Introduction: The Value Chain Overshadowed by GPUs
Everyone is counting GPU shipments, but few are counting electricity. JPMorgan's latest report applies first-principles reasoning to the complete power delivery chain of AI data centers. The core conclusion: the AI power semiconductor market will grow from approximately $2.7 billion in 2025 to $19.2 billion by 2028, representing a three-year compound annual growth rate of 82%. An even larger variable is the 800V high-voltage direct current (HVDC) architecture revolution, where silicon carbide solid-state transformers and gallium nitride converters replace traditional electromechanical equipment, driving semiconductor content per watt from $175 to $260. This suggests a value chain obscured by the GPU spotlight is now emerging.
Current Power Delivery Chain: Five Conversion Stages, Only 85-88% Efficiency
Today's data center power delivery is an inefficient long chain: grid 10-35kV AC is first stepped down to 400-480V via transformers, then fed into UPS, through PDU, then converted from AC to DC in server power supplies, and finally regulated to sub-volt levels for GPU cores via VRM. Five conversion stages, each losing 2-5%, result in end-to-end efficiency of only 85-88%. For a 100kW rack, 15kW becomes waste heat that must be removed by cooling systems. This severely limits the power density of AI clusters.
81 GW Capacity Additions by 2028, Driving a $19.2B Semiconductor Market
Based on internal AI server models, JPMorgan estimates that global AI data centers will add approximately 81 GW of capacity by 2028, comprising about 63 GW of new builds and 18 GW of replacements. AI chip power consumption accounts for roughly 54 GW, with network equipment and PUE factors bringing the total to the final figure. The power semiconductor market supporting this 81 GW capacity is projected to see per-watt semiconductor content rise from $175 to $260, pushing the total market to $19.2 billion.
800V HVDC Architecture: The Core Technological Shift
The report's most critical technological insight is the replacement of conventional AC architecture with 800V HVDC. The physics is simple: power equals voltage times current, and thermal losses are proportional to the square of the current. Raising voltage from 400V to 800V halves the current, reducing copper losses to one-quarter. But the real significance of the architecture shift lies in the qualitative change in semiconductor content. In conventional designs, many stages are electromechanical, with semiconductor concentration limited to PSU and VRM. The 800V architecture introduces four new nodes: SiC solid-state transformers replacing copper-wound transformers; SiC solid-state circuit breakers for microsecond fault isolation; native DC battery backup units with bidirectional DC-DC converters and BMS chips; and rack-level 800V-to-low-voltage DC-DC conversion.
Deployment Timeline: Solid-State Transformers Mature by 2028
JPMorgan provides a clear timeline: 2026-2027 will still be dominated by conventional 400V architecture, but retrofits have already begun, with sidecar power racks and power shelves appearing. From the second half of 2027 to 2028, NVIDIA's Kyber rack (600kW per rack) will drive volume deployment of native 800V solutions. After 2028, solid-state transformers mature, merging sidecar power racks and transformers into a single SST device. This makes 2028 a pivotal year for the 800V ecosystem.
Semiconductor Material Landscape: SiC, GaN, and Silicon Share the Market
The report quantifies the share evolution for different semiconductor materials. Silicon carbide (SiC) per-watt content rises from $30 to $60 long-term, dominating high-voltage stages from grid to rack. Gallium nitride (GaN) jumps from $3 to $46, winning the intermediate conversion stage from 800V to low voltage. Silicon increases modestly from $150 to $180, still holding the largest pool of VRM/load point, leveraging cost-performance. All three materials will coexist, but SiC and GaN will grow much faster than silicon.
Key Players: Infineon Leads Full Chain, MPS and Renesas Follow
The competitive landscape is taking shape. Infineon has the strongest full-chain layout; MPS is the VRM leader and a core supplier to NVIDIA; Renesas captures significant share in intermediate conversion and load points. NVIDIA has selected multiple suppliers among them. The report covers 12 core companies: Infineon, MPS, Renesas, TI, STMicroelectronics, Navitas (GaN leader), ADI, onsemi, Rohm, Innoscience, AOS, and Wolfspeed. Investors should watch these names closely.
Risk Factors: Grid Expansion Lag and NVIDIA's Pricing Power
JPMorgan's report excels at framework building rather than offering specific price targets. The $19.2 billion scale is modest within the overall AI infrastructure landscape, but the key insight is: without sufficient power semiconductors, no amount of GPUs can run. Two underdeveloped assumptions warrant attention. First, grid expansion delivery times (median 3-5 years in the US) conflict with the two-year data center build cycle, posing execution risk to the 81 GW forecast for 2028. Second, NVIDIA's pricing power across the entire value chain means its choice of power suppliers for the Kyber rack will directly reshape the competitive landscape. Additionally, JPMorgan has investment banking relationships with Infineon, STMicroelectronics, and other covered companies, which should be considered when evaluating specific stock recommendations.