OnSemi forecast third-quarter revenue at $1.72 billion to $1.82 billion Monday, the midpoint $120 million above the Street's $1.57 billion consensus. The Phoenix-based chipmaker is betting on sustained demand for power management silicon in AI data centers, where hyperscalers are burning through infrastructure budgets faster than their own internal projections suggested six months ago.
The company manufactures silicon carbide MOSFETs and intelligent power modules that regulate voltage delivery to GPU clusters and manage thermal dissipation in high-density server racks. OnSemi's Q2 bookings showed 23% quarter-over-quarter growth from three hyperscale customers, none of whom the company named but whose capex schedules align with Microsoft, Meta, and AWS datacenter expansion timelines published in July earnings calls. The guidance implies OnSemi expects $410 million to $455 million in AI-related revenue for Q3, up from an estimated $340 million in Q2.
What matters here is timing and durability. The inference buildout wave—where trained models get deployed at scale—requires different power architecture than training infrastructure. OnSemi's power modules sit at the rack level, managing distribution regardless of whether the silicon is training GPT-5 or serving ChatGPT queries to enterprise customers. That positioning insulates the company from the training-to-inference transition risk that has pressured GPU-adjacent plays since April. Hyperscalers are now ordering power infrastructure on 18-month forward contracts instead of the traditional six-month cycles, a structural shift OnSemi's CFO noted without elaboration during the Monday earnings call.
The guidance also signals capital allocation priorities among the hyperscalers. If OnSemi is seeing sustained bookings growth, it means datacenter construction timelines are holding or accelerating despite the broader semiconductor inventory correction that has pressured analog chipmakers since late 2023. The company's lead times—currently 22 to 26 weeks for silicon carbide modules—suggest supply is tight enough that customers are willing to lock in pricing six quarters out, an unusual dynamic outside of memory markets.
Operators should track OnSemi's silicon carbide fab utilization rates in the November earnings call, expected around November 4th. The company is running its New Hampshire facility at 94% capacity as of Q2 and has flagged potential expansion capex decisions by year-end. Watch for hyperscaler capex guidance revisions in the next reporting cycle—Microsoft reports October 24th, Meta October 25th, AWS parent Amazon October 26th—to confirm whether infrastructure spend is genuinely decoupling from model development budgets. If hyperscaler capex holds or rises while GPU revenue growth moderates, it validates OnSemi's positioning in the power layer rather than the compute layer.
The forward contracts also create a twelve-month visibility window that OnSemi hasn't had since the automotive chip shortage of 2021. The company is now selling into a market where customers are willing to commit capital based on projected utilization rates, not spot demand.