Hyperscalers pour into AI infra

Microsoft, Amazon Web Services, Alphabet and Meta are all raising spending on artificial intelligence infrastructure, turning 2026 into a data-center buildout race. (microsoft.com, aws.amazon.com, abc.xyz, investor.atmeta.com) Microsoft said on July 30, 2025 that it expected first-quarter capital expenditures to top $30 billion as it kept scaling artificial intelligence capacity. In its 2025 annual report, the company said it would keep investing in cloud and artificial intelligence infrastructure and training. (microsoft.com, microsoft.com) Amazon Web Services said Amazon planned to raise fiscal 2025 capital expenditures to $100 billion, with “the vast majority” going to new artificial intelligence data centers. Amazon Web Services is also expanding the physical backbone around those centers, with 123 Availability Zones across 39 Regions and nearly 20 million kilometers of fiber-optic cabling. (aws.amazon.com, aws.amazon.com) Alphabet said capital expenditures were $27.9 billion in the fourth quarter and $91.4 billion for full-year 2025. Chief financial officer Anat Ashkenazi said about 60% of that technical-infrastructure spend went to servers and 40% to data centers and networking equipment. (abc.xyz) Meta reported $72.22 billion in capital expenditures for full-year 2025, including principal payments on finance leases, after spending $22.14 billion in the fourth quarter alone. Chief executive Mark Zuckerberg said on January 28, 2026 that he was focused on “advancing personal superintelligence” in 2026. (investor.atmeta.com) The spending is not just for chips. Alphabet’s breakdown shows money moving into servers, buildings and networking gear, while Amazon Web Services says new designs now mix air and liquid cooling for processors, storage servers and network switches inside artificial intelligence clusters. (abc.xyz, aws.amazon.com) That shift lines up with Nvidia’s pitch that an artificial intelligence data center works like a factory, where processors do the computing and high-speed links move work between them. Nvidia says its Spectrum-X Ethernet platform is built to connect hyperscale artificial intelligence systems across hundreds of thousands of graphics processing units, not just a few racks. (nvidia.com, nvidia.com) Blackwell is central to that push. Nvidia said Blackwell data-center revenue grew 17% sequentially in the quarter ended July 27, 2025, and the company’s data-center revenue reached $62.3 billion in the quarter ended January 25, 2026, up 75% from a year earlier. (nvidianews.nvidia.com, nvidianews.nvidia.com) Nvidia’s GB200 NVL72 system shows why the bills are so large: one rack ties together 72 Blackwell graphics processing units and 36 Grace central processing units in a liquid-cooled system designed to act like one giant machine. Amazon Web Services says it is already designing cooling systems for rack-scale setups including Nvidia GB200 NVL72 and its own Trainium3 chips. (nvidia.com, aws.amazon.com) The near-term question is whether revenue rises fast enough to justify the buildout. Alphabet said Google Cloud revenue grew 48% and backlog jumped to $240 billion in its February 4, 2026 earnings call, while Microsoft said Azure revenue topped $75 billion for fiscal 2025. (abc.xyz, microsoft.com) For now, the biggest cloud companies are spending as if artificial intelligence demand will keep climbing, and the winners are spreading beyond the chip itself to power, cooling, fiber and networking. Nvidia is still at the center, but the checkbook is now reaching across the whole stack. (nvidia.com, aws.amazon.com, abc.xyz)