POWI · Power Integrations, Inc. · Stock research

Current Thesis

Power Integrations makes high-voltage power ICs for chargers, appliances, industrial gear and automotive. Between April and June 2026 it re-rated roughly 2.7x off a $30.86 low to $91.18 on a single story: NVIDIA's Kyber rack architecture moves AI data-center power delivery to 800VDC, and POWI's 1250V/1700V PowiGaN is one of the few volume-production parts that fits that socket.

That leg has broken. The stock closed $70.45 on 2026-07-17 after a -20.21% week, cutting clean through the $74–78 May breakout shelf that held the June-5 sector test and produced the June high. It now trades below the $76.20 average analyst target after months above it. The cohort broke in the same window: Navitas fell from above $26 in mid-June to the mid-teens, -41% in a month, on a Wolfspeed patent suit and a $500M ATM. The June read's secondary invalidation — the GaN complex rolling over together — happened.

What remains is a company guiding to real sequential growth with a narrative premium that the tape is actively repricing. The 2026-08-05 Q2 print is where reference designs either become disclosed data-center dollars or stay a story.

Bullish and bearish views on Power Integrations, Inc.

The model's bull view on Power Integrations, Inc. (POWI), in brief: Q2 guide is genuinely accelerating (2026-05-07): $115–120M revenue, ~+8.5% sequential off Q1's $108.3M, with non-GAAP gross margin 54–55% and operating margin 13.5–15.5%. The bear view: The breakout failed. The $74 level named as the structural line broke decisively. A shelf that gets cut through on a -20% week is not support anymore; it becomes overhead supply. Cluster break, with company-specific damage in the cohort. Wolfspeed's Delaware suit against Navitas… Both cases follow in full.

Bull Case

• Q2 guide is genuinely accelerating (2026-05-07): $115–120M revenue, ~+8.5% sequential off Q1's $108.3M, with non-GAAP gross margin 54–55% and operating margin 13.5–15.5%. Growth across industrial, data center and automotive.
• Cyclical trough is turning in the reported numbers: Q1 2026 revenue $108.3M beat the $106.7M consensus, with industrial +23% YoY on renewables, battery storage, home automation and automotive.
• Product position is real, not vapor (2026-06-01): DER-1110 (35W six-rail) and DER-1114 (15W single-rail) auxiliary PSU reference designs for Kyber 800VDC liquid-cooled racks, built on 1700V PowiGaN single-HEMT InnoMux-2 — roughly 30% PDB space saving and 30% fewer BOM components.
• Volume-production claim (2025-10-13): POWI states it is the only supplier of 1250V and 1700V GaN switches in volume production, with more than 175M GaN switches deployed and >90.3% 12V system efficiency in a fan-less 800VDC architecture.
• Street is still net-constructive: Stifel $95 (2026-06-24), Needham initiated at $90, Benchmark raised. Spot at $70.45 sits ~8% below the $76.20 blended target.
• The July drawdown was macro-led: rising Treasury yields, a soft ADP print and hawkish Fed signalling hit the whole chip complex. No company-specific negative news accompanied the decline.

Bear Case

• The breakout failed. The $74 level named as the structural line broke decisively. A shelf that gets cut through on a -20% week is not support anymore; it becomes overhead supply.
• Cluster break, with company-specific damage in the cohort. Wolfspeed's Delaware suit against Navitas covers GaNFast, GaNSlim, GaNSafe, GeneSiC MOSFETs and SiCPAK modules. NVTS also announced a $500M ATM on 2026-07-02. IP risk and dilution are now live themes across GaN, and NVTS reports 2026-07-27 — ahead of POWI.
• The scarcity argument is eroding. ON Semiconductor launched GaNEXUS in June 2026 aimed squarely at AI data centers, industrial automation and energy infrastructure. A large balance sheet entering the socket compresses the "only volume supplier" premium.
• Fundamentals do not support the multiple. ~236x trailing P/E on TTM revenue of $443.5M (+5.86% YoY) with earnings -31.46% YoY. Q1 revenue grew +2.6% YoY. The re-rate was multiple expansion.
• First sell-side crack: Northland cut to Market Perform from Outperform on consumer exposure. When price falls below the average target, target cuts tend to follow rather than lead.
• AI revenue is still undisclosed. Reference boards and partner-ceremony appearances, no named design win, no bookings figure. Nothing yet quantifies the data-center TAM in a filing.
• Insider sales clustered in late May 2026, into the parabolic leg.

Setup & Price Structure

Spot $70.45 (2026-07-17 close, $72.00 after hours), against a 52-week range of $30.86–$91.18. The name is ~23% off the June high and gave back a third of the year's gain in about two weeks; YTD sits near +79% versus roughly +110% in late June.

The structure is a failed breakout, not a pullback to support. The $74–78 shelf that defined the May base and absorbed the June-5 sector selloff is now broken and overhead. The next reference is the July swing low near $60, which coincides with the pre-breakout base built before the April–May advance. Between $60 and $74 there is no meaningful prior consolidation to lean on — the stock moved through that range quickly on the way up, which means it can move through it quickly on the way down.

Momentum and moving-average feeds on this name have returned stale values in past pulls (sub-$55 EMA prints against a $70+ tape); those numbers should not be operationalized. The usable structure is $60, the void $74–78 shelf, and $91.18.

On the beginner-trap matrix: this is no longer stretched above its averages and no longer sits at peak retail enthusiasm — the mania has already deflated. The relevant trap here is the opposite one. A name that has fallen 23% from a high, into a broken shelf, with a binary print 13 trading days out, is precisely the setup that invites buying weakness on the argument that the story is unchanged. It is not a base until it makes a higher low.

Catalyst Calendar (next 30 days)

• Ongoing through August — Fed commentary and Treasury yield path. The July decline was macro-driven; the same channel can reverse it.
• Undated, live — Wolfspeed v. Navitas docket developments. Any ruling or expansion of GaN IP claims re-prices the whole cohort.

Elapsed catalysts

• 2026-07-27 — Navitas (NVTS) Q2 2026 results. The nearest read on GaN power demand, 800VDC design-win timing and pricing pressure. This print sets the tone for the complex before POWI reports. (passed 13d ago)
• 2026-08-05 (after close, call 1:30pm PT) — Power Integrations Q2 2026 results. Confirmed. Guide was $115–120M with 54–55% gross margin. The specific things that matter: any named AI data-center customer, any quantified data-center revenue or bookings, and the Q3 guide. (passed 4d ago)

What Would Change Our Mind

The bearish read on the current setup gets overturned by structure repairing itself, not by the narrative sounding good. Concretely: a higher low above $60 followed by a weekly close back above $78 would reclaim the broken shelf and re-establish the trend. Absent that, strength into the print is a bounce inside a broken structure.

On the fundamental side, the 2026-08-05 call converting the Kyber reference designs into a named customer or a disclosed data-center revenue line would justify re-underwriting the premium, because the entire bull case currently rests on inference. A Q3 guide meaningfully above the $115–120M Q2 range would do similar work.

The read gets worse from here on a weekly close below $60, on NVTS guiding down on 2026-07-27, or on further target cuts following Northland's downgrade.

Correlation Notes

POWI now trades as a high-beta expression of the GaN/AI-power complex rather than as a power-IC cyclical. The tightest comp is NVTS (Navitas), the higher-profile NVIDIA 800V GaN partner — its -41% month dragged the cohort's multiple down regardless of POWI's own news flow, and its 2026-07-27 print is the leading indicator. ON Semiconductor matters now as a competitor rather than a peer following the GaNEXUS launch. MPWR and VICR round out the power-delivery cohort; joint lows across the group confirm a theme-level de-rate rather than single-name weakness.

Second-order linkage runs through NVIDIA's rack roadmap: Kyber and MGX timing slippage would hit every 800VDC supplier at once. The name also carries residual consumer-electronics exposure — the basis of Northland's downgrade — which means it is not a pure AI-infrastructure vehicle and can be pressured by handset and appliance weakness independent of the data-center story.

Index mechanics add noise: the June 2026 Russell 2000 Dynamic Index addition brought one-time flow that has now passed. Broad semiconductor beta (SOX) and the rates complex explain most of the July move; the theme is cooling from ACCELERATING toward MATURING, and a failed Q2 print would push it further.