Overview
Barclays has published a cross-sectional list of firms it considers essential to the expansion of artificial intelligence infrastructure, noting that technology sector estimates point to annual AI infrastructure spending from Western hyperscalers and AI labs potentially topping $1 trillion before peaking in 2028. The bank’s work covers more than 400 public and private companies operating across 19 digital and power infrastructure subcategories, and it highlights possible additional demand from sovereign AI programs and activity in China.
Why compute matters
At the center of Barclays’ roster are companies that supply the compute, accelerators, chips and packaging that train and run large AI models, plus firms that provide the networking and software glue for hyperscale deployments. Barclays singled out a group of market-leading names that it sees as particularly well-positioned to capture the bulk of near-term AI infrastructure spending.
Highlighted companies and roles
Nvidia - Identified as the AI bellwether, Nvidia maintains leadership in AI infrastructure through its integrated Blackwell and Rubin GPU architectures. Barclays notes that Nvidia’s chief executive recently said the company has visibility to more than $1 trillion in orders for its Blackwell and Rubin platforms spanning 2025 through 2027.
Microsoft - Microsoft is deploying proprietary Maia AI accelerators and Cobalt CPUs to enhance Azure’s AI infrastructure. The company has also made NVIDIA Nemotron models available on Microsoft Foundry and is rolling out the Vera Rubin NVL72 supercomputer into its data centers.
Alphabet - Google develops proprietary Tensor Processing Units for its internal model training and for Google Cloud customers. Recent product updates from Google include new Gemini assistant features such as a chat import tool and the Gemini 3.1 Flash Live audio model that developers can use.
Meta - Meta Platforms is building in-house accelerators, termed Meta Training and Inference Accelerators (MTIA), and the firm is reported to be in discussions with the Adani Group about potential data-center partnerships in India.
Amazon - Amazon Web Services creates purpose-built chips like Trainium and Inferentia to optimize price-performance for cloud customers. Barclays noted that AWS is reportedly developing AI to automate internal sales functions, and JPMorgan has observed elevated demand for AWS services.
AMD - Advanced Micro Devices supplies high-performance Instinct GPUs and EPYC CPUs used for both training and inference workloads in data centers. AMD also announced a collaboration with Celestica to produce the Helios rack-scale AI platform and struck a multi-year licensing agreement with Adeia.
Broadcom - The company provides custom accelerator solutions and XPU design services for hyperscale data centers. Broadcom and partner Carahsoft won a five-year, $970 million contract to deliver cloud software to the U.S. Defense Information Systems Agency, and Broadcom has begun volume shipments of its Tomahawk 6 switch chip.
Alibaba - In China, Alibaba develops proprietary GPUs for its cloud services to meet AI compute demand. The company recently introduced a next-generation AI chip called the XuanTie C950, built on RISC-V architecture and reported to be more than three times faster than its predecessor.
Arm - The chip designer provides foundational architecture used across many hyperscale custom silicon projects, powering designs such as Cobalt, Graviton and Axion, and supporting NVIDIA’s Grace-Blackwell clusters. Following an "Arm Everywhere" event, the company unveiled a new AGI CPU chip aimed at agentic AI and received multiple analyst upgrades and price target increases.
Taiwan Semiconductor Manufacturing - TSMC fabricates the most advanced AI chips and supplies crucial CoWoS advanced packaging capacity. The foundry reported that its revenue for the first two months of 2026 rose 29.9% versus the same period a year earlier.
Other names - Barclays also highlights Intel, Marvell, Qualcomm and Tencent as important compute-oriented participants in the digital and power infrastructure required for AI deployments.
Market implications and Barclays’ capex view
Barclays projects that current consensus estimates for hyperscaler capital expenditures could see upside of more than $300 billion as AI infrastructure needs ramp. The bank also expects that spending growth will eventually decelerate as recursive self-improvement in AI models reduces the marginal requirements for training over time.
Summary
Barclays’ mapping of suppliers across compute, chips, packaging and data-center hardware points to a concentrated group of technology companies that stand to benefit from multi-year AI infrastructure investment. The bank’s work emphasizes both the scale of potential spending - with Western hyperscalers and labs possibly exceeding $1 trillion annually ahead of a 2028 peak - and the range of specialized providers needed to build and operate those systems.
Context limitations
The bank’s assessment includes more than 400 entities across 19 subcategories and references potential additional support from sovereign AI programs and activity in China. Barclays also notes that longer-term reductions in training demand driven by model improvements could slow spending growth after the peak period.