I infrastructure is approaching its second bottleneck. The first was securing enough GPUs; the next is delivering enough power to use them. As NVIDIA pushes toward 600-kilowatt and megawatt-scale systems, the industry is being forced beyond the traditional 54V rack and toward 800V DC. More than 80 suppliers have joined NVIDIA's ecosystem, but only a handful could capture enough content per rack to move earnings. Three U.S.-listed companies stand out—and the purest 800V play may not be the best stock.
Why 800 Volts Changes the Profit Pool
Today's AI racks receive AC power, convert it to 54V DC and step it down again inside the compute tray. That becomes unwieldy as NVIDIA moves toward 600-kilowatt and one-megawatt systems: delivering one megawatt at 54 volts implies roughly 18,500 amps, versus 1,250 amps at 800 volts. Higher current means more copper, heat and resistive loss. Raising voltage and eliminating conversions allows more of a scarce grid connection to reach the GPUs.
NVIDIA's power rack, expected in the second half of 2026, is designed to sit inside an existing AC facility and deliver 800V DC within the row. A larger system supporting up to two megawatts per row is expected in 2027. NVIDIA, Google and Microsoft have developed a common architecture through the Open Compute Project, giving suppliers a standard around which to qualify real products.
The transition redistributes the bill of materials toward power racks, solid-state transformers, energy storage, DC busways and new conversion stages. Silicon carbide handles high voltage and large loads; gallium nitride enables faster switching and greater density. The winners will be suppliers able to capture multiple sockets, qualify at scale and manufacture when deployments begin—not everyone carrying an NVIDIA partner badge.
The sequence matters. The first 800V products will not replace the entire electrical chain: NVIDIA's hybrid power rack keeps the building's AC infrastructure and changes power delivery within the row. That puts the earliest revenue opportunity in high-voltage conversion, backup power and rack-level DC-to-DC components. Facility-scale solid-state transformers represent a larger prize, but one that arrives later. Suppliers covering both the transitional architecture and the eventual native-DC design have a better chance of earning revenue before 800V reaches full adoption.
onsemi: The Auto Stock With an AI Power Business Hidden Inside
onsemi is still discussed primarily through automobiles and industrial demand, which contributed $781 million and $423 million of its $1.60 billion in second-quarter revenue. That exposure is a cyclical risk, but it obscures the fastest-growing part of the company. Management expects AI data-center revenue to more than double in 2026 and has secured platform wins with Great Wall while expanding its NVIDIA MGX role.
The advantage is not the NVIDIA relationship alone. onsemi supplies across the power tree—from controllers and low-voltage MOSFETs near the processor to SiC and GaN in 800V distribution and solid-state transformers. As the architecture removes some conversion stages and makes the survivors more valuable, onsemi can follow the profit pool instead of defending one legacy socket.
The company estimates its addressable content could rise from roughly $15,000 in a current 120-kilowatt rack to $115,000 in a megawatt-class rack by 2030. That nearly eightfold increase is a projection, not an order, but higher power does add SiC and GaN content. Its GaNEXUS portfolio spans 40V to 650V, while 700V and 1,200V vertical-GaN devices are sampling from Syracuse, New York, with volume targeted for late 2026. onsemi also has U.S. manufacturing across New York, Oregon, Idaho and Pennsylvania.
Second-quarter non-GAAP gross margin reached 39.3%, operating margin was 20.8% and free cash flow represented approximately 27% of revenue. Third-quarter guidance calls for $1.65–$1.75 billion of revenue and 40%–42% gross margin. AI is not yet large enough to drive the company alone, but it is arriving as utilization and margins recover, creating leverage unavailable to a pre-revenue pure play.
onsemi is no longer priced as a forgotten cyclical. At approximately $83 in mid-August, its equity value was around $32 billion, more than five times 2025 revenue. The planned $7 billion all-stock Synaptics acquisition will give Synaptics holders roughly 12% of the combined company and could distract management.
Still, onsemi offers the best asymmetry here. The market believes AI power will grow; it may not price a shift from one small revenue stream to content across the 0.8V-to-800V architecture. The thesis needs sustained AI growth, timely vertical-GaN production and larger platform wins. It fails if the $115,000 estimate remains a slide or Synaptics dilutes focus.
Navitas: The Purest 800V Option—and the Most Expensive Promise
Navitas Semiconductor is the name most likely to attract speculative capital. Its GaN and SiC portfolio is built specifically for the conversion stages that 800V requires. The company has demonstrated 800V-to-6V and 800V-to-50V boards, including a 10-kilowatt all-GaN platform targeting 98.5% peak efficiency, plus a 250-kilowatt solid-state transformer using high-voltage SiC. It is shipping production samples for several customer programs and expects selected hyperscaler and XPU platforms to ramp in 2027. A GlobalFoundries partnership is also intended to bring GaN manufacturing to the U.S., with initial availability targeted for late 2026.
That gives Navitas more percentage upside if its products become standard components: GaN near compute and SiC in higher-voltage power racks and grid infrastructure. Its $557 million cash balance provides time to fund qualifications and capacity.
But investors are already paying generously for revenue that has not arrived. Second-quarter sales were only $10.5 million, up 22% sequentially but still below the prior year, while the non-GAAP operating loss was $11.4 million. Third-quarter guidance calls for $13.5 million of revenue at the midpoint. Against a recent equity value above $3 billion, and even after subtracting cash, Navitas trades at roughly 60 times likely 2026 sales. The cash cushion came partly through capital raising, so survival risk has fallen at the cost of dilution.
Navitas is therefore an option on 800V adoption, not the value pick. The milestone that matters is not another demonstration at an NVIDIA event; it is conversion of production samples into named, repeatable programs during 2027. Until that happens, a small position can participate in the upside, but the valuation leaves little protection against delayed hyperscaler ramps or a competing GaN/SiC design.
Eaton: The Most Certain Winner, With the Least Hidden Upside
Eaton represents the other end of the spectrum. Its NVIDIA reference architecture already combines supercapacitors, DC busbars, connectors, containment and power distribution. The company can also supply medium-voltage solid-state transformers that convert grid power directly to high-voltage DC. That grid-to-chip breadth makes Eaton difficult to displace: it can sell the system around the semiconductor rather than compete for one device socket. New manufacturing investment in Virginia and transformer expansions in Texas and South Carolina strengthen its North American supply position.
The numbers confirm the demand, although they do not isolate 800V. Second-quarter sales reached $8.5 billion, organic growth was 14%, Electrical Americas orders increased 41% and total Electrical backlog grew 43% year over year. Management raised full-year adjusted EPS guidance to $13.40–$13.60. Yet around $455, the stock traded near 34 times the guidance midpoint and carried an equity value around $175 billion.
Eaton is the most dependable way to own the transition, but 800V must compete with transformers, aerospace and the rest of a $27 billion revenue base for financial relevance. The stock should work if electrical demand remains tight; it is less likely to rerate solely because one new data-center architecture succeeds.
The Best Exposure Is Not the Purest One
The 800V transition now has a deployment schedule. The power rack begins the bridge in the second half of 2026; U.S. GaN production should follow; row-level systems and hyperscaler programs become the 2027 proof points. Delays are likely, but megawatt racks cannot be powered efficiently with an architecture designed for tens of kilowatts.
Among these three, onsemi offers the strongest balance between evidence and optionality. It has real revenue, free cash flow, U.S. manufacturing and enough breadth to benefit even if the precise mix of SiC, GaN and silicon changes. Navitas offers the largest theoretical upside, but its valuation already assumes that engineering demonstrations become production programs. Eaton offers the greatest certainty, but that certainty is reflected in both its multiple and scale.
The preferred exposure is therefore onsemi, with Navitas reserved for investors willing to treat the position as a speculative option and Eaton serving as the quality benchmark. The critical question is no longer whether NVIDIA can assemble an 800V ecosystem. It is which suppliers turn that ecosystem into qualified content before the market discovers that the power architecture—not the GPU count—is setting the ceiling on the next AI factory.
Read full article here »
View the Zacks #1 Rank List
NVIDIA's Next AI Bottleneck Isn't GPUs—It's 800 Volts. These 3 Stocks Could Win
I infrastructure is approaching its second bottleneck. The first was securing enough GPUs; the next is delivering enough power to use them. As NVIDIA pushes toward 600-kilowatt and megawatt-scale systems, the industry is being forced beyond the traditional 54V rack and toward 800V DC. More than 80 suppliers have joined NVIDIA's ecosystem, but only a handful could capture enough content per rack to move earnings. Three U.S.-listed companies stand out—and the purest 800V play may not be the best stock.
Why 800 Volts Changes the Profit Pool
Today's AI racks receive AC power, convert it to 54V DC and step it down again inside the compute tray. That becomes unwieldy as NVIDIA moves toward 600-kilowatt and one-megawatt systems: delivering one megawatt at 54 volts implies roughly 18,500 amps, versus 1,250 amps at 800 volts. Higher current means more copper, heat and resistive loss. Raising voltage and eliminating conversions allows more of a scarce grid connection to reach the GPUs.
NVIDIA's power rack, expected in the second half of 2026, is designed to sit inside an existing AC facility and deliver 800V DC within the row. A larger system supporting up to two megawatts per row is expected in 2027. NVIDIA, Google and Microsoft have developed a common architecture through the Open Compute Project, giving suppliers a standard around which to qualify real products.
The transition redistributes the bill of materials toward power racks, solid-state transformers, energy storage, DC busways and new conversion stages. Silicon carbide handles high voltage and large loads; gallium nitride enables faster switching and greater density. The winners will be suppliers able to capture multiple sockets, qualify at scale and manufacture when deployments begin—not everyone carrying an NVIDIA partner badge.
The sequence matters. The first 800V products will not replace the entire electrical chain: NVIDIA's hybrid power rack keeps the building's AC infrastructure and changes power delivery within the row. That puts the earliest revenue opportunity in high-voltage conversion, backup power and rack-level DC-to-DC components. Facility-scale solid-state transformers represent a larger prize, but one that arrives later. Suppliers covering both the transitional architecture and the eventual native-DC design have a better chance of earning revenue before 800V reaches full adoption.
onsemi: The Auto Stock With an AI Power Business Hidden Inside
onsemi is still discussed primarily through automobiles and industrial demand, which contributed $781 million and $423 million of its $1.60 billion in second-quarter revenue. That exposure is a cyclical risk, but it obscures the fastest-growing part of the company. Management expects AI data-center revenue to more than double in 2026 and has secured platform wins with Great Wall while expanding its NVIDIA MGX role.
The advantage is not the NVIDIA relationship alone. onsemi supplies across the power tree—from controllers and low-voltage MOSFETs near the processor to SiC and GaN in 800V distribution and solid-state transformers. As the architecture removes some conversion stages and makes the survivors more valuable, onsemi can follow the profit pool instead of defending one legacy socket.
The company estimates its addressable content could rise from roughly $15,000 in a current 120-kilowatt rack to $115,000 in a megawatt-class rack by 2030. That nearly eightfold increase is a projection, not an order, but higher power does add SiC and GaN content. Its GaNEXUS portfolio spans 40V to 650V, while 700V and 1,200V vertical-GaN devices are sampling from Syracuse, New York, with volume targeted for late 2026. onsemi also has U.S. manufacturing across New York, Oregon, Idaho and Pennsylvania.
Second-quarter non-GAAP gross margin reached 39.3%, operating margin was 20.8% and free cash flow represented approximately 27% of revenue. Third-quarter guidance calls for $1.65–$1.75 billion of revenue and 40%–42% gross margin. AI is not yet large enough to drive the company alone, but it is arriving as utilization and margins recover, creating leverage unavailable to a pre-revenue pure play.
onsemi is no longer priced as a forgotten cyclical. At approximately $83 in mid-August, its equity value was around $32 billion, more than five times 2025 revenue. The planned $7 billion all-stock Synaptics acquisition will give Synaptics holders roughly 12% of the combined company and could distract management.
Still, onsemi offers the best asymmetry here. The market believes AI power will grow; it may not price a shift from one small revenue stream to content across the 0.8V-to-800V architecture. The thesis needs sustained AI growth, timely vertical-GaN production and larger platform wins. It fails if the $115,000 estimate remains a slide or Synaptics dilutes focus.
Navitas: The Purest 800V Option—and the Most Expensive Promise
Navitas Semiconductor is the name most likely to attract speculative capital. Its GaN and SiC portfolio is built specifically for the conversion stages that 800V requires. The company has demonstrated 800V-to-6V and 800V-to-50V boards, including a 10-kilowatt all-GaN platform targeting 98.5% peak efficiency, plus a 250-kilowatt solid-state transformer using high-voltage SiC. It is shipping production samples for several customer programs and expects selected hyperscaler and XPU platforms to ramp in 2027. A GlobalFoundries partnership is also intended to bring GaN manufacturing to the U.S., with initial availability targeted for late 2026.
That gives Navitas more percentage upside if its products become standard components: GaN near compute and SiC in higher-voltage power racks and grid infrastructure. Its $557 million cash balance provides time to fund qualifications and capacity.
But investors are already paying generously for revenue that has not arrived. Second-quarter sales were only $10.5 million, up 22% sequentially but still below the prior year, while the non-GAAP operating loss was $11.4 million. Third-quarter guidance calls for $13.5 million of revenue at the midpoint. Against a recent equity value above $3 billion, and even after subtracting cash, Navitas trades at roughly 60 times likely 2026 sales. The cash cushion came partly through capital raising, so survival risk has fallen at the cost of dilution.
Navitas is therefore an option on 800V adoption, not the value pick. The milestone that matters is not another demonstration at an NVIDIA event; it is conversion of production samples into named, repeatable programs during 2027. Until that happens, a small position can participate in the upside, but the valuation leaves little protection against delayed hyperscaler ramps or a competing GaN/SiC design.
Eaton: The Most Certain Winner, With the Least Hidden Upside
Eaton represents the other end of the spectrum. Its NVIDIA reference architecture already combines supercapacitors, DC busbars, connectors, containment and power distribution. The company can also supply medium-voltage solid-state transformers that convert grid power directly to high-voltage DC. That grid-to-chip breadth makes Eaton difficult to displace: it can sell the system around the semiconductor rather than compete for one device socket. New manufacturing investment in Virginia and transformer expansions in Texas and South Carolina strengthen its North American supply position.
The numbers confirm the demand, although they do not isolate 800V. Second-quarter sales reached $8.5 billion, organic growth was 14%, Electrical Americas orders increased 41% and total Electrical backlog grew 43% year over year. Management raised full-year adjusted EPS guidance to $13.40–$13.60. Yet around $455, the stock traded near 34 times the guidance midpoint and carried an equity value around $175 billion.
Eaton is the most dependable way to own the transition, but 800V must compete with transformers, aerospace and the rest of a $27 billion revenue base for financial relevance. The stock should work if electrical demand remains tight; it is less likely to rerate solely because one new data-center architecture succeeds.
The Best Exposure Is Not the Purest One
The 800V transition now has a deployment schedule. The power rack begins the bridge in the second half of 2026; U.S. GaN production should follow; row-level systems and hyperscaler programs become the 2027 proof points. Delays are likely, but megawatt racks cannot be powered efficiently with an architecture designed for tens of kilowatts.
Among these three, onsemi offers the strongest balance between evidence and optionality. It has real revenue, free cash flow, U.S. manufacturing and enough breadth to benefit even if the precise mix of SiC, GaN and silicon changes. Navitas offers the largest theoretical upside, but its valuation already assumes that engineering demonstrations become production programs. Eaton offers the greatest certainty, but that certainty is reflected in both its multiple and scale.
The preferred exposure is therefore onsemi, with Navitas reserved for investors willing to treat the position as a speculative option and Eaton serving as the quality benchmark. The critical question is no longer whether NVIDIA can assemble an 800V ecosystem. It is which suppliers turn that ecosystem into qualified content before the market discovers that the power architecture—not the GPU count—is setting the ceiling on the next AI factory.
Read full article here »