Aug 4, 2026 earnings call — AI data center growth and manufacturing actions
AI data center revenue now expected to more than double in 2026; Q2 revenue was $1.6 billion.
Revenue $1.6B (est $1.6B) · EPS $0.74 (est $0.71) · Next-quarter revenue outlook $1.6B–$1.8B
Key quotes
"We now expect AI data center revenue to more than double in 2026, driven by stronger demand, further accelerated by broader customer wins and expanding content across the entire Power Tree."
"We expect our ESS revenue to grow approximately 40% year-over-year in 2026, driven by higher year-over-year growth in North America with microgrid customers."
"We expect approximately $35 million of annualized savings with the initial benefit starting in 2027 and the full savings realized in 2028."
Analyst Q&A 11
automotive demand and pricing
Q (Vivek Arya · BofA): Asked how the automotive business will perform in Q3 and typical Q4 seasonality, and whether pricing actions apply to automotive customers.
A: Management said automotive demand is stable and Q2 was seasonally down. They said pricing actions are being used across all markets, including automotive, to offset material input cost increases; gross margin effects from pricing were limited in Q2 and utilization gains will help margins later.
"So overall, we see automotive kind of, how do you say, very stable with a good outlook." — Hassane El-Khoury (CEO)
utilization impact and exits
Q (Timothy Arcuri · UBS): Asked if the right rule is 25–30 bps gross margin per point of utilization, and asked about the $35 million exits and the $300 million full-year impact.
A: Thad confirmed the math is about 25 to 30 basis points of gross margin improvement per point of utilization. He said the exits are done; $35 million was the Q2 impact and annualizing Q1 and Q2 gets you roughly about $300 million for the year.
"The math is 25 basis points to 30 basis points of gross margin improvement for every point of utilization." — Thad Trent (CFO)
prioritizing AI vs auto/industrial
Q (Joseph Quatrochi · Wells Fargo): Asked whether prioritizing AI data center over auto/industrial impacted revenue and asked for guidance by subsegment for 3Q.
A: Management said they prioritized AI data center when power technologies were constrained, shifting some orders from other businesses, and expect manufacturing to catch up in Q3 and Q4. They guided auto up low single digits, industrial relatively flat, and other (which includes AI data center) up high teens.
"As you look forward, we expect auto to be up low single digits, industrial to be relatively flat, and we expect other, which has our AI data center to be up high teens." — Thad Trent (CFO)
AI data center growth and SiC role
Q (Quinn Bolton · Needham & Company): Asked whether AI data center revenue will exceed the prior $500 million target and whether silicon carbide is the main growth driver.
A: Management would not give a new numeric breakout but said momentum in AI data center is strong and sustained. They said AI data center growth is not only high-voltage silicon carbide — revenue gains come across the whole power chain from high to low voltage.
"So the growth that we are seeing in AI data center is not necessarily anchored on the high voltage." — Hassane El-Khoury (CEO)
why AI and EV TAM rose
Q (Christopher Rolland · Susquehanna): Asked what drove the large increase in AI data center TAM and what drove the EV TAM increase.
A: Management said the AI TAM increase is tied to a projected gigawatt-scale install by 2030 and using content per rack to derive the market size. They also said their TAM rose because of new products and market opportunities (vertical GaN, Treo, silicon carbide JFETs) and greater silicon carbide penetration plus hybrid designs in EVs.
"And the projection we are basing on the increase is really correlated to a -- I'm going to call it, a gigawatt scale that would be installed by 2030." — Hassane El-Khoury (CEO)
lead times and flat utilization
Q (Blayne Curtis · Jefferies): Asked to confirm if AI data center was the primary driver of the 'other' growth, and to explain longer lead times and why utilization stayed flat.
A: Thad said AI data center was the primary driver but other markets also grew. He said lead times stretched from about 27 weeks to around 32 weeks on average, and utilization stayed flat because fab cycle times are 4 to 6 months so it takes time to ramp manufacturing to meet the higher demand.
"So lead times stretched out from about 27 weeks to somewhere around 32 weeks on average." — Thad Trent (CFO)
depreciation and distributor growth
Q (Thomas O'Malley · Barclays): Asked whether depreciation trends will change and what caused the distributor (mass market) step-up.
A: Thad said they took about 12% capacity offline which reduced depreciation and that depreciation is now steady state; CapEx this year is maintenance and below their 5% target. He said the mass market revenue increased 20% quarter-over-quarter, mostly industrial with some automotive exposure.
"So the mass market actually increased 20% quarter-over-quarter." — Thad Trent (CFO)
ramping capacity with partners; rack timing
Q (Tore Svanberg · Stifel): Asked how much flexibility they have with external partners to ramp more capacity, and whether the content per rack increase will be gradual or a step function (timing).
A: Management said external supply is constrained but they have good allocation and can support demand; the issue is how quickly demand arrived and wafer lead times. They still do over 60% internal manufacturing and said an 800-volt rack is a design step-function, but revenue will grow gradually as racks deploy with a ramp around end of '27 to early '28.
"We still do over 60% on internal manufacturing at onsemi even with the divestitures." — Hassane El-Khoury (CEO)
customer inventory and input costs
Q (James Schneider · Goldman Sachs): Asked whether OEM customers are building inventory now and what suppliers say about input cost trends into 2027 and pricing plans.
A: Hassane said they are not seeing an inventory build and are shipping to end demand; they remain disciplined and reduced channel inventory days. He said suppliers are not signaling cost decreases — 'it's not going down' — so they expect to continue pricing actions to offset input cost increases and do not expect a softer pricing environment in 2027.
"The only consistent message I can say here is it's not going down." — Hassane El-Khoury (CEO)
utilization level and capacity to take share
Q (Joseph Moore · Morgan Stanley): Asked how their low-80s utilization compares to competitors and whether they can take share if supply tightens; also asked where constraints exist.
A: Thad declined to comment on competitors but said fully utilized is the low 90s (92–93%) and they can flex capacity inside and outside so they are not close to being capped unless revenue rises 25–30% from today. Hassane said some lanes across high to low voltage are constrained, but they are increasing capacity, have available runway in invested areas (East Fishkill), and are winning where their product differentiates.
"For us, getting to kind of in that low 90% with 92%, 93% is fully utilized." — Thad Trent (CFO)
rack content split and geography trends
Q (Vijay Rakesh · Mizuho): Asked how content splits between high and medium/low voltage at different content levels, and how order trends differ by geography for auto, industrial, and data center.
A: Hassane said roughly for a $115,000 rack the split is 50% high voltage and 50% medium/low, while today on $15,000 of content the split is about 30% high voltage and 70% medium to low. He said order trends: automotive is China first, then U.S., then Europe; industrial is China, U.S., Europe; AI data center is primarily U.S. with some China.
"Today, on the $15,000 of content, that split between high voltage and then the rest is 70-30, 30 is high voltage, 70 is medium to low." — Hassane El-Khoury (CEO)