onsemi's guided-up AI data-center demand is shifting the “power bottleneck” from GPUs to SiC and power-silicon

earnings

The “AI bottleneck” is broadening: guided-up AI demand now reads like a server-power buildout story

The non-obvious angle in onsemi’s latest Q2 update is that the AI opportunity is no longer only about compute (GPUs/HBM). Management is explicitly connecting AI data-center momentum to the company’s power-silicon and silicon-carbide exposure—i.e., the parts of the rack that turn electrical input into stable, efficient power rails.

The market often treats these as adjacent beneficiaries. But if power conversion, protection, and high-voltage switching scale in step with AI racks, then power semiconductors can move with AI capex before the memory/compute complex fully shows it.

Q2 2026 revenue (reported)

$1.604B

per onsemi Q2 2026 results release

AI data-center revenue outlook

more than doubles

per Q2 2026 management guidance language in the filed Exhibit 99.1

Q2 2026 gross margin (reported)

38.4%

GAAP gross margin per Q2 2026 results release

event verification

What onsemi actually guided (and what it proves)

A key part of the thesis must be anchored in primary-source language, not paraphrase. In the Q2 2026 results materials filed for onsemi, management reported strong revenue performance and included forward-looking guidance tied to the AI data-center end market.

What matters for investors is the direction and shape of the guidance: it frames AI data-center as a fast-scaling end demand rather than a slow catch-up. When that framing is paired with onsemi’s product focus (power conversion/protection + SiC devices), it supports the interpretation that AI buildouts increasingly pull through power-silicon and SiC supply.

Load-bearing facts collected this session

Q2 2026 revenue

$1.604B

Q2 2026 results release at onsemi IR (Exhibit/press materials)

Q2 2026 gross margin (GAAP)

38.4%

Q2 2026 results release

AI data-center revenue growth (outlook)

more than doubles

Exhibit 99.1 forward-looking language

For power-semiconductor investors, the key is that the guide ties AI data center strength to onsemi’s addressable power products—meaning AI rack additions can translate into discrete SiC/power-silicon demand before you see a clean “AI-only” revenue line in every quarter.

supply chain mechanism

Why AI compute demand can “leak” into power-silicon and SiC: a rack-level power equation

At the system level, AI buildouts increase (1) total power draw, (2) the need for fast transient response and efficient conversion, and (3) the reliability/thermal constraints on server power supplies.

That changes the bill of materials around the compute complex. Even if the GPU/HBM vendors capture the highest attention, the electrical front end—DC-DC conversion, power switching, regulation, and protection—still scales with the number of populated accelerators and the power delivery design they require.

Hyperscaler AI rack scale increases the volume of power conversion stages that must meet efficiency and stability targets, pulling through discrete power devices.
Higher voltages and switching losses create stronger incentives for next-gen power technologies, where SiC’s efficiency/conversion advantage becomes more attractive at the margin.
Server-power reliability requirements can increase demand for protection/robust switching devices even when compute mix changes quarter to quarter.

onsemi’s framing supports a model where AI capex expands the “power conversion” spend bucket, not just the compute/HBM spend bucket.

fundamental sanity check

Is onsemi’s earnings power consistent with an AI-led mix shift, or is it temporary?

onsemi’s quarterly revenue trend around the latest quarter

FMP-derived quarterly revenue line items used only as a directional check; the AI linkage is from onsemi’s Q2 guidance language.

Unit: USD

2025 Q2

quarterly revenue

1,468,700,000

2025 Q3

quarterly revenue

1,550,900,000

2025 Q4

quarterly revenue

1,530,100,000

2026 Q1

quarterly revenue

1,513,300,000

The directional check matters because if revenue is choppy but margins improve, it can hint at a mix tailwind. On the fundamentals side from data tools, onsemi has recently shown revenue at roughly the ~$1.5B/quarter scale with a reported Q2 2026 gross margin of 38.4% in the results release.

This doesn’t prove AI mix is the sole driver of every margin move. But it reduces the chance that the guided AI data-center strength is purely a one-quarter accounting bump.

Q2 2026 gross margin (GAAP)

38.4%

Q2 2026 results release

Q1 2026 revenue

$1.513B

FMP quarterly line item

investor angles

Five investment angles to test after this onsemi AI data-center guide

Validate AI data-center demand via continued guide-up language: watch whether AI data-center revenue growth stays “more than doubles” in subsequent quarters (the investor question is persistence, not the first guide).
Decompose mix pressure on gross margin: if SiC/power-silicon share rises, gross margin should remain resilient versus broader discrete price competition.
Track working-capital and cash conversion: AI buildouts can create both inventory risk and favorable cash dynamics if shipments track burn rates.
Cross-check capacity constraints: if SiC supply or wafer/device capacity limits become binding, near-term shipments may lag and can cap upside—even with strong AI demand.
Map competitive re-rating: if investors shift valuation models toward “power-per-rack share,” peer multiples for pure compute/memory may decouple from power-silicon beneficiaries.

horizons

What changes in the next quarter vs. over 1–3 years

Catalyst map built from what onsemi actually guided plus what power-silicon/SiC demand transmission implies.
Horizon	What moves first	Investor tell	What would falsify it
Days–quarters	Shipment/booking momentum for AI power-related devices	Management reiterates strong AI data-center end-market wording while margins hold	AI guide normalizes quickly or gross margin compresses despite revenue strength
1–3 years	Design-win durability across server generations and more varied power architectures	AI rack power scaling sustains a higher portion of revenue from power silicon/SiC versus cyclical peers	SiC/power-silicon demand growth lags hyperscaler capex growth due to adoption timing

Related listed stocks this thesis can realistically touch

OonsemiON--

--Vol --

Bullish

AI data-center guide-up supports sustained demand pull-through into power devices, so onsemi can outgrow the broader semiconductor cycle if the AI end-market persists.
Reported Q2 2026 GAAP gross margin of 38.4% supports a base case that mix tailwinds are not immediately reversing.

ABroadcomAVGO--

--Vol --

Mixed

If AI rack buildouts expand server infrastructure spend, Broadcom’s custom silicon and networking exposure can benefit, but power-silicon share may dilute pure “compute AI” optics in the near term.
Broadcom can be a beneficiary, yet the relative rotation effect depends on whether AI capex shifts toward power conversion components faster than it shifts toward networking.

MMicron TechnologyMU--

--Vol --

Mixed

AI demand is still memory-positive, but an adjacent power-silicon pickup implies not all AI optimism gets monetized first in HBM/DRAM—timing can decouple across the supply chain.
If hyperscalers prioritize power/efficiency upgrades alongside memory, Micron’s near-term pacing can face mix-driven volatility.

WWolfspeedWOLF--

--Vol --

Watch

SiC plays a direct role in the AI rack power equation; if onsemi’s guided-up SiC exposure broadens, Wolfspeed can catch up on demand expectations, so AI power builds can strengthen SiC unit demand over 1–3 years.
The falsifier is operational: if SiC capacity constraints or customer qualification delays dominate, Wolfspeed may not monetize the AI pull-through immediately.