon boasted $75.9 billion in capex in Q3, primarily tied to AWS cloud and AI investments, Amazon SVP and CFO Brian Olsavsky said during an earnings call Thursday.
“We still expect 2026 cash capex of approximately $220 billion, the majority for AWS, and expect to spend meaningfully more in 2027,” Olsavsky said.
Google surpassed Microsoft this year as Amazon held the lead
Hyperscaler capex is a KPI in the AI economy. Spending translates into compute to train and deploy large language models, chips to run AI applications and cloud capacity to house massive data estates.
Amazon President and CEO Andy Jassy underscored the urgency of infrastructure buildouts. “Would it be more convenient if this type of acceleration and opportunity happened over 25 years and you could spread the capex over 25 years?” he asked. “Yes, it might be more convenient. But that's not the world that we live in — and that's not the opportunity.”
Microsoft executives took a different tack. Hood attributed the company’s comparatively modest capex during Q1 of its 2027 fiscal year to a delay in third-party capacity deliveries and pledged to spend more than $50 billion on AI-related investments during the current quarter, as memory and storage chip prices continue to rise.
“We currently expect capex spend to accelerate primarily from CPUs, inclusive of higher memory and storage costs, and GPUs as well as the third-party capacity deliveries that shifted to H2,” Hood said. “As we continue to improve the efficiencies of our infrastructure, each dollar of capex has the potential to support more revenue and higher returns over time.”
Chip flexing
In addition to buying compute capacity, hyperscaler capex funds AI chip design and production.
Google introduced the eighth generation of its homegrown tensor processing units in April — TPU 8t for LLM training and TPU 8i for running agentic workloads. While TPUs are a key part of Google’s cloud infrastructure, the company has only recently ramped up a full sales operation.
“For customers who want TPU systems in their data centers, we're making TPU systems available in their data centers,” Google Cloud SVP and CEO Thomas Kurian said. “We're also broadening availability to other providers.”
Oracle will become the first cloud provider outside of Google to deploy TPUs next year, the two companies announced on Oct. 27. In April, Google agreed to supply Blackstone-funded neocloud Crux AI with TPU processors that are also expected to come online in 2027.
“Nearly 100 customers use TPUs today,” Kurian said. “In addition to our work with Anthropic, both directly and via Broadcom, and public reference customers like Apple, who are scaling pre‑training of their foundation models on the latest generation of TPUs, we're also working with new frontier model companies.”
Microsoft is newer to the AI chip business. The company rolled out the first version of its Azure Maia AI accelerator chip in November 2023, brought the Maia 200 into production this year and plans to have the third generation Maia 300 available in 2027.
“Maia 200 is already delivering up to 1.7 times more tokens per watt across AI experiences in Teams and Copilot,” Microsoft Chairman and CEO Satya Nadella said. “Maia 300 will drive further gains, delivering 40% more tokens per watt compared to leading GPU-based systems.”
Cost optimization is also baked into Amazon’s chip development strategy. AWS Graviton and Trainium chips saved customers more than $2 billion in Q3, according to Jassy. The company plans to make Graviton available outside of AWS next year, he added.
Market shifts
As AI steers hyperscaler capex strategy, it’s accelerating growth in the massive global market for cloud services.
Spending on cloud has nearly tripled since OpenAI unleashed ChatGPT in late 2022, according to Synergy Research Group data. In Q3, cloud revenue grew 55% year over year to $168 billion, setting a new high for quarterly cloud spend, the analyst firm said in a Thursday report.
“These are simply unprecedented numbers,” SRG Chief Analyst John Dinsdale said in an email. “We thought we had aggressive market forecasts in place, but they just weren’t aggressive enough.”
The three largest cloud providers are reaping the rewards.
Amazon’s revenues grew 12% year over year to $201.9 billion, with AWS net sales increasing nearly 40% to $184 billion in annualized revenues. Microsoft reported $92.3 billion in revenues, a 19% year-over-year bump, and Google Cloud revenues more than doubled year over year, contributing $30.7 billion to Alphabet’s $127.6 billion in Q3 revenues.
The balance of power among the three cloud giants hasn’t shifted much in the last several years. AWS leads the pack with 26% of the market, followed by Microsoft and Google Cloud at 19% and 16% respectively, according to SRG. However, their combined share has declined from more than three-quarters of the market in Q3 2022 to just 61% in the most recent quarter.
AWS remained the largest cloud provider in Q3
“What we now see is AI technology propelling cloud markets in a way that was unimaginable just four years ago,” Dinsdale said in the Friday report. “The leading cloud providers are benefiting from surging revenues, but they are also facing a much-changed market landscape, with AI labs and neoclouds emerging as both partners and competitors.”
Oracle, Alibaba, SpaceX/xAI, CoreWeave, Salesforce and Anthropic together captured 17% of the global market in Q3. OpenAI, Snowflake, and Databricks made it into the top 20 and Nebius wasn’t far off, according to Dinsdale.
Neocloud providers outfitted with AI processing power could capture as much as 20% of the market by 2030, according to Gartner. Connectivity gaps remain a problem for rising neoclouds and a potential opportunity for channel partners that can provide network infrastructure, Channel Dive sister company Omdia noted in an April report.
Securing reliable revenue streams will also be an issue for cloud upstarts.
“Many of the newcomers will find profitability a huge challenge over the coming years, and there will be consolidation,” Dinsdale in an email. “Meanwhile, they are causing market disruptions and bringing more investment into the infrastructure market.”
