Ambarella

Phase A — Understand the business

Lens 1 · Company Overview

Ambarella designs low-power system-on-a-chip (SoC) semiconductors and software for edge AI and computer vision — chips that let a camera or sensor perceive, reason, and act locally instead of streaming raw video to the cloud. Incorporated in the Cayman Islands on Jan 15, 2004; fabless; ~75% of employees are in R&D, mostly in Asia with an executive core in Santa Clara, CA.

What it actually sells. A family of vision/AI-inference SoCs built around its proprietary CVflow® AI accelerator (a deep-learning neural-network engine fused with a best-in-class video/image pipeline). Newest generation is the CV7 line. Installed base is >42 million AI SoC units shipped. The pitch is "physical AI at the edge" — autonomous decision-making on-device, with power and bandwidth as the binding constraints Ambarella optimizes for.

End markets (the 10-K's own framing): automotive camera systems (ADAS, e-mirrors, data-loggers/dashcams, commercial-vehicle safety), video security cameras (the legacy heart of the business, now AI-upgrading), fixed robots + autonomous mobile robots (AMRs), industrial + intelligent transportation systems, and consumer devices (action/drone/360° cameras — the old GoPro-era business, now a small tail).

Customer/channel structure — the single most important structural fact. Ambarella sells to OEMs and ODMs globally, but routes the overwhelming majority of revenue through one distributor: WT Microelectronics (formerly Wintech), its non-exclusive sales rep + fulfillment partner across Asia ex-Japan. WT accounted for ~70% of total FY2026 revenue (FY25: 63%, FY24: 53%) — and the concentration is rising every year. A second name, ODM Chicony, was a 10%+ customer (14%) only as recently as FY24. Design cycles are long (12–18 months, longer for automotive safety), which creates stickiness once designed-in but also a slow ramp. No long-term volume contracts with the foundry/assembly chain; revenue recognition is largely on shipment to distributors/ODMs.

Verdict on the model: a real, defensible technology franchise (low-power vision AI is hard and Ambarella is genuinely good at it) wrapped in a commercially fragile go-to-market — one distributor, Taiwan-routed, with the end-demand one layer removed from Ambarella's visibility.

Lens 2 · Supply Chain

Map upstream → Ambarella → end customer, named at every link:

• EDA / IP: standard cadence of EDA tools + third-party IP (not disclosed by name in the 10-K; Arm-based application processors are typical for the category).
• Foundry (the chokepoint): Samsung Electronics is the named foundry, with capacity in Austin, Texas and South Korea; Ambarella can buy fully-assembled/tested parts or tested die in wafer form. For advanced nodes (the 10-K explicitly names 4nm and 2nm), Ambarella states it is "increasingly dependent" on the handful of foundries — Samsung and TSMC — that offer those processes. So: Samsung today, TSMC the named alternative/future for leading-edge AI parts. Single-/dual-source foundry is the hardest chokepoint in the chain.
• Assembly & test: Samsung subcontracts the assembly + initial test; other third-party OSATs in Asia. No long-term agreements — explicitly flagged as a concentration risk.
• Ambarella (fabless design house): Santa Clara HQ + Hong Kong; design centers Asia-heavy.
• Distribution: WT Microelectronics (Asia ex-Japan) is the dominant fulfillment artery — ~70% of revenue. Direct sales force covers US/Asia/Europe; business-development offices in China, Germany, Japan, South Korea, Taiwan.
• ODMs: Chicony and others build finished devices incorporating Ambarella SoCs on behalf of multiple end-customers.
• OEMs / end-customers (named/known): security — Hanwha (Korea; subject of a new >$800M long-term agreement — see Lens 5/6), legacy Hikvision/Dahua exposure historically (China surveillance); automotive — commercial-vehicle and ADAS OEMs (named in aggregate, not individually, in the 10-K); consumer — the residual action/drone-camera base.

Geographic "bill-to" concentration (an artifact of the WT/Taiwan routing, NOT true end-demand): Taiwan $271.9M = 70% of FY26 revenue, Asia-Pacific ex-Taiwan $71.0M, North America ex-US $21.8M, Europe $20.1M, US $5.8M. Read Taiwan as "where WT/ODMs invoice," not where the cameras are used.

Chokepoints, ranked: (1) leading-edge foundry (Samsung/TSMC, no second source at 2nm), (2) WT distributor single-point-of-failure, (3) Asia-concentrated assembly/test with no long-term contracts, (4) China geopolitical exposure of the security end-market.

Lens 3 · Competitive Advantages (moats)

The real moat is a power-efficiency + software-stack advantage in vision AI, defended by long design-in cycles. Ambarella's CVflow architecture pairs a class-leading image/video pipeline (its 20-year heritage) with a purpose-built AI accelerator, delivering more inference per watt than general-purpose GPUs/SoCs in the thermal/power envelope of a camera. That matters because the binding constraint at the edge is power and bandwidth, not raw TOPS. CEO Fermi Wang's inaugural 2026 IEEE Arun N. Netravali Award for video-analytics systems is a (soft) external signal of genuine technical depth.

Moat sources, honestly graded:

• Switching costs / design-in lock-in — STRONG. Once an SoC is designed into an automotive or security platform (12–24+ month cycles, longer for safety), it ships for the product life (often >24 months in auto/robotics). The new multi-year LTAs (Hanwha >$800M/10yr) institutionalize this.
• IP / architecture — MEDIUM-STRONG. CVflow + the video pipeline is hard to replicate; but a recent patent-ruling headline shows the IP estate is litigable terrain.
• Process/scale — WEAK (this is the vulnerability). Ambarella is small (~$390M revenue) competing against giants (NVIDIA, Qualcomm, TI, Renesas, Mobileye/Intel) with vastly more R&D dollars and foundry leverage. It rents leading-edge capacity from Samsung/TSMC rather than commanding it.
• Brand — MEDIUM in security/pro-AV, LOW in automotive where Mobileye/NVIDIA own mindshare.

Bargaining power: weak over suppliers (rents foundry capacity from Samsung/TSMC who serve far bigger customers first) and weak over its channel (70% through one distributor that is not exclusive). Power over end-customers is moderate and rising via LTAs. Net: a technology moat without a commercial moat — the defining tension of the name.

Lens 4 · Segments

Ambarella reports as a single operating + reportable segment (Note 16) — the CODM (CEO) runs it on consolidated net loss. So there is no GAAP segment P&L by product/end-market; the only ``-grounded cuts are geography (Lens 2) and the qualitative end-market commentary management gives on calls.

What the filings + calls let us assert:

• Revenue trajectory (consolidated, audited): FY24 $226.5M → FY25 $284.9M (+25.8%) → FY26 $390.7M (+37.1%). That is a sharp re-acceleration after the FY24 trough.
• Mix shift is the story: FY26 growth came "primarily due to higher product unit shipments and an increased percentage of sales from higher-ASP AI inference processors" on strong edge-AI demand, partly offset by lower NRE service revenue. Edge-AI/AI-inference is now >75% of revenue.
• By end-market (qualitative, ``): the two engines are (1) video security AI-upgrade cycle — anchored by the Hanwha LTA, and (2) automotive, which hit an all-time quarterly record in Q1 FY27 on commercial-vehicle AI penetration. Consumer (action/drone) is a declining tail.

Direction: revenue accelerating, mix enriching toward higher-ASP AI parts, services revenue decelerating. The absence of segment-level margin disclosure is itself a (minor) transparency limitation — you cannot see whether automotive is dilutive or accretive to gross margin from the filings alone; management's 59–62% long-term GM target implies the mix shift is accretive.

---

Phase B — Measure performance

Lens 5 · Earnings Result (latest print: Q1 FY27, quarter ended Apr 30, 2026)

The numbers:

• Revenue $100.357M vs $85.872M PYQ = +16.9% YoY; first $100M+ quarter; down ~0.5% sequentially. Landed slightly above the $97–103M guide midpoint.
• GAAP gross profit $58.589M → GAAP GM 58.4%; non-GAAP GM 59.9%. Within the 59–62% long-term target band.
• Opex $78.005M — of which R&D $58.140M (≈58% of revenue) and SG&A $19.865M. R&D essentially equals gross profit; this is a company spending nearly its entire gross margin on engineering.
• GAAP loss from operations −$19.416M (vs −$25.858M PYQ — loss narrowing). GAAP net loss −$18.093M; EPS −$0.41. Non-GAAP net income +$5.0M, +$0.11/sh — a non-GAAP beat. The ~$24M GAAP↔non-GAAP gap is almost entirely stock-based comp ($21.9M SBC in the quarter).
• Guidance (Q2 FY27): revenue $105–111M, midpoint $108M — strong sequential step-up; reiterated 59–62% long-term GM.
• Balance sheet: cash + equivalents $114.4M, total cash + marketable securities $277.8M, no debt. But operating cash flow was −$25.6M this quarter, driven by a −$28.0M inventory build (inventories up sharply) — a flag worth watching (building ahead of demand, or stuffing the channel?). AR with WT was ~$24.6M at FY26 year-end.
• Market reaction: stock DIPPED despite the EPS beat. Classic "priced for more" — the bar set by the prior +37% year and the M&A bid was higher than an in-line-to-slight-beat print cleared.

Full-year context (FY26, audited): revenue $390.7M (+37%), GAAP operating loss −$82.5M (vs −$126.6M FY25, −$154.6M FY24 — losses shrinking fast), GAAP net loss −$75.9M (−$1.78/sh). Operating cash flow +$73.5M for the full year (the business is FCF-positive on a full-year basis once you exclude the quarterly inventory swing). The trajectory is unambiguous: revenue up, losses compressing, full-year cash generation positive.

Unusual vs. its own history: the inventory build (+$28M in one quarter on $100M revenue) and the sequential revenue dip into a guided sequential jump — read the Q2 guide as the tell that Q1's flatness was timing, not demand softening.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on disk (transcripts/ empty), so this is ``-grounded from call coverage.

What management is focused on (Q1 FY27 call, May 28 2026): (1) edge-AI leadership — demand signals "very strong"; (2) a deliberate pivot to Long-Term Agreements to make revenue "more predictable" — headlined by a Hanwha LTA that could exceed $800M over 10+ years; (3) automotive hitting record quarterly revenue on commercial-vehicle AI; (4) margin discipline — reiterated 59–62% long-term GM as mix enriches.

Tone shift over time: the arc from the FY24/FY25 calls (defensive, post-trough, inventory-correction language) to FY26/Q1-FY27 (offensive — "new phase of market development," LTAs, record automotive) is a clear sentiment upgrade. The recurring new phrases are "edge AI," "physical AI," "long-term agreements," "predictable revenue." The thing they've stopped emphasizing: the consumer/action-camera business and NRE services. Management is narrating a transition from lumpy, customer-concentrated, project-driven revenue to a contracted, AI-platform revenue base. The market's skepticism (stock dip on the beat) says investors want to see the LTAs convert to GM-accretive shipped revenue before re-rating further.

Lens 7 · Comps

Peer multiples are ``, point-in-time June 2026, and databases disagree materially — treat as directional. AMBA itself:

• Market cap ~$2.7–2.95B
• EV/Sales ~6.5x–8.8x (sources vary: gurufocus EV/Rev 6.48; another 8.76)
• P/S (TTM) ~5.4x; forward P/E ~83x (near-meaningless — GAAP unprofitable; non-GAAP basis only)
• 5-yr avg ROE: negative (cumulative GAAP losses; accumulated deficit −$328M)

Read: AMBA trades at a premium P/S to Mobileye (~5.4x vs ~3.9x) despite being ~10x smaller and GAAP-unprofitable, while Mobileye is the scaled, profitable automotive-vision leader. The premium is justified only by (a) higher growth (FY26 +37% vs Mobileye's slower line) and (b) the embedded M&A bid. On fundamentals alone the stock is rich; on growth-plus-takeout it is defensible. I will not fabricate the peer multiples I could not source this run — they are marked n/a.

Lens 8 · Stock-Price Catalysts (what actually moves AMBA)

The historical tape reveals a clear pattern — AMBA moves on customer-concentration shocks, China/geopolitics, and now M&A, more than on broad-market beta:

• Jan 2018: −14% in a day alongside GoPro (−17%) when GoPro cut guidance; GoPro later designed Ambarella out — the original concentration-shock lesson.
• 2018–19: revenue expected −22.9% (FY19) as the GoPro/drone/VR business collapsed.
• Oct 2019: −12% after-hours when customer Hikvision was added to the US entity list — China surveillance/geopolitics directly hit the tape.
• June 24, 2025: +19–21% in a day on the Bloomberg report that Ambarella is exploring a sale. This is the single biggest up-catalyst in the recent record and remains unresolved.
• 2026: episodic ~9% drawdowns on growth/valuation wobbles and "growth not translating to profit" downgrades; modest pops on LTA/award news.

What the market reacts to for this name: (1) anything touching customer concentration (a key customer win or loss moves it hard), (2) China/geopolitical headlines on the security end-market, (3) the M&A bid (binary), and (4) the gap between revenue growth and GAAP profitability (the bear's recurring lever). Pure earnings beats — like Q1 FY27 — can fall if expectations are stretched.

---

Phase C — Judge people & books

Lens 9 · Management

CEO: Fermi (Feng-Ming) Wang — co-founder (Jan 2004), President & CEO, 30+ years in digital video/HPC/AI; PhD-level technologist; just won the inaugural 2026 IEEE Arun N. Netravali Video Analytics Award.

1. Track record — STRONG technically, MIXED commercially. Built Ambarella from scratch into the edge-vision leader with >42M units shipped, and navigated two near-death concentration shocks (GoPro design-out, Hikvision ban) to a re-accelerating, mix-enriched FY26 (+37%). That's real operational resilience. The commercial blemish: the business has been GAAP-unprofitable for years and is still one-distributor-dependent on his watch.
2. Tenure & skin in the game — a real flag. Wang directly holds only ~1.8% / ~806k shares and has been a consistent net seller — ~124k shares sold LTM, zero open-market buys. The CFO and multiple VPs have also been selling through 2026 (CFO sold >$400k in March; VPs selling $660k blocks). For a founder-led "story" stock pricing a re-acceleration, the insiders voting with their feet is not the alignment you want to see. Institutions own ~82% (Vanguard ~13%).
3. Capital allocation — disciplined but uninspiring. No debt; ~$278M net cash; minimal buyback (token ~$1–2.4M repurchases that barely offset SBC dilution); no dividend. The "allocation" is really just funding heavy R&D ($238.5M FY26) — appropriate for the stage, but the share count keeps creeping up via SBC (42.7M → 43.9M and rising). ROE/ROIC negative throughout. No value-destroying M&A — a point in their favor.
4. Red flags — moderate. Persistent insider selling; SBC at ~25% of revenue flattering the non-GAAP narrative; goodwill of $303.6M (~38% of total assets) sitting on the balance sheet from prior deals, untested by an impairment but a latent risk.
5. Archetype — founder-technologist, not operator-capitalist. Implication: superb at staying on the technology frontier (CVflow, CV7, 2nm roadmap), structurally less focused on the commercial de-risking (diversifying off WT, driving to GAAP profitability) that the market now demands. Which is exactly why a sale to a larger platform is strategically logical — and why management's "there are advantages to having more scale" comment lands.

Lens 10 · Forensic Red Flags

Forensic lens, every figure labeled.

• SBC dependence (the headline accounting watch-item). Full-year SBC: FY24 $111.3M, FY25 $107.8M, FY26 $98.0M. In FY26 SBC = ~25% of revenue and roughly 42% of gross profit. The entire GAAP→non-GAAP bridge to "profitability" is SBC. Non-GAAP "$5M net income" in Q1 FY27 exists only because ~$22M of SBC is added back. Real economic dilution is occurring (share count rising), masked in the non-GAAP framing the market quotes.
• Inventory outrunning revenue. Inventories: $34.4M (Jan-25) → $52.2M (Jan-26) → built another +$28.0M in Q1 FY27. Inventory growth (+52% YoY into FY26-end, then another large step) is running ahead of revenue growth. Benign read: building for the LTA ramp. Bearish read: demand pull-in / channel risk given the 70%-through-one-distributor structure. Watch the next 1–2 quarters of inventory days.
• Revenue recognition + channel. ~70% of revenue ships through WT (sell-in to a distributor). Sell-in vs. sell-through visibility is inherently weaker; a distributor inventory correction (as happened across semis in 2023) would hit reported revenue with a lag. Not an irregularity — but a structural quality-of-revenue discount.
• Goodwill $303.6M — unchanged YoY, ~38% of total assets, no impairment taken. Latent write-down risk if the AI-inference thesis disappoints.
• Cash flow vs. earnings: healthy at the full-year level (FY26 OCF +$73.5M vs net loss −$75.9M — the gap is SBC + D&A, legitimate non-cash adds), but the quarterly OCF turned negative (−$25.6M) on working capital. No divergence that signals manipulation; the non-cash adds are well-understood (SBC, D&A).
• Receivables: AR $39.2M at FY26-end, of which ~$24.6M is WT — concentrated credit risk in one counterparty.
• Tax: small provisions on pre-tax losses; valuation allowance dynamics on deferred tax assets (FY24 had a $20.9M tax provision swing from DTA treatment) — worth noting but not a red flag.

Regulatory findings (required sub-section). Read from regulatory/regulatory-findings.md (fetched 2026-06-20):

• SEC Litigation Releases: None found naming Ambarella in the 2021-06-20 → 2026-06-20 window (EDGAR EFTS, form LR).
• AAERs: None found in the same window (EDGAR EFTS, form AAER).
• 10-K Item 3 (Legal Proceedings): Ambarella discloses only ordinary-course commercial/employment/IP matters, with no accruals for contingent liabilities as of Jan 31, 2026 or 2025, and management's belief that outcomes will not be materially adverse. (Note: a 2026 patent-ruling headline surfaced in the press — IP litigation is normal terrain for a chip designer and not disclosed as material.)
• Non-SEC enforcement (web search): No material FTC/DOJ/FDA/CFPB enforcement actions, consent decrees, or fines against Ambarella surfaced. The relevant geopolitical exposure is indirect — its customers (e.g., Hikvision) have faced US entity-list/sanctions actions, which is a demand risk, not an Ambarella compliance finding.

Net: clean regulatory/accounting record. The legitimate forensic watch-items are SBC-as-profitability-mask and inventory/channel quality-of-revenue, not fraud risk.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection (FY27 / FY28 / FY29; fiscal year ends Jan 31)

Built bottom-up from FY26 actuals + the Q2 FY27 guide. Ambarella is GAAP-loss-making, so I project non-GAAP EPS (the basis the company and Street use); GAAP will lag by ~$0.40–0.50/qtr of SBC.

Anchors: FY26 revenue $390.7M, non-GAAP GM in the 59–62% target band, ~44M shares (rising ~3%/yr via SBC). Q1 FY27 non-GAAP EPS $0.11; Q2 FY27 guide $108M midpoint. FY27 company-framed growth signaled ~19–25%.

• Bear (FY27): revenue ~$430M (+10%, a distributor/channel correction or auto pause); GM ~58%; opex flat-ish → non-GAAP EPS ~$0.60.
• Base (FY27): revenue ~$470M (+20%), in line with the company's own 19–25% frame and the strong Q2 guide; non-GAAP GM ~60%; modest opex growth + operating leverage → non-GAAP EPS ~$1.30.
• Bull (FY27): revenue ~$500M (+28%) as LTAs (Hanwha) + auto ramp pull forward; GM ~61%; leverage → non-GAAP EPS ~$1.90.

FY28 base: revenue ~$560M (+19%), GM ~61%, leverage → non-GAAP EPS ~$2.20. FY29 base: revenue ~$660M (+18%), GM ~61% → non-GAAP EPS ~$3.00. The thesis math: at $660M revenue and a 60%+ GM with R&D no longer growing dollar-for-dollar with revenue, GAAP profitability arrives ~FY28–29 and non-GAAP EPS compounds fast off a tiny base — which is the entire bull re-rating case.

Lens 12 · Bull vs Bear

Bull case. Ambarella is the purest public way to own low-power edge-AI vision silicon at the exact moment "physical AI" (cameras/robots/vehicles that perceive and act on-device) inflects. FY26 +37% with mix enriching to high-ASP AI parts proves the inflection is real, not a story. The LTA pivot (Hanwha >$800M/10yr) structurally de-risks the historically lumpy revenue and should compress the volatility discount. Operating leverage is finally visible: losses went −$155M → −$127M → −$83M while revenue grew, and full-year OCF is +$73.5M. Net cash ~$278M, no debt — fully funded to GAAP profitability (~FY28–29). And there is a live, unresolved M&A bid from June 2025 that a larger platform (Qualcomm, NVIDIA, TI, Renesas, or PE) could revive at a premium because the technology is strategically scarce. Heads (re-rate on growth + GAAP profitability), tails (taken out at a premium).

Bear case (permanent-impairment risks).

1. Customer/channel concentration is a single point of failure and getting WORSE — 53%→63%→70% through WT. The GoPro and Hikvision episodes prove this company can lose a third of its revenue from one counterparty event. A WT dispute, a distributor inventory correction, or a China-security demand shock could permanently reset the revenue base.
2. It may never out-scale the giants. Mobileye, NVIDIA, Qualcomm, TI, Renesas, Horizon Robotics all target the same automotive/edge sockets with far deeper pockets. A small fabless designer renting Samsung/TSMC capacity can be out-roadmapped at the leading node.
3. Valuation prices perfection. ~6.5–8.8x EV/Sales and ~83x forward (non-GAAP) P/E on a company that is still GAAP-unprofitable leaves no margin for a growth stumble — and the stock already fell on a Q1 beat.

Pre-mortem (it's Dec 2027, thesis broke — what happened?). Most likely: the M&A bid evaporated (no buyer at an acceptable price), a semiconductor/channel inventory correction hit WT and reported revenue dropped ~15–20% YoY, automotive design-win ramps slipped, and the market de-rated a still-loss-making name from ~7x sales toward ~3–4x (Mobileye-like), halving the stock. The fingerprints would be: rising inventory days (already visible), WANG/CFO continuing to sell, and "growth not translating to profit" downgrades.

Are multiples too high? On standalone fundamentals, yes — a ~7x sales, GAAP-loss name needs flawless execution. The premium is only defensible if you underwrite both sustained 20%+ growth to GAAP profitability and meaningful M&A optionality.

Contrarian view (what the market refuses to see): the bears fixate on "unprofitable + concentrated," but the market may be under-pricing the LTA-driven business-model transition — Ambarella is quietly converting from a project-revenue chip designer into a contracted edge-AI platform with multi-year visibility. If the Hanwha-style LTAs proliferate, the right comp shifts from "lumpy small-cap chip" toward "sticky AI-platform," and both the multiple and the M&A logic strengthen. The flat-on-a-beat reaction may be late-cycle skepticism that a stickier, more profitable Ambarella is about to disprove.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case.

• The 70%-through-WT structure is the whole short. This is not a diversified franchise; it is one distributor away from a revenue cliff. Bulls call the LTAs "de-risking" — but an $800M/10yr Hanwha headline averages ~$80M/yr and is still one customer; it concentrates security exposure rather than diversifying it. Where revenue is concentrated, that's where the next blow-up comes — and AMBA's own history (GoPro, Hikvision) is the proof of concept.
• The moat is narrower than bulls think. "Power-efficient vision AI" is a real edge today, but NVIDIA (Jetson), Qualcomm, and Horizon Robotics are pouring resources into the same edge-inference sockets, and Mobileye owns automotive ADAS mindshare. A small designer cannot win a TOPS-per-watt arms race against opponents with 20–50x the R&D budget indefinitely.
• Most dangerous competitor bulls underestimate: Qualcomm — it has the scale, the auto-cockpit/ADAS beachhead, the foundry leverage, and is exactly the kind of acquirer that could either buy AMBA or simply crush it on price. Horizon Robotics (China) is the low-cost disruptor in the exact security/auto markets AMBA leads.
• Capital-allocation / incentives: founder owns ~1.8% and is a net seller; CFO and VPs selling; 25%-of-revenue SBC is the only thing turning losses into "non-GAAP profit." The insiders' revealed preference contradicts the bull narrative.
• Assumptions that must hold for ~$66 / ~7x sales: sustained 20%+ revenue growth, GM holding 60%+, opex leverage delivering GAAP profitability by ~FY28, and WT/channel staying intact, and the M&A bid either reviving or growth re-rating the name. That's a lot of ANDs.
• If growth disappoints 20–30% (revenue +0% to +10% instead of +20%): the loss-making + concentrated profile re-rates toward Mobileye's ~3.5–4x sales → stock ~$35–40, roughly −40% to −45%.
• Single permanent-impairment scenario, plausibility ~20–30%: a US-China escalation that cuts the security end-market AND a WT distributor correction in the same window — a repeat of 2018/2019 but off a higher base. Not the base case, but the tail is fatter than a 7x-sales multiple admits.

Lens 14 · Management Questions (ordered by information value)

1. WT Microelectronics is now ~70% of revenue and rising. What is the concrete plan and timeline to diversify the channel, and at what concentration level would you consider it a strategic risk worth acting on?
2. Walk us through the economics of the Hanwha LTA: what annual revenue and gross margin does it actually contemplate, what are the volume commitments vs. ceilings, and is it incremental or a re-papering of existing business?
3. When do you expect GAAP (not non-GAAP) operating profitability, and what revenue level and opex trajectory underpin that?
4. SBC is ~25% of revenue. What is the multi-year path to bring it down as a % of revenue, and how do you think about the real dilution it creates?
5. The June 2025 strategic-review/sale process was never formally closed. What is its current status, and how is the board weighing independence vs. a sale given your "advantages to scale" comment?
6. Inventories jumped ~$28M in Q1 FY27. Is that an LTA-driven build, and how do you get comfortable it isn't channel over-shipment given the WT structure?
7. At the 2nm node you're dependent on Samsung/TSMC capacity behind much larger customers. How do you secure leading-edge allocation, and what's the second-source plan?
8. In automotive specifically, name the design-win pipeline and expected production-revenue ramp by year — how much of the "record" auto quarter is durable vs. initial-ramp lumpiness?
9. How do you defend TOPS-per-watt leadership against NVIDIA Jetson, Qualcomm, and Horizon Robotics over a 3–5 year horizon as they out-spend you on R&D?
10. What is your realistic long-term operating-margin model at, say, $700M–$1B revenue, and what gets you there besides gross-margin mix?
11. How exposed is the security end-market to further US-China actions, and what's the contingency if a major security customer is restricted?
12. Capital allocation: with ~$278M net cash and minimal buybacks, why not offset SBC dilution more aggressively — or is M&A/independence optionality the reason to hoard cash?
13. What share of revenue is now genuinely "edge-AI inference" at premium ASPs vs. legacy video, and how fast is the legacy tail declining?
14. Founder and CFO insider selling has been consistent with zero buys. How should investors read that against your stated long-term conviction?
15. What is the single risk on the 18-month horizon you worry about most that the Street is not focused on?

---