NVIDIA's Feynman GPU generation pushes power semiconductor content to $191,000 per system, a 17× increase over the Blackwell architecture. The leap reflects the industry's shift to 800V DC power delivery, which demands more expensive silicon carbide (SiC) and gallium nitride (GaN) components.
Key facts
- Feynman power semiconductor content: $191,000 per system.
- 17× increase over Blackwell's ~$11,000 power semi content.
- 800V DC architecture requires SiC and GaN components.
- Google plans $190B AI infrastructure buildout.
- NVIDIA's China market share dropped to zero in 2026.
Wccftech reports that NVIDIA's upcoming Feynman GPU architecture will require $191,000 worth of power semiconductors per system, up from roughly $11,000 for Blackwell [According to Wccftech]. The 17× increase is not a GPU cost hike — it reflects the move to 800V DC architectures, which demand silicon carbide (SiC) and gallium nitride (GaN) power devices rather than traditional silicon MOSFETs.
Why 800V DC Drives the Cost
The 800V DC architecture shift is the primary driver behind the 17× cost increase. Higher voltage reduces current for the same power level, slashing I²R losses in datacenter distribution. But it requires power semiconductors with higher breakdown voltages — SiC and GaN components that cost 5–10× more per watt than silicon equivalents [Industry estimates]. Each Feynman rack likely contains dozens of 1,200V-class SiC MOSFETs and GaN HEMTs for DC-DC conversion and bus regulation.
The Unique Take
This cost increase is not inflationary — it's substitutional. The $191,000 figure represents power delivery, not compute. As hyperscalers like Google plan $190B AI infrastructure buildouts [According to TradingView], the 800V DC standard is becoming the default for next-generation datacenters. Feynman's power semi content is a leading indicator: the cost of moving electrons is now a first-order design constraint for AI clusters.
Industry Context
NVIDIA's China market share has dropped to zero due to US export controls [Per Nvidia's Jensen Huang, May 2026]. The company invested $2 billion in Marvell to deepen the NVLink Fusion partnership [Previously reported]. Feynman's power architecture suggests NVIDIA expects hyperscaler demand to absorb the higher system cost — each Feynman system at $191K in power semi content alone implies total system prices well above $1 million.
Google's $190B AI buildout, reported by TradingView, signals demand for the 800V DC systems that Feynman targets. The shift also creates opportunities for power semiconductor suppliers: Wolfspeed, Infineon, and ON Semiconductor are the primary SiC/GaN vendors positioned to supply the Feynman generation.
Wccftech notes the figure is based on supply chain estimates, and NVIDIA has not confirmed per-system BOM breakdowns.
What to watch
Watch for NVIDIA's official Feynman launch event, expected late 2026 or early 2027, which should confirm system pricing and power architecture details. Also track Wolfspeed and Infineon earnings for SiC order volumes that would validate the $191K figure.
