Hyperscaler Capex to Exceed $710B in 2026
The top eight cloud service providers are projected to spend over $710 billion on capital expenditures in 2026, according to a new analysis from TrendForce. The report highlights that Google leads in custom ASIC deployment with its proprietary TPUs, underscoring a broader hyperscaler strategy to build custom silicon for core workloads to control costs and performance.
- The "Big Five" hyperscalers—Amazon, Microsoft, Google, Meta, and Oracle—are projected to spend over $600 billion on infrastructure in 2026, a 36% increase from 2025, with approximately 75% of that capital dedicated to AI-related infrastructure. This massive spending is creating a surge in demand for data centers, with worldwide capex expected to reach $1.7 trillion by 2030. - Microsoft's latest custom AI accelerator, the Maia 200, is built on TSMC's 3nm process and is designed for inference workloads. The company claims it offers three times the performance of Amazon's third-generation Trainium and surpasses Google's seventh-generation TPU at 8-bit precision, while being 30% more cost-efficient than previous Microsoft hardware. - Meta is developing its own custom silicon, the Meta Training and Inference Accelerator (MTIA), to reduce reliance on external suppliers like Nvidia and optimize for its own recommendation models. The second-generation MTIA, running at 1.35GHz with a 90W thermal envelope, aims to provide denser capabilities for workloads on platforms like Facebook and Instagram. - Amazon's custom silicon strategy includes the Trainium chip for AI training and the Inferentia chip for inference, which are designed to offer a significant cost reduction compared to competitor GPUs. AWS claims that its Trainium2 chip delivers comparable performance to NVIDIA's H100 GPUs at approximately 25% of the cost for some workloads. - Google's Tensor Processing Units (TPUs) are highly specialized for large-scale tensor operations, making them efficient for neural network training and inference, particularly for those heavily invested in the TensorFlow ecosystem. For large language model training, reports suggest that the cost-effectiveness of TPUs can be 4–10 times higher than GPUs. - The massive capital expenditure by hyperscalers is leading to a significant portion of their operating cash flow being reinvested into infrastructure, reducing the funds available for stock buybacks and dividends. In 2026, it's expected that capex will consume about 92% of hyperscalers' operating cash flows. - The intense investment in AI has led to a surge in venture capital funding for AI startups, with these companies attracting a third of all VC capital. In 2025, 49 U.S.-based AI startups raised over $100 million each, absorbing roughly 80% of the total venture capital flowing into the AI sector. - Enterprise spending on AI is also rapidly increasing, with Gartner forecasting that AI infrastructure software spending will reach nearly $230 billion in 2026. The total global AI infrastructure market is projected to reach $758 billion by 2029, according to IDC.
