Bernstein Lifts Server Forecast as AI-Driven Capital Spending Accelerates

Bernstein has raised its outlook for the global server market, citing accelerating capital spending on artificial intelligence infrastructure by major technology firms. The investment firm now projects the global server market will exceed $1 trillion by 2028, according to its second quarter 2026 AI tracker report released Monday.

Upgraded Growth Projections

The report details a significant upward revision in shipment forecasts, driven by the voracious appetite for computing power. Bernstein's analysts now anticipate strong, sustained growth across several key segments through 2028.

• Global server shipments: Forecasted to see a compound annual growth rate (CAGR) of 15%.
• GPU AI server shipments: Expected to grow even faster, with a projected CAGR of 22%.
• 8-GPU equivalent servers: Shipments are anticipated to surge 48% this year alone.

Bernstein also projects total server rack shipments will reach 61,000 units in 2026 and grow to 88,000 units in 2027. This growth aligns with new product cycles, as both Nvidia’s Rubin rack and AMD’s Helios rack are scheduled to begin shipping in the fourth quarter of 2026.

The Trillion-Dollar Investment Wave

Ad

Underpinning these bullish forecasts is an unprecedented level of investment in data center construction and equipment. Total planned and in-progress data center investment currently stands at approximately $1.2 trillion, according to the report.

Following second-quarter earnings, market estimates for 2026 capital expenditure (capex) by major cloud service providers jumped nearly 15% from March projections. Total capex from this group is now expected to hit roughly $1.1 trillion by 2027. Specific corporate initiatives highlighted by Bernstein include Meta's expanded $50 billion Hyperion project and SoftBank Group's plan to invest up to $85 billion in data center capacity in France.

Impact on the Semiconductor Sector

The demand for servers translates directly to a robust outlook for semiconductor companies. Bernstein noted that custom chips, or ASICs, are projected to account for approximately 45% of total CoWoS-based AI chip shipments. Google's custom Tensor Processing Units (TPUs) are a key part of this trend, with shipments expected to grow roughly 90% year-over-year in 2026.

This trend benefits a range of chip designers. Broadcom has secured a long-term deal with Google to develop future TPU generations, while Marvell is reportedly ramping up its ASIC production for all four major cloud providers, according to Bernstein.