Cloud giants are on track to spend more than $710 billion on AI infrastructure in 2026

A new TrendForce report shows the world's eight largest cloud providers increasing capital expenditure by 61% year on year, and shifting away from off-the-shelf chips toward custom silicon designed in-house.

The eight largest cloud service providers in the world, Google, Amazon Web Services, Meta, Microsoft, Oracle, Tencent, Alibaba, and Baidu, are collectively on track to spend more than $710 billion on infrastructure this year, according to a new report from TrendForce, a market research firm specialising in technology supply chains. The projected figure represents a 61% increase over 2025, driven largely by surging demand for AI computing capacity.

The scale of that spending reflects a fundamental shift in how these companies are building their data centres. While all eight continue to purchase graphics processing units from NVIDIA and AMD, the dominant hardware for training and running AI models, a growing number are moving toward custom chips designed for specific workloads. These application-specific integrated circuits, or ASICs, allow companies to handle AI tasks more efficiently and at lower cost than general-purpose processors.

Google is the furthest along that path. TrendForce projects Alphabet, Google's parent company, will spend more than $178 billion this year, a 95% increase year on year, the largest absolute figure among the eight providers. The company's Tensor Processing Units, its proprietary AI chips, are expected to account for nearly 78% of all AI servers shipped to Google in 2026. That would make it the only major cloud provider where in-house chips outnumber GPU-based systems.

Amazon, which operates AWS, the world's largest cloud infrastructure business , is continuing to scale its GPU deployments using NVIDIA's latest rack-scale systems, while also advancing its own Trainium chip line. The third generation of Trainium is expected to begin shipping in the second quarter of this year, with volume potentially accelerating in the second half as software development catches up with the hardware.

Meta, the company behind Facebook, Instagram, and WhatsApp, is forecast to spend more than $124 billion in 2026, up 77% from last year. Its AI infrastructure remains heavily reliant on NVIDIA and AMD GPUs, which are expected to account for more than 80% of its AI server build-out. The company is developing its own ASIC platform, called MTIA, though TrendForce notes that software and hardware integration challenges may slow adoption relative to initial targets.

Microsoft continues to invest in NVIDIA's rack-scale systems while introducing its own Maia 200 chip for AI inference, the process of running a trained model to generate outputs. Oracle is expanding its GPU deployments in support of large-scale AI data centre projects, including its involvement in the Stargate initiative alongside OpenAI.

Among the Chinese providers, ByteDance, the parent company of TikTok,  is expected to direct more than half of its 2026 capital expenditure to AI chip procurement. Tencent is sourcing NVIDIA GPUs for its cloud and generative AI services while also working with domestic chipmakers on ASIC alternatives. Alibaba and Baidu are both advancing proprietary chip development: Alibaba through its T-head semiconductor unit and Alibaba Cloud, and Baidu with its next-generation Kunlun chips, which are planned for release after 2026. All four Chinese companies are navigating ongoing US export controls on advanced semiconductors, which continue to shape their procurement strategies.

TrendForce concludes that the divergence in chip strategies, between those relying on third-party GPU suppliers and those building custom silicon, is becoming one of the defining competitive factors among cloud providers, with implications for cost, performance, and long-term positioning in the AI market.