• Global server shipments are forecast to grow at a 15% compound annual growth rate (CAGR) through 2028, raised from the prior tracker .
• GPU AI server shipments were upgraded to a 22% CAGR over the same period .
• 8-GPU equivalent servers are expected to grow 48% in 2026 alone .
• Total AI rack shipments are projected to reach 61,000 units in 2026 and 88,000 units in 2027 .

New hardware catalysts are on the horizon: Nvidia's Rubin rack and AMD's Helios rack are both expected to begin shipping in Q4 2026 .

Goldman Sachs released a separate forecast in June 2026 projecting the global server market reaching $1.1 trillion by 2028, with AI server rack revenue growing at a 118% CAGR to $561.4 billion, or 51% of total market revenue .

Data Center & Cloud Capex: A $1.2 Trillion Pipeline

Hyperscale capital expenditure is the engine behind these server shipment forecasts, and the numbers are staggering:

• Total planned and under-construction data center investment: Bernstein estimates this has reached approximately $1.2 trillion .
• Four major cloud providers (Microsoft, Amazon, Google, Oracle) are on track for $623 billion in capex in 2026, with their combined backlog exceeding $2 trillion — an increase of 176% year-over-year .
• Five largest hyperscalers: Expected to grow capex from $438 billion in 2025 to more than $1 trillion in 2027, with a potential path toward $1.6 trillion by 2030, according to Bernstein's own midyear outlook .
• Post-Q2 2026 earnings, consensus 2026 capex estimates for major cloud service providers were raised by nearly 15% vs. March 2026, with total capex modeled to grow at a 61% CAGR to approximately $1.11 trillion by 2027 .

Supporting data from other sources paints a consistent picture:

• Goldman Sachs (July 2026) sees Big Tech combined capex reaching $750 billion by end of 2026 and nearing $1.2 trillion in 2027 .
• Morgan Stanley (Q1 2026 analysis) found the five largest hyperscalers collectively raised 2026 capex guidance in the first half of 2026, projecting combined capex of approximately $800 billion in 2026 and $1.2 trillion in 2027 .
• Bank of America (late 2025) saw global hyperscale spending rising 67% in 2025 and another 31% in 2026, with total outlays climbing to $611 billion .
• Gartner forecasts worldwide AI spending (full stack) will total $2.59 trillion in 2026, up 47% year-over-year .

Custom ASICs Take Center Stage

Perhaps the most significant structural shift documented in the tracker is the rapid rise of custom application-specific integrated circuits (ASICs) for AI workloads. Hyperscalers are increasingly deploying their own silicon to reduce dependence on merchant GPUs and optimize for specific inference and training tasks .

Broadcom: The Dominant Force

Broadcom has emerged as the leading AI ASIC design partner. Key developments include:

• Long-term agreement with Google: Broadcom signed a deal to co-develop next-generation TPUs . Broadcom already designs Google's TPU and OpenAI's custom chip .
• Market share: Broadcom commands roughly 60% market share among AI ASIC design partners .
• Financial performance: Broadcom's AI semiconductor revenue hit $10.8 billion in Q2 FY2026, up 143% year-over-year, with the company reiterating a target of $100 billion in AI semiconductor revenue for FY2027 . Q2 AI semiconductor bookings exceeded $30 billion .

Marvell: Ramping with All Major Cloud Providers

Marvell is now ramping ASIC production for all four major cloud service providers (AWS, Microsoft, Google, and others) . Marvell designs AWS Trainium and Microsoft Maia . The company projects up to ~$11 billion in AI ASIC revenue in 2026 .

Together, Broadcom and Marvell control roughly 95% of the custom AI ASIC design market .

Market Projections

• ASICs share of CoWoS-based AI chip shipments: Bernstein expects ASICs to account for approximately 45% of total shipments, with TPU shipments growing about 90% year-over-year in 2026 .
• TrendForce / Counterpoint Research: AI server compute ASIC shipments are projected to triple between 2024 and 2027, with Broadcom retaining a 60% share . ASIC-based AI server shipments are expected to reach 27.8% of the market in 2026, the highest share since 2023 .
• Broadcom's FY2025 AI revenue was approximately $12 billion, and the company reported $8.4 billion in AI revenue in Q1 FY2026 alone, up 106% year-over-year .

The Big Picture: No Sign of Slowing Down

The overarching theme from Bernstein's Q2 2026 tracker is that AI infrastructure investment remains on a steep upward trajectory, driven by both GPU-based systems (Nvidia and AMD) and a rapidly scaling custom ASIC ecosystem led by Broadcom and Marvell, with no sign of a near-term capex pullback among the largest cloud providers. While merchant GPUs still dominate — TrendForce projects GPUs will account for 69.7% of AI server shipments in 2026 — the ASIC share is growing at a faster clip and represents a fundamental shift in how hyperscalers are building out their AI compute capacity.

• Global server shipments are forecast to grow at a 15% compound annual growth rate (CAGR) through 2028, raised from the prior tracker .
• GPU AI server shipments were upgraded to a 22% CAGR over the same period .
• 8-GPU equivalent servers are expected to grow 48% in 2026 alone .
• Total AI rack shipments are projected to reach 61,000 units in 2026 and 88,000 units in 2027 .

New hardware catalysts are on the horizon: Nvidia's Rubin rack and AMD's Helios rack are both expected to begin shipping in Q4 2026 .

Goldman Sachs released a separate forecast in June 2026 projecting the global server market reaching $1.1 trillion by 2028, with AI server rack revenue growing at a 118% CAGR to $561.4 billion, or 51% of total market revenue .

Data Center & Cloud Capex: A $1.2 Trillion Pipeline

Hyperscale capital expenditure is the engine behind these server shipment forecasts, and the numbers are staggering:

• Total planned and under-construction data center investment: Bernstein estimates this has reached approximately $1.2 trillion .
• Four major cloud providers (Microsoft, Amazon, Google, Oracle) are on track for $623 billion in capex in 2026, with their combined backlog exceeding $2 trillion — an increase of 176% year-over-year .
• Five largest hyperscalers: Expected to grow capex from $438 billion in 2025 to more than $1 trillion in 2027, with a potential path toward $1.6 trillion by 2030, according to Bernstein's own midyear outlook .
• Post-Q2 2026 earnings, consensus 2026 capex estimates for major cloud service providers were raised by nearly 15% vs. March 2026, with total capex modeled to grow at a 61% CAGR to approximately $1.11 trillion by 2027 .

Supporting data from other sources paints a consistent picture:

• Goldman Sachs (July 2026) sees Big Tech combined capex reaching $750 billion by end of 2026 and nearing $1.2 trillion in 2027 .
• Morgan Stanley (Q1 2026 analysis) found the five largest hyperscalers collectively raised 2026 capex guidance in the first half of 2026, projecting combined capex of approximately $800 billion in 2026 and $1.2 trillion in 2027 .
• Bank of America (late 2025) saw global hyperscale spending rising 67% in 2025 and another 31% in 2026, with total outlays climbing to $611 billion .
• Gartner forecasts worldwide AI spending (full stack) will total $2.59 trillion in 2026, up 47% year-over-year .

Custom ASICs Take Center Stage

Perhaps the most significant structural shift documented in the tracker is the rapid rise of custom application-specific integrated circuits (ASICs) for AI workloads. Hyperscalers are increasingly deploying their own silicon to reduce dependence on merchant GPUs and optimize for specific inference and training tasks .

Broadcom: The Dominant Force

Broadcom has emerged as the leading AI ASIC design partner. Key developments include:

• Long-term agreement with Google: Broadcom signed a deal to co-develop next-generation TPUs . Broadcom already designs Google's TPU and OpenAI's custom chip .
• Market share: Broadcom commands roughly 60% market share among AI ASIC design partners .
• Financial performance: Broadcom's AI semiconductor revenue hit $10.8 billion in Q2 FY2026, up 143% year-over-year, with the company reiterating a target of $100 billion in AI semiconductor revenue for FY2027 . Q2 AI semiconductor bookings exceeded $30 billion .

Marvell: Ramping with All Major Cloud Providers

Marvell is now ramping ASIC production for all four major cloud service providers (AWS, Microsoft, Google, and others) . Marvell designs AWS Trainium and Microsoft Maia . The company projects up to ~$11 billion in AI ASIC revenue in 2026 .

Together, Broadcom and Marvell control roughly 95% of the custom AI ASIC design market .

Market Projections

• ASICs share of CoWoS-based AI chip shipments: Bernstein expects ASICs to account for approximately 45% of total shipments, with TPU shipments growing about 90% year-over-year in 2026 .
• TrendForce / Counterpoint Research: AI server compute ASIC shipments are projected to triple between 2024 and 2027, with Broadcom retaining a 60% share . ASIC-based AI server shipments are expected to reach 27.8% of the market in 2026, the highest share since 2023 .
• Broadcom's FY2025 AI revenue was approximately $12 billion, and the company reported $8.4 billion in AI revenue in Q1 FY2026 alone, up 106% year-over-year .

The Big Picture: No Sign of Slowing Down

The overarching theme from Bernstein's Q2 2026 tracker is that AI infrastructure investment remains on a steep upward trajectory, driven by both GPU-based systems (Nvidia and AMD) and a rapidly scaling custom ASIC ecosystem led by Broadcom and Marvell, with no sign of a near-term capex pullback among the largest cloud providers. While merchant GPUs still dominate — TrendForce projects GPUs will account for 69.7% of AI server shipments in 2026 — the ASIC share is growing at a faster clip and represents a fundamental shift in how hyperscalers are building out their AI compute capacity.