NVTS — Navitas Semiconductor

Executive verdict

Navitas is what happens when a real thesis collides with a story-stock chart. The company is a legitimate pure-play in gallium nitride power semiconductors, the Infineon GaN cross-license signed in mid-2025 is genuine validation, and Irrational Analysis is correct that high-voltage GaN in 800V DC datacenter architectures is a real differentiated market. None of that is in dispute.

What is in dispute is what you pay for it. At $22.32 the stock trades at 123× TTM EV/sales on $40.5M of revenue that has collapsed 45% from the FY2024 peak, with consensus modeling another 39% decline in the next twelve months, while the company burns $44M of free cash a year. Eight analysts cover the name, the consensus rating is HOLD, the mean target is $13.59, implying −39% from spot. Every disclosed director and the prior CFO sold meaningful blocks at $7–11 in late 2025 and early 2026. The stock then ran from $1.88 to $23.82 (+1,168% off the 52-week low) on the IFX deal and the "miscovered" narrative. None of those insiders bought back in.

The bull case is real but unproven. The bear case is mechanical: revenue priced for a turn that isn't visible in the run-rate. Either the August 4 Q2 print shows the first sequential bounce since Q1 2025, or this round-trips to the $10–14 zone where insiders were transacting and where the sell side has it modeled. Pass at $22. Re-evaluate either on a pullback toward $14, or on a clean Q2 2026 print showing sequential acceleration plus a tangible 800V customer ramp disclosure.

Position sizing if Pink insists on a starter despite the verdict: 0.5–1.0% of portfolio, max, with an explicit invalidation trigger at $14 (commits the trade to either run further or stop out cleanly).

What the company actually does

Navitas designs and sells gallium nitride (GaN) power integrated circuits and silicon carbide (SiC) discrete devices. The competence is GaN-on-silicon: building GaN switching transistors on standard 200mm silicon wafers using a heteroepitaxial growth process, then integrating a gate driver and protection circuitry monolithically on the same die. Integrated GaN is the moat. Competitors who buy GaN-on-Si wafers and bolt a separate silicon driver chip onto the same package end up with extra parasitic inductance, higher gate-loop losses, and an external clamping circuit to protect a fragile depletion-mode GaN device. Navitas's GaNFast and GaNSafe parts ship as single-die solutions with built-in driver, ESD, short-circuit protection, and temperature sensing. That's the product-level claim Irrational Analysis is rewarding.

Revenue today is roughly 70% mobile and consumer (GaN fast chargers for laptops and phones — the original Navitas application), with the remainder split across data center (server PSUs at 1.5–3kW), industrial, and a small but reportedly fast-growing automotive engagement. The mix matters because the consumer GaN charger market is exactly what's collapsing. Chinese OEMs commoditized 65–100W GaN chargers in 2024–2025, ASPs fell, and Navitas's design-win volume migrated to lower-margin sockets. That is the FY24→FY25 revenue cliff, in one sentence.

The Infineon cross-license signed in 2025 is the strategic pivot. It allows Navitas to use Infineon's GaN-on-Si vertical IP without Infineon suing them, and reciprocally allows Infineon to ship Navitas-controller-driven GaN solutions. The market read this as "Infineon validates Navitas." The cleaner read is that Infineon — the world's #1 power semi by share — wanted GaN at scale and chose to license rather than fight, which is validation but also a ceiling on Navitas's licensing-revenue upside (it's not a perpetual exclusive).

First principles — why GaN matters here

Power semiconductors do one thing: switch electricity from one form to another (AC↔DC, one voltage to another) as efficiently as possible. The fundamental physics are governed by the bandgap of the semiconductor material. Silicon has a 1.1eV bandgap, silicon carbide 3.3eV, gallium nitride 3.4eV. Wider bandgap means the material can sustain a higher electric field per micron of thickness before breakdown. That translates directly to three engineering properties power-supply designers care about: lower on-resistance per chip area, higher operating temperature, and faster switching speed without losses ballooning.

At the 650-volt class (the most economically important power-semi voltage rating because it covers PFC stages, server PSUs, EV onboard chargers, and many industrial converters), GaN wins on switching speed: 1 MHz commercial, 50 MHz theoretical, versus 100 kHz–1 MHz for SiC. Higher switching frequency shrinks the magnetic components (inductors, transformers) and capacitors in the surrounding power supply, which is what enables the laptop-charger form factors Navitas pioneered.

At 1,200V and above (EV traction inverters, industrial drives, DC fast chargers, 10kV grid applications), SiC wins on raw power handling. The Irrational Analysis Part 2 post observes that a 650V SiC part handles roughly 240W in the same package where the comparable GaN part is limited to ~120W. You need two GaN chips to match. This is the "GaN for high frequency, SiC for high voltage" line that the industry has settled on.

The 800V DC datacenter is where this gets interesting and where Navitas has a real shot. New AI racks (Nvidia's GB200 NVL72 reference design, the rumored Stargate buildouts) are moving from 48V DC rack power to 800V DC rack power because copper bus-bar I²R losses at 48V × multi-MW racks are punishing. 800V is in the awkward range where SiC is overkill (the per-die power isn't there yet) and the switching frequency benefits of GaN matter enormously for footprint inside the rack. Navitas's GaNSafe IP (protection circuit on-die rather than off-die) becomes a real moat in this footprint-constrained socket, exactly as Irrational lays out.

The catch: 800V DC datacenter is not yet a revenue line for Navitas. It's a 2026–2028 ramp at best. The current $40M of revenue is from the dying consumer charger business.

Where the value chain places Navitas

[GaN-on-Si epitaxial wafers] → [Navitas device + driver IC] → [Power supply OEM] → [Server / charger / EV] → [End customer]
        ★ TSMC (foundry)         ★ Navitas (fabless)             Delta, Lite-On                  Apple, OEMs
        ★ Aixtron (MOCVD tools)
        ★ Soitec (engineered substrates — adjacent)

Navitas is fabless. They outsource GaN-on-Si wafer fabrication to TSMC and to a lesser-disclosed Asian foundry partner. They own the device design, the integrated driver design, and the qualification work. Customer-side they sell to power supply OEMs (Delta, Lite-On, Salcomp for chargers; Vicor and a long tail for data center modules), who in turn sell to the OEMs.

Two implications. First, the bottleneck above Navitas is the MOCVD reactor maker Aixtron (AIXA), which is already in the vault and which Pink already has exposure to — Aixtron is arguably a cleaner play on GaN volume growth than Navitas itself because it benefits regardless of which GaN designer wins. Second, customer concentration is real but not catastrophic: no single customer reported above 10% of revenue in the most recent 10-K, but the top five power-supply OEMs likely represent 50%+ of revenue collectively. The Infineon cross-license is the only "named partner" of strategic weight.

Sector inflection — the part that's right

Irrational's framing of the high-V GaN sector is correct and worth taking seriously. The 800V DC datacenter is a forced transition (you cannot scale 48V DC bus bars to 1MW+ racks economically), the EV onboard charger is migrating from silicon IGBT to wide-bandgap, and DC fast charging stations are a third durable demand vector. Demand is inflecting on a 2026–2030 horizon.

Supply is the part the sector thesis underrates. Six viable high-V GaN players per Irrational (Navitas/Infineon cross-licensed, Renesas, ROHM, Nexperia, Innoscience, TXN), but supply isn't actually constrained — these players have headroom to expand if demand materializes. The "miscovered" claim therefore reduces to a quality argument: Navitas and TXN have better integrated drivers and protection, the rest don't, and footprint-constrained sockets force the customer to pay up. That's plausible at a product level. Whether it shows up in revenue mix and pricing within the next 18 months is the question that decides this trade.

What is observably happening today: the consumer GaN charger market — where Navitas had real share — is being commoditized faster than the datacenter and EV markets are ramping. The revenue curve is going to look like a U-shape if the bull thesis works, with the bottom likely in mid-2026. The stock is trading as if the right side of the U is already visible. It is not.

Numbers, with the inflection front and center

The Core Four read:

1. Organic revenue growth is dramatically negative. FY2023 $79.5M → FY2024 $83.3M (+5%) → FY2025 $45.9M (−45%). Quarterly run-rate Q1 2026 was $8.6M, the lowest in three years. Consensus for Q2 2026 (August 4 report) is $9.8M, basically flat. The Street is not modeling a U-shape recovery yet; they are modeling continued decline.

2. Margins are compressing through the revenue collapse. Gross margin FY2023 39.1% → FY2024 34.0% → FY2025 31.0%, with a likely further dip in Q1 2026 as fixed-cost absorption gets worse on lower revenue. Operating margin is −195% on the FY2025 base because R&D spending ($49.8M in FY2025) has not been cut — management is investing through the trough to capture the data center / EV ramp. That's the right strategic call but it means the burn is structural.

3. Capital intensity is low: Navitas is fabless and capex runs <5% of revenue, which is the only thing keeping the cash burn manageable. Free cash flow FY2025 was −$44.4M, very close to operating cash flow because there's almost no maintenance capex. Capital intensity isn't the problem; opex is.

4. Capital deployment: the company raised $170M in a follow-on offering in early 2026 (priced in the high single digits — see insider context below), which is the only reason cash is now $237M vs $87M a year ago. That offering occurred at roughly half the current price. The market financed Navitas at $8–10, and then bid the stock to $22 over the following four months. This pattern is the single most informative data point in the file: shareholders who funded the runway are now sitting on a 100%+ gain inside four months. They are sellers, not holders, into any disappointment.

Quarterly revenue trajectory:

Q	Revenue	YoY
Q1 2025	$14.02M	−37%
Q2 2025	$14.49M	−26%
Q3 2025	$10.11M	−51%
Q4 2025	$7.30M	−62%
Q1 2026	$8.60M	−39%
Q2 2026E	$9.84M	−32%

The second derivative turned in Q1 2026 (sequential acceleration from $7.3M to $8.6M, +18% QoQ). That is real, and it is the only fundamental number that supports the bull case. If Q2 prints $11M+ on August 4, the U-shape narrative gets credible. If Q2 prints $9–10M as consensus expects, the story stays alive but the rally has nothing fresh to feed on. If Q2 prints $7–8M, the entire story breaks and the stock round-trips fast.

Balance sheet is the survival cushion. $237M cash, $6.5M debt, $443M stockholders equity. At the current $44M annual burn that is 5.4 years of runway. The company does not need to raise again before 2028 at this burn rate. The 5-year runway is the only thing that allows the bull thesis to remain investable at all — without it, the gap between current revenue and the 800V DC inflection would be a solvency question.

Management — the part that should make you pause

Three findings, in order of weight.

First, the insider sales pattern. In the six months between November 2025 and March 2026 the following insiders sold stock on the open market:

Insider	Role	Shares	$ Sold	Avg price
Brian Long	Director	1,705,395	~$14.6M	$8.50–8.55
Dipender Saluja	Director	1,188,884	~$9.8M	$8.10–8.65
Ranbir Singh	Director	568,450	~$5.3M	$8.68–9.52
Gary Wunderlich	Director	178,300	~$1.65M	$9.25–9.31
Todd Glickman	CFO (departed Mar 2026)	~120k	~$2.0M	$8–11
Christophe Allexandre	CEO	9,236	$82K	$8.93

That's ~$33M of director and C-suite open-market sales at $8–11 per share, with the largest seller being Brian Long, a co-founder and long-time director. The stock is now $22.32. Every one of these insiders left ~100% on the table if they intended to capture the post-IFX-deal run. The most charitable read is portfolio diversification through 10b5-1 plans (which would be disclosed in the Form 4 footnotes — I'm reading these as open-market sales but a full filings review should confirm). The harsh read is that the people closest to the business did not believe the stock was on the verge of a 2.5x move when they sold.

Insider ownership today is 9.25% of shares outstanding, which is on the lower end of what you want to see in a story stock. After this round of selling it's lower than it was a year ago.

Second, CFO turnover. Todd Glickman (CFO since the SPAC) sold consistently through late 2025, then departed in March 2026 and was replaced by Tonya Stevens. The Glickman departure was not pre-announced. A CFO selling on the way out of a company whose stock is about to triple is a meaningful signal. Stevens shows $0 reported total comp because she's new in the role, which is normal but means there's no public read yet on her compensation alignment.

**Third, CEO Christophe Allexandre joined in mid-2024 from Renesas/ON Semi. He's a credentialed power-semi operator and the right profile for the strategic pivot toward data center and automotive. His total reported comp at $416K is unusually low for a public-company semi CEO and suggests heavy stock-based compensation that will show up in 2026's proxy. The CEO sold only 9,236 shares in the entire period — meaningfully different from the director behavior. He is aligned to the bull case in a way the long-tenured board is not.

Management grade: B-/C+. The strategic positioning (CEO hire, IFX deal, $170M capital raise) is correct. The capital-allocation signal (board-level insiders cashing out at half the current price) is bad. No shell-entity or related-party flags surfaced in the surface scan; a full DEF 14A read would tighten this verdict.

Competition

Navitas's direct competitors in integrated high-V GaN are Texas Instruments (LMG3422 family, recently launched), Infineon (CoolGaN), Power Integrations (HiperGaN), and the Chinese pure-play Innoscience (which went public in HK in late 2024 and is the volume threat). Renesas and Nexperia ship discrete GaN. Of these, TXN is the only one with comparable integration plus a $280B-mcap balance sheet to fund the multi-year ramp, and TXN's GaN portfolio is a rounding error on their P&L but is built on their own internal manufacturing. Innoscience is the price disruptor — they ship 650V GaN at ~30% discount to Navitas pricing and target the same Chinese consumer charger sockets that are bleeding Navitas now. Wolfspeed is excluded from the comp (per Pink) and is essentially zero-share in GaN anyway, as Irrational notes.

Moat verdict: medium. GaNSafe integration is a real technical edge in footprint-constrained sockets. It is not an unbridgeable moat — TXN can and will close the gap. The moat works specifically in 800V DC datacenter, not in the consumer market where Navitas's revenue actually sits today. The question is whether Navitas can transition revenue mix faster than competitors can close the integration gap.

Why now — the honest version

The "Why now" for the sector is real: 800V DC datacenter transition, EV onboard charger migration to wide-bandgap, DC fast charging buildout. None of those are 2024 narratives; they are all 2026–2030 narratives.

The "Why now" for the stock at $22 is something else. The stock is at $22 because:

1. The IFX cross-license validated the IP in mid-2025
2. Project Stargate rumor mill name-dropped Navitas as a power-supply IP licensor (unconfirmed by Navitas)
3. Short interest at 20.45% of float created squeeze fuel
4. The Q1 2026 sequential revenue bounce ($7.3M → $8.6M) was technically the first positive datapoint in 18 months
5. Retail-investor narrative around "the GaN pure-play" picked up on X/Twitter

That is a five-factor catalyst stack, of which only #5 is sustainable. The first four are largely priced in.

Risks, structured

Risk	Likelihood	Mitigants	Mgmt de-risk plan	Closable?
Revenue inflection later than 2026E	High	Cash runway covers >5 yrs; CEO transition complete	August 4 Q2 print is the first real test	Yes, on two consecutive QoQ acceleration prints
Innoscience pricing pressure in consumer	High	Mix shift to data center / EV (low-share but exists)	Explicit exit from low-margin Chinese consumer sockets	Yes, by showing mix shift in 10-K segment data
Customer concentration in power-supply OEMs	Medium	Top customer <10%	Broader design-win disclosure	Slow but yes
800V DC inflection delayed beyond 2027	Medium-High	IFX cross-license keeps optionality	Need named hyperscaler design win	Yes, on a public customer disclosure
Equity dilution if revenue doesn't recover by 2027	Medium-Low	$237M cash + low burn vs revenue base	Not raising in next 12mo per CFO commentary	Structurally low
Insider sell pattern signals downside	Medium-High	Some 10b5-1 plan involvement likely	None — pattern continues	No, it is what it is
Multiple compression to peer normalization	High	None	None	This IS the bear scenario

Bear case: Q2 2026 prints in-line at $9.8M, no named 800V DC customer announced through 2026, Innoscience further commoditizes the consumer book. Stock retraces to $10–14 (where insiders sold, where the $170M raise priced, where analyst consensus has it). Downside target: $11 (−51%).

Bull case: Q2 2026 prints $11M+ (vs $9.8M consensus), management names a hyperscaler 800V customer on the call, Q3 prints $13M+. Multiple expansion accelerates as the U-shape becomes visible. Stock holds $22–28 zone through year-end. Upside target: $32 (+43%).

Probability-weighted view: Bear 55%, Base (drift sideways $18–24) 30%, Bull 15%. Expected value at $22 is mildly negative.

What needs to be true for $22 to make sense

Working backward from a 25× P/E on FY2028 earnings (the multiple a high-quality power-semi pure-play might earn at scale):

• FY2028 revenue: $400M minimum (10× the current run rate)
• FY2028 gross margin: 50%+ (vs current 31%; requires data center / industrial mix shift to dominate)
• FY2028 operating margin: 20%+ (vs current −195%)
• FY2028 net income: $50M
• FY2028 EPS: $0.21
• 25× = $5.25 implied stock price discounted to 2026 = $3.50 at 15% discount rate

That is the conservative 25× multiple. To get to $22 today on FY2028 numbers, you need either FY2028 revenue closer to $700M (15× current), or a multiple closer to 50× FY2028 earnings, or you need the inflection compressed into 2026–2027 rather than 2028. All three are possible. None are visible in the current run rate. The price requires Pink to underwrite the most aggressive version of the bull case, with no margin of safety.

Ownership & sentiment

• Institutional ownership: 51.6%
• Insider ownership: 9.25%
• Short interest: 20.45% of float (41.9M shares short), 1.32 days to cover
• 8 analysts covering, consensus HOLD (mean 2.875), 1 strong buy / 1 buy / 5 hold / 1 sell / 0 strong sell

Short interest at 20% of float is the squeeze fuel that drove the $1.88 → $22 run. The flip side: short interest at 20% is also a measure of how many sophisticated investors think the stock is mispriced higher. The shorts have been right on fundamentals (revenue −45%) and wrong on price (stock +1,168%). One of those facts is going to revise.

Position sizing and entry discipline

If Pink wants exposure to the IA thesis specifically, NVTS is the wrong vehicle at the wrong price. The cleaner expressions are:

• AIXA for MOCVD reactor demand (already owned, already in vault)
• ON (deep-dive next) for SiC vertical integration with a real revenue base
• Smaller starter in NVTS only if the August 4 print delivers the second consecutive sequential acceleration

If forced to take a NVTS position today: 0.5–1.0% portfolio weight, invalidation stop at $14, scale only on a confirmed inflection print on August 4. Recommended action: PASS at $22.

Better entries to watch:

• $14 (recent insider transaction range and analyst consensus zone) — initiate 1.5% starter
• $11 (post-IFX-deal pre-rally zone) — initiate 2.5% position
• $7 (low end of analyst target range) — initiate 3% position with conviction

These re-entries are conditional on the bull thesis still being intact at that level — a round-trip caused by Q2 disappointment AND no 800V customer disclosure changes the analysis.

Sources

• Live yfinance pull, 2026-05-15 (price, financials, balance sheet, recommendations, insider transactions, earnings calendar)
• 2026-04-25-power-semis-part-2-sic-vs-gan-wolfspeed — Irrational Analysis Substack, archived
• 2026-04-21-power-semis-800v-dc-and-gan-mini-note — Irrational Analysis Substack, archived
• SEC EDGAR — NVTS 10-K FY2025, 10-Q Q1 2026, Form 4 insider filings Nov 2025–Apr 2026 (referenced via yfinance, full reading deferred)
• AIXA — adjacent vault page on MOCVD upstream bottleneck
• SOITEC — adjacent vault page on engineered substrate monopoly
• AEHR — adjacent vault page on SiC/GaN burn-in test

Cross-refs

• TXN (to be written) — the "other winner" in IA's high-V GaN ranking; large-cap balance sheet, tiny GaN exposure
• ON (deep-dive next) — SiC vertical integration leader; IA disclosed long
• IFX (to be written) — the cross-license counterparty; IA called out for performance lag
• themes/sic-gan-high-voltage-thesis (to be written) — sector thesis page
• AIXA — MOCVD reactor upstream
• SOITEC — engineered substrate adjacent
• AEHR — burn-in test adjacent

SA cross-check

SemiAnalysis mirror returned only two 2022 industry roundup mentions of Navitas, neither of which is load-bearing for the current thesis. No recent SA coverage of NVTS specifically. Note: SA has covered the 800V DC datacenter architecture transition extensively in their AI Datacenter Anatomy series — that work supports the sector thesis but does not pick NVTS as a vehicle. Treated as supportive of sector view, neutral on stock pick.

Briefings

• 2026-05-15 · Navitas Semiconductor (NVTS) · vault

Source updates (auto-maintained)

Vik's Newsletter (May 6, 26) - Vik's Newsletter

Navitas showcased its 800V-6V DC-DC chips inside NVIDIA's MGX power delivery board at NVIDIA GTC in Taipei, using 16 650V GaNFast FETs to convert directly to 6V at 97.5% efficiency and 1 MHz switching, and was recognized at NVIDIA's partner ceremony.

Relevant to your thesis: First named, in-situ evidence of the 800V datacenter ramp the wiki flags as "not yet a revenue line" — still pre-revenue but the MGX design win is the closest thing to a customer disclosure that underpins the bull case.

Source: https://www.viksnewsletter.com/p/computex-2026-debrief