Skip to content

Dossier · POWI · Dormant

POWI

Last analysed · · source: watchlist_research

Current thesis

Legacy power-IC name re-rated ~3x ($31→$85) on AI-data-center power optionality; 2026-06-01 NVIDIA Kyber 800VDC GaN reference designs are the latest leg. Theme ACCELERATING but POWI itself is extended at 52-wk highs, above every analyst PT ($69.50), with next hard catalyst (Q2 print) ~2 months out — chase risk high here.

Invalidation trigger

Daily close below ~$74 (post-May breakout shelf / rising 50-DMA); OR GaN-power peers (NVTS, MPWR, VICR) rolling over together (cluster break); OR Q2 print ~2026-08-06 shows no AI data-center design-win/bookings traction to justify ~58x fwd P/E.

Thesis status

Open commitment scored if the trigger above fires How this is scored →

Current Thesis

Legacy high-voltage power-IC name (chargers, appliances, industrial) re-rated roughly 3x off its $30.86 52-week low to ~$84–85 (2026-06-04) on an AI-data-center power optionality story. The latest leg is the 2026-06-01 launch of ultra-slim auxiliary PSU reference designs for NVIDIA's Kyber 800VDC AI racks. The 800V data-center power transition is a genuinely ACCELERATING theme, but POWI itself is extended at the top of its 52-week range, trades ABOVE every analyst PT ($69.50 avg), and has no scheduled catalyst until the ~Aug-6 Q2 print. We'd be chasing strength after the catalyst already fired — probe-size only, not a fat pitch here.

Bull Case

  • 2026-06-01: Unveiled two ultra-slim auxiliary PSU reference designs (15W at 30×30×7mm; 35W six-rail at 80×60×8mm) for NVIDIA Kyber 800VDC liquid-cooled racks, built on InnoMux-2 + 1700V PowiGaN; claims ~30% PDB space saving and ~30% BOM-count reduction. Direct association with NVIDIA's flagship AI rack architecture.
  • Structural fit: The 1700V-rated InnoMux-2 supports 1000VDC nominal flyback input at ~90% flat efficiency — one of few parts positioned for the multi-year 800VDC/HVDC data-center power shift (NVIDIA Rubin/Kyber roadmap).
  • Q2 2026 guide $117.5M midpoint = +8.5% QoQ vs Q1's $108.3M and +2.2% above consensus (Q1 call ~2026-05-06); GAAP gross margin guide 53.5–54.5%, op margin recovering to 13.5–15.5% as restructuring normalizes.
  • Industrial +23% YoY in Q1 2026 — the cyclical recovery engine is already turning before any AI revenue shows up; inventory levels improving.
  • Trend fully intact: tripled off the $30.86 low, sitting just under the $89.00 52-week high — momentum structure is not broken.

Bear Case

  • Valuation divorced from fundamentals: 285x trailing P/E, ~58x forward (2026-06-04); analyst avg PT $69.50 is ~18% BELOW spot — price has run past even bullish sell-side.
  • Core is flat: Q1 2026 revenue +2.6% YoY ($108.3M) and a slight consensus miss. The 3x move is a re-rating on story, not earnings growth.
  • AI contribution is reference-design stage (announced 2026-06-01), NOT booked revenue or named design wins. The data-center TAM is narrative, not yet in the P&L.
  • Catalyst vacuum: the June-1 news already fired; the next hard catalyst is the Q2 print ~2026-08-06 — ~2 months of nothing scheduled into a stretched chart.
  • Not sole-source: Navitas (NVTS) is the higher-profile NVIDIA 800V GaN partner; MPWR, Vicor (VICR), Infineon, onsemi all chase the same data-center power sockets.

Setup & Price Structure

  • 2026-06-04: ~$84–85, market cap ~$4.7B, 52-week range $30.86–$89.00 — pinned at the top of the range after a parabolic run.
  • Uptrend intact but extended; price trades above all analyst PTs. This is a buy-strength-after-catalyst entry, not a pullback entry.
  • MA/RSI feeds came back stale/conflicting — do NOT trust the $54.77 EMA / sub-40 RSI prints; they lag the run-up. Treat the ~$74–78 May breakout shelf as first structural support and $89.00 as overhead resistance. Assume RSI is elevated given proximity to highs.
  • Cleanest R/R is a pullback to the rising 50-DMA holding with the theme still accelerating, OR an actual AI design-win/bookings disclosure — not a chase at $85.

Catalyst Calendar (next 30 days)

  • None hard/scheduled through ~2026-07-04. The 2026-06-01 NVIDIA Kyber reference-design news has already been released.
  • Watch (unscheduled): additional NVIDIA Rubin/Kyber ecosystem design-win announcements; investor-conference appearances; any 800VDC supply-chain headlines that move the GaN-power cohort.
  • Next HARD catalyst (outside window): Q2 2026 earnings ~2026-08-06 (est., per prior-year 2025-08-06 print) — the binary for whether the AI narrative converts to numbers.

What Would Change Our Mind

  • Upgrade to HIGH: a named AI data-center design win or bookings figure (not a reference design) — re-rating becomes earnings-justified.
  • Better entry / size up: clean pullback to a rising 50-DMA (~$74–78) that holds while the theme stays ACCELERATING.
  • Exit / SKIP: daily close below ~[entry redacted] breaks the breakout shelf; OR the GaN-power cohort (NVTS, MPWR, VICR) rolls over together (cluster break); OR the ~Aug-6 Q2 print shows no AI data-center traction → re-rating deflates toward the $69.50 PT.
  • Theme flip: mainstream "AI power" saturation (CNBC headlines, retail piling into the NVTS/POWI/VICR basket) → mark theme SATURATED, trim.

Correlation Notes

  • Moves with the AI-data-center-power cohort: NVTS (closest comp — GaN, NVIDIA 800V), MPWR, VICR, onsemi (ON), Infineon (IFNNY).
  • 2nd-order to NVIDIA (NVDA) data-center capex and the Rubin/Kyber 800VDC roadmap — POWI is a beta-play on the power-architecture transition, not a primary AI compute name.
  • Also carries industrial-cyclical beta (Industrial +23% YoY) — recovery-correlated, making it less pure-AI than NVTS.
  • Basket-crowding risk: a single NVIDIA roadmap delay or 800V timeline slip hits the entire power-GaN cohort simultaneously.