onsemi AI Data Center Revenue Targets $2.5 Billion, but Execution Will Decide the Margin Story
onsemi expects its AI data center revenue to exceed $2.5 billion annually by 2030, starting from more than $500 million in 2026. That fivefold expansion sits inside a broader plan for faster growth, higher margins, and greater cash generation. The conflict is clear: the company must turn a developing power architecture into profitable, repeatable semiconductor sales.
The opportunity does not center on processors or memory. onsemi wants to supply the power-conversion, protection, control, and sensing components surrounding increasingly energy-hungry AI accelerators. It argues that higher rack power will move data centers toward high-voltage distribution, where its automotive and industrial expertise becomes more relevant.
That strategy puts onsemi against established power-semiconductor suppliers while exposing it to a second opponent: its own execution history. The company is again targeting a 53% non-GAAP gross margin after setting the same percentage as a 2027 objective in 2023. Investors must decide whether AI demand changes the trajectory or merely extends the timetable.
The onsemi AI Data Center Revenue Target Starts With Power
onsemi is betting that power delivery will capture a larger share of every AI infrastructure budget.
At its September 16, 2026 Investor Day, onsemi presented a plan covering AI infrastructure, automotive systems, industrial equipment, and emerging physical AI applications. Its central financial forecast calls for company revenue to grow at a 12% to 14% compound annual rate through 2030.
The company expects the AI data center operation to grow considerably faster. Management projects more than 100% growth in 2026, followed by another doubling in 2027. It then expects annual growth near 50% through 2030.
That path takes the business from more than $500 million in 2026 to over $2.5 billion in annual revenue by 2030. The forecast represents approximately fivefold growth in four years. Management says that rate would run about ten percentage points above its assumed market growth.
These figures remain company forecasts, not completed results. The distinction matters because onsemi is combining several uncertain variables. AI rack deployment, electrical architecture, design wins, manufacturing output, and customer qualification must all advance on schedule.
The broader ambition is equally large. onsemi estimates a $213 billion addressable market across automotive, industrial, AI data centers, and emerging applications by 2030. Its growth strategy treats those categories as connected markets with overlapping power and sensing requirements.
That connection is important to the business case. A high-voltage device developed for an electric vehicle can inform products used in energy storage or data center power conversion. Shared technology can spread research spending across more customers and applications.
Management also wants to sell more than isolated components. It is moving toward platforms and system-level solutions that address larger portions of a customer's power architecture. Capturing more content per system would support both revenue growth and better margins.
The strategy therefore depends on two expansions occurring together. AI infrastructure must consume more power-semiconductor content, while onsemi must win a larger portion of that content. General growth in AI spending will not automatically produce the forecast.
The company's estimate starts with more than $500 million of 2026 AI data center revenue. Management said it already has visibility into another doubling during 2027. That near-term statement gives investors a measurable checkpoint before the longer-range assumptions dominate the model.
What changed at Investor Day was not simply the publication of another large market estimate. onsemi connected a specific revenue target to a specific electrical transition. It also tied that transition directly to its margin plan.
Megawatt Racks Move the Market Toward onsemi's Strengths
The revenue thesis depends on AI racks shifting from low-voltage equipment toward connected, high-voltage power systems.
AI accelerators require large amounts of electricity, but the challenge extends beyond total consumption. Operators must deliver that electricity within limited space while controlling conversion losses, heat, reliability, and infrastructure requirements.
Power density measures how much useful power a system can deliver within a given physical area. As rack power increases, small efficiency losses create more heat and waste more electricity. Those losses also reduce the energy available for computing.
Traditional data centers repeatedly convert electricity while moving it from the grid toward processors. Each conversion needs equipment, occupies space, and loses some energy. More direct high-voltage distribution can remove conversion stages and improve overall efficiency.
The transition will occur in phases rather than through one immediate replacement. Near-term systems can add high-voltage power shelves or sidecars beside compute racks. Later designs can distribute high-voltage direct current deeper into the facility.
onsemi expects 800-volt direct-current distribution to reach compute trays, with even higher voltages possible in later architectures. Direct current, or DC, moves electrical charge in one direction rather than alternating direction.
The company also expects solid-state transformers and circuit breakers to replace some mechanical or conventional equipment. A solid-state device controls electricity through semiconductors, enabling faster switching and more compact designs.
These changes expand the number of relevant semiconductor functions. Power conversion remains important, but protection, energy storage, telemetry, sensing, and system control also gain value. The opportunity stretches from the grid connection to the processor core.
According to onsemi's power architecture, semiconductor content normalized per AI rack could rise from roughly $15,000 to more than $115,000. That nearly eightfold estimate comes from the company and has not been independently validated.
Still, the underlying mechanism is understandable. Higher voltages require components that can switch, convert, and protect electricity safely. Higher rack density also increases the value of components that reduce heat and wasted space.
onsemi sells silicon, silicon carbide, and gallium nitride power devices. Silicon carbide, commonly shortened to SiC, handles high voltages and temperatures efficiently. Gallium nitride, or GaN, supports fast switching and compact power designs.
The company also highlighted its Embedded Power Platform, or EPP, which integrates power functions into a compact architecture. Management says EPP and its vertical GaN technology can improve rack-level power density by as much as 20%.
That percentage should be treated as a company claim until customers validate it in deployed systems. Even so, it shows how onsemi wants to compete. The pitch focuses on total system efficiency, not the price or specifications of one transistor.
Management says AI infrastructure is moving onto the company's established playing field. Automotive and industrial customers have long required efficient, reliable high-voltage devices. AI data centers are beginning to demand similar capabilities at a different scale.
The transfer is not automatic. Data center customers have their own reliability testing, architecture decisions, and qualification schedules. Technologies proven in vehicles or industrial equipment still need acceptance from hyperscalers, server manufacturers, and power-system suppliers.
Timing creates another complication. Customers will not adopt the same architecture simultaneously. Some will retain conventional designs, while others deploy sidecars or 800 VDC systems at different speeds.
That staggered transition can produce uneven revenue. It may also delay the most valuable grid-to-core configurations beyond the planning assumptions behind the 2030 target.
Yet the phased rollout gives onsemi several entry points. It can sell components into current power supplies, near-term high-voltage sidecars, energy storage, and later direct-current systems. Revenue does not depend entirely on one final architecture arriving at once.
The Real Opponent Is onsemi's Margin Execution
The $2.5 billion target matters because onsemi expects AI products to improve its business mix, not merely add sales.
The company's 2030 plan calls for non-GAAP gross margin to rise from an estimated 40.3% in 2026 to 53%. It also targets a 38% operating margin and a free-cash-flow margin between 30% and 35%.
Free cash flow is the cash remaining after operating expenses and capital investment. onsemi expects that figure to more than double and reach approximately $3.5 billion annually by 2030. Capital spending is projected to remain near 5% of revenue.
Management expects gross profit dollars to grow around 22% annually through 2030. Operating income dollars are expected to increase about 30% per year, more than twice the projected rate of revenue growth.
Those targets require operating leverage, which means profits grow faster than sales as factories and operating costs become more productive. Higher utilization spreads fixed manufacturing costs across more units. A richer product mix can add another lift.
The company divided its planned gross-margin improvement into several contributors. Higher factory utilization represents the largest portion, while manufacturing optimization and new products provide the rest.
New products include Treo, EPP, and vertical GaN. The plan also assumes a mix shift within silicon power products and growing AI data center revenue. In other words, AI is one part of the margin bridge rather than its only support.
The financial model follows five years of portfolio and manufacturing changes. During that period, onsemi says it exited about $900 million of non-core revenue.
The company reports that non-GAAP gross margin rose from roughly 26% to above 40% across those five years. It also says operating margin nearly tripled and free cash flow more than doubled.
That record gives management a credible argument that portfolio changes can affect profitability. However, it does not settle whether the next 13 percentage points of gross-margin expansion will arrive by 2030.
The 53% objective is familiar. At its 2023 analyst event, onsemi targeted the same gross margin for 2027. The earlier plan connected margin growth to silicon carbide expansion, new products, and manufacturing optimization.
Holding the target while extending the horizon changes how investors should read the latest forecast. The percentage remains ambitious, but it is not a newly raised destination. The burden has shifted toward proving the path.
AI demand can strengthen that path in three ways. It can raise factory utilization, create demand for differentiated high-voltage products, and increase the semiconductor content sold into each system.
However, the benefits can arrive at different times. Factory utilization can improve before new platforms reach scale. Design wins can precede meaningful revenue by several quarters. Revenue growth can also occur before the best product mix becomes dominant.
The company must manage all three layers without creating excess capacity. If it prepares for demand that arrives late, underutilized factories can pressure margins. If it invests too cautiously, supply constraints can limit its ability to capture the opportunity.
That tension explains why the onsemi AI data center revenue target cannot be assessed in isolation. Revenue quality matters as much as revenue quantity. Investors need evidence that the sales carry the margins assumed in the financial model.
The best outcome would combine fast-growing AI sales with higher-value products and better manufacturing absorption. The weaker outcome would produce revenue growth through competitive components while margin gains remain delayed.
Competition Shrinks With Voltage, but It Does Not Disappear
onsemi's broad portfolio creates an opening, although customers can still divide the power tree among multiple suppliers.
Power delivery begins at the grid and ends at processor voltage levels below one volt. No single semiconductor material performs every conversion equally well. Suppliers must combine high-voltage devices, low-voltage control, packaging, protection, and software-supported system knowledge.
Competition is broadest near the processor core, according to onsemi. Management said numerous companies advertise solutions there, although it believes only a smaller group has the scale and engineering capabilities required by major customers.
The field narrows as designs move toward higher-voltage SiC and GaN components. It narrows further when grid-side equipment, energy storage, protection, and low-voltage control must operate as one coordinated system.
During the Investor Day transcript, onsemi claimed only one other company could compete across the entire grid-to-core power tree. It did not identify that competitor in the cited remarks.
That claim illustrates the company's preferred competitive frame. Instead of comparing individual power devices, management wants customers to evaluate architecture breadth and co-design capability.
Co-design means optimizing connected components together instead of improving each part independently. The approach becomes more valuable when a fault or voltage change can propagate across an interconnected DC system.
Customers still have reasons to maintain multiple suppliers. Diversification protects supply, supports negotiation, and lets engineers select specialized components for each conversion stage. A broad portfolio does not guarantee exclusive system ownership.
Established semiconductor manufacturers also have deep expertise in power devices, analog control, and industrial systems. Some possess strong customer relationships or leadership in particular materials and voltage ranges.
Equipment companies add another layer. Data center electrical designs involve power-system vendors, server manufacturers, cooling specialists, utilities, and hyperscalers. Their architectural choices determine where semiconductor value accumulates.
onsemi therefore needs more than technically competitive components. It needs customer designs to progress from evaluation into qualified production. It must also secure meaningful content across those systems.
The company disclosed a major processor-core power socket win with revenue expected to begin by the end of 2026. That is a useful near-term signal, but management did not publicly identify the customer in the cited presentation.
A socket win means a supplier's component has been selected for a defined position within a system design. It indicates technical acceptance, although shipment volume still depends on the customer's production ramp.
onsemi also described demand for power shelves, sidecars, energy storage, solid-state protection, and future grid-side equipment. Those applications create multiple ways to participate without controlling the entire architecture.
This breadth reduces dependence on one component category. It also makes the revenue forecast harder to evaluate externally because different products follow different qualification and deployment schedules.
The competitive question is therefore not whether onsemi participates in AI infrastructure. Its existing revenue indicates that it already does. The question is how much share and value it can capture as high-voltage systems spread.
If the market remains fragmented, onsemi can still grow with total AI infrastructure spending. However, reaching 50% annual growth may require share gains beyond simple market expansion.
Management explicitly assumes those gains. Its baseline uses approximately 40% market growth and adds at least ten percentage points for onsemi. That assumption makes competitive execution central to the $2.5 billion outcome.
What the 2030 Targets Do Not Prove
The forecast combines credible technical pressure with adoption, market, and manufacturing risks that remain unresolved.
The strongest part of onsemi's argument is the physical constraint. Denser AI systems need more power, and every conversion loss becomes more costly as total consumption rises. Efficient delivery clearly has economic value.
The more uncertain part is how quickly customers standardize around the architectures onsemi describes. Sidecars, 800 VDC distribution, solid-state transformers, and vertical power delivery will not reach production at the same time.
The company expects sidecar architecture to deploy within roughly 12 to 18 months. It places more extensive grid-to-rack DC distribution further out. Slippage in those schedules would shift revenue toward conventional products.
A second risk concerns the size of the AI infrastructure market. Management acknowledged that third-party growth estimates vary widely. Its baseline assumes approximately 40% compound annual growth through 2030.
If the market expands more slowly, onsemi would need larger share gains to preserve its 50% growth target. If demand grows faster, capacity and supply execution become more important.
A third risk is product validation. Claims about smaller size, lower temperatures, and higher power density need confirmation in customer systems. Laboratory performance does not always translate directly into large-scale data center operation.
Reliability standards are particularly demanding for power infrastructure. A processor failure can remove compute capacity, but a power-system fault can affect a wider portion of a facility. Customers will test new technologies carefully.
The transition to connected high-voltage DC systems also introduces design complexity. Protection, fault isolation, grounding, thermal management, and control must work across the architecture. That requirement supports onsemi's co-design argument while slowing qualification.
Manufacturing creates a fourth risk. The gross-margin plan assumes factories become more fully utilized and more efficient. It also assumes the company allocates capacity toward differentiated, higher-value products.
These assumptions can conflict during volatile demand cycles. A company may increase utilization by accepting lower-value volume, but that volume may not deliver the desired mix. Alternatively, strict product selection can leave capacity unused.
The company also needs to scale several new platforms while serving automotive and industrial customers. Shared technology creates leverage, yet overlapping demand can produce capacity-allocation decisions.
Investor reaction has reflected this uncertainty. After attending the event, Stifel said it was encouraged by the strategy but considered the 2030 model potentially below expectations. Its analyst also said the margin path remained unclear.
That response captures the tension inside the announcement. The company presented a large AI target, but investors were already evaluating the assumptions required to reach it. A distant destination matters less without visible intermediate milestones.
The 53% gross-margin target deserves particular caution. onsemi has maintained that objective while moving the relevant time horizon from 2027 to 2030. The new plan needs quarterly evidence, not just another long-range model.
None of these risks invalidate the opportunity. They define the proof required. Revenue must arrive alongside product mix, utilization, and cash generation rather than through volume alone.
Three Signals Will Test the $2.5 Billion Forecast
The next evidence should come from 2027 revenue visibility, customer adoption, and measurable margin progression.
The first signal is whether the AI data center business doubles again in 2027. Management says it has visibility into that outcome after expecting more than 100% growth during 2026.
A completed doubling would support the early slope of the fivefold forecast. It would also indicate that current design wins are converting into shipments before the broadest 800 VDC transition.
A weaker result would not eliminate the 2030 opportunity, but it would force more growth into later years. That would increase dependence on architectures that remain under development or early deployment.
Investors should also examine what drives the growth. Revenue from existing power supplies carries different implications than sales tied to sidecars, solid-state protection, or grid-to-core systems.
The second signal is verified customer adoption of high-voltage architectures. Named production programs, platform qualifications, and shipment schedules would carry more weight than total addressable market estimates.
Particular attention should go to 800 VDC deployments and the disclosed processor-core socket win. Commercial shipments beginning by the end of 2026 would show that onsemi can compete at both high and low voltages.
Customer evidence would also help evaluate management's claim that architectural breadth creates an advantage. Multiple wins across the power tree would support that thesis more strongly than one isolated component award.
The third signal is the progression from an estimated 40.3% non-GAAP gross margin in 2026 toward 53% in 2030. That movement should accompany rising utilization and a greater contribution from differentiated products.
Margin improvement without AI growth could still reflect manufacturing recovery. AI growth without margin improvement might indicate competitive pricing, delayed product mix, or continued factory inefficiency.
The most persuasive quarters will show both trends together. Revenue should rise while gross margin, operating income, and free cash flow expand faster. That combination would validate the company's operating-leverage thesis.
Watch capital spending as well. onsemi expects it to remain around 5% of revenue through the expansion. A sustained increase could indicate that the growth plan needs more manufacturing investment than management currently assumes.
The onsemi AI data center revenue forecast is ultimately a claim about where value moves when computing becomes constrained by electricity. The company argues that power components will capture more value as racks become denser and voltages increase.
That judgment is credible enough to deserve attention, but it is not proven by the Investor Day targets. The decisive evidence will arrive through shipments, customer architectures, and financial results.
Readers evaluating the forecast should ask three questions each quarter. Did AI data center sales follow the promised trajectory? Did customers adopt more high-voltage onsemi content? Did those sales improve margins and cash generation?
If all three answers become consistently positive, the $2.5 billion target will look like an operating plan rather than an aspiration. If one remains missing, the gap between the opportunity and its financial payoff will stay open.
