Goldman Sachs has increased its forecast for the global server market size, with AI server racks emerging as the primary growth engine.
In a research report published on June 24, Goldman Sachs' Allen Chang team revised upward the total global server market, projecting that global server market revenue will reach $1.1 trillion by 2028.
Within this, AI server rack revenue is expected to expand at a compound annual growth rate of 118% to $561.4 billion, accounting for 51% of total global server market revenue.
Goldman Sachs also raised its capital expenditure forecasts for major US and Chinese cloud service providers, anticipating that the leading US firms' capex will grow year-over-year by 76%, 35%, and 8% from 2026 to 2028, with the cumulative three-year total reaching $1.145 trillion by 2028.
Analysts at the firm noted that high memory costs and the sustained growth of AI workloads are the core factors driving this forecast revision.
Regarding AI chip architecture, custom ASICs are rapidly gaining market share. The proportion of ASICs in the total AI chip volume is expected to rise from 50% in 2026 to 55% in 2028, an increase from prior estimates.
Goldman Sachs maintains its positive outlook for the AI infrastructure investment cycle continuing through 2028 and has assigned buy ratings to several related stocks in the AI server supply chain.
AI Server Racks: Shipment Forecasts Significantly Raised, Foxconn Leads ODM Market
Goldman Sachs has raised its forecast for AI server rack (NVL72 equivalent) shipments to 55,000, 105,000, and 163,000 racks for 2026 through 2028, respectively. The forecasts for 2027 and 2028 represent increases of 16% and 20% over previous estimates.
Broken down by chip platform:
Shipments of Nvidia NVL72 racks are projected to reach 92,000 and 148,000 in 2027 and 2028, respectively, up approximately 19% to 22% from old forecasts.
AMD rack shipments remain largely unchanged, forecast at 5,000, 13,000, and 15,000 from 2026 to 2028.
With specifications continuing to upgrade, Goldman Sachs expects the dollar value per AI server rack to keep rising, driving the global total addressable market value for AI server racks to grow at a 118% CAGR between 2025 and 2028, reaching $561.4 billion by 2028.
On the supplier landscape, Goldman expects major global cloud service providers to continue dominating demand, with the ODM direct supply model remaining the mainstream source of supply.
Among these, Foxconn/Industrial FII is viewed favorably by Goldman as the largest ODM supplier in the AI server rack space, with its global market share expected to increase from 55% in 2026 to 69% by 2028.
Accelerating ASIC Penetration Drives Overall Server Shipment Growth
Goldman Sachs has raised its forecast for AI server (8-GPU equivalent) shipments by 18%, 24%, and 11% for 2026 through 2028, corresponding to shipments of 1.9 million, 2.4 million, and 2.56 million units. The primary driver for the upward revision is the rapid rise in demand for ASIC servers.
The analysis suggests that custom ASIC designs can provide more efficient solutions for AI computing, including lower latency and memory cost savings by reducing expensive read/write cycles. Major global cloud providers' continued rollout of next-generation ASIC chips supports this growth in AI servers.
At the total AI chip demand level, Goldman projects implied demand of 18.85 million, 26.75 million, and 32.21 million chips for 2026 through 2028.
Within this:
GPU demand is expected to increase from 9.51 million chips in 2026 to 14.57 million in 2028.
ASIC demand is projected to surge from 9.34 million to 17.65 million chips, with its share rising correspondingly from 50% to 55%.
Goldman Sachs forecasts the global market size for AI servers (8-GPU) to grow at a 26% CAGR between 2025 and 2028 to $295.5 billion, representing 27% of total global server market revenue in 2028.
The firm also raised its growth forecast for general server revenue, revising the expected growth rates for 2026-2028 from the previous 15%/6%/6% to 28%/17%/11%. The main drivers include high memory costs and specification upgrades driven by continuously evolving AI workloads.
General server global market size is projected to grow at an 18% CAGR from 2025 to 2028 to $245.2 billion, accounting for 22% of global server market revenue in 2028.
Cloud Provider Capex: US and Chinese Cloud Firms Accelerate Expansion, Memory Costs a Key Driver
On the client demand side, Goldman Sachs' US internet team projects that capital expenditures for leading US cloud service providers, including Microsoft, Amazon, Meta, Google, and Oracle, will grow year-over-year by 76%, 35%, and 8% from 2026 to 2028.
The three-year forecast is raised by 8%, 27%, and 25% from prior estimates, with the cumulative total reaching $1.145 trillion by 2028.
In the Chinese market, Goldman expects capital expenditures for leading Chinese cloud platforms, represented by ByteDance, Alibaba, Tencent, and Baidu, to grow year-over-year by 80%, 20%, and 18% from 2026 to 2028. The three-year forecast is raised by 37%, 44%, and 55% from previous estimates.
The analysis indicates that the strong growth in cloud provider capex primarily reflects high memory costs and the continued expansion of AI workload demand.
In terms of capex structure, chips (GPUs, ASICs, CPUs) remain the largest expenditure item, followed by memory (HBM), with networking and other components (such as cooling, power supplies, chassis, rail kits, etc.) following.
Compared to global cloud providers, Chinese cloud providers have a more diversified configuration structure between GPUs and ASICs in their chip spending, influenced by the diversification of supply from local chip manufacturers and differing requirements across AI application scenarios.
Comments