Tech giants raise ai investment plans to $725 billion in 2026
Major technology companies are scaling up spending on artificial intelligence infrastructure to unprecedented levels, with combined capital expenditure plans nearing $725 billion for 2026. The surge follows quarterly earnings reports released on April 29 by Alphabet, Amazon, Meta and Microsoft, all of which confirmed or increased their commitments to AI-driven expansion.
Each company outlined aggressive investment targets tied to data centers, cloud capacity and advanced computing hardware. Meta raised its projected capital spending for 2026 to between $125 billion and $145 billion. Alphabet now expects annual investment to reach between $180 billion and $190 billion, with further increases planned for 2027. Microsoft disclosed $190 billion in capital expenditure for 2026, while Amazon reaffirmed plans to spend close to $200 billion. A significant share of this spending is directed toward Nvidia, whose graphics processing units remain central to global AI training and inference systems.
Nvidia’s Blackwell Ultra platform has become a key focus of this investment cycle. Introduced at GTC 2025 and expanded at GTC 2026, the system has been shipping since the second half of 2025 and is ramping up production. Its GB300 NVL72 rack integrates 72 Blackwell Ultra GPUs and 36 Grace CPUs, delivering up to 50 times higher throughput per megawatt and reducing cost per token by a factor of 35 compared with the previous Hopper generation, according to company data and independent benchmarks. For training workloads, Blackwell Ultra offers roughly 1.9 times the performance of standard Blackwell chips at scale.
Cloud providers are accelerating deployment timelines. AWS plans to install more than one million Nvidia GPUs across its global infrastructure starting this year, covering both Blackwell and the upcoming Rubin architecture. Google Cloud has confirmed it will be among the first to offer Nvidia’s Vera Rubin NVL72 rack systems in the second half of 2026. Nvidia’s chief executive Jensen Huang has projected that combined demand for Blackwell and Rubin systems could approach $1 trillion by 2027, doubling earlier estimates.
The next phase of Nvidia’s roadmap centers on the Vera Rubin platform, expected to launch later this year with a reported tenfold improvement in performance per watt compared with the Grace Blackwell generation. The company also previewed Kyber, a future architecture featuring 144 GPUs in vertically integrated compute racks, designed to underpin Rubin Ultra systems planned for 2027. Nvidia’s stock reached a record high earlier in the week, supported by large-scale deployment commitments, including plans by OpenAI to roll out more than 10 gigawatts of Nvidia-based systems, alongside the release of GPT-5.5.
-
11:45
-
11:30
-
11:16
-
11:15
-
11:00
-
10:59
-
10:58
-
10:45
-
10:39
-
10:30
-
10:22
-
10:15
-
10:05
-
10:00
-
09:45
-
09:41
-
09:38
-
09:30
-
09:15
-
09:11
-
09:00
-
08:55
-
08:45
-
08:43
-
08:30
-
08:15
-
08:01
-
08:00
-
07:45
-
07:39
-
07:30
-
07:21
-
07:15
-
07:05
-
07:04
-
16:45
-
16:33
-
16:30
-
16:15
-
16:12
-
16:02
-
16:00
-
15:46
-
15:45
-
15:32
-
15:30
-
15:15
-
15:14
-
15:00
-
14:49
-
14:45
-
14:36
-
14:30
-
14:20
-
14:15
-
14:00
-
13:56
-
13:45
-
13:40
-
13:30
-
13:20
-
13:15
-
13:03
-
13:01
-
12:45
-
12:30
-
12:15
-
12:00