Ambarella
Fabless chip design: sells system-on-chip (SoC) processors + software toolchain, manufactured at external foundry (Samsung for advanced nodes; TSMC historically); revenue overwhelmingly via a Taiwan sales-rep/fulfillment partner (WT Microelectronics) serving Asian ODM/OEM customers. Design-win driven, multi-year automotive cycles.
Earnings, margins, COGS & capex
Ambarella earns by selling edge-AI vision SoCs (CVflow architecture) plus its software/dev toolchain. FY2026 was a strong recovery year (+37.2%) off a cyclical FY2025 trough, led by higher-priced AI-inference SoCs - the large majority of revenue - across automotive, security/IoT, and nascent robotics. Gross margin sits high-50s to low-60s%; the company is non-GAAP profitable but GAAP-unprofitable due to large R&D and stock-based compensation relative to a ~$390M revenue base. FY2027 has decelerated (Q1 +16.9% YoY; Street models ~10-15% full-year) as consumer IoT softens while automotive hits records.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~41¢ is cost of goods and ~0¢ operating expense, leaving ~59¢ of operating profit (~-19¢ net).
Revenue trend
Margins
down ~130bps YoY (from 60.5%) on mix
down ~200bps YoY (from 62.7%)
down from 60.0% in Q1 FY2026 on mix
down from 62.0% in Q1 FY2026; guided ~59.0-60.5% for Q2
loss, driven by SBC + R&D; narrowed from $117.1M loss in FY2025
positive, up from a $6.8M non-GAAP loss in FY2025
COGS structure
COGS is primarily foundry wafer + assembly/test/packaging cost (fabless - Samsung is the named advanced-node foundry partner: 5nm for current flagship SoCs, 2nm for the next-generation edge-AI platform; TSMC supplied earlier nodes). Gross margin high-50s/low-60s% reflects differentiated SoC + IP content; margin gives back ~100-200bps on product mix (more automotive/commercial vs high-margin legacy). No fab depreciation in COGS.
Capex
Minimal - fabless, no wafer fabs; capex is lab/test equipment + IP/EDA tooling. Not a capital-intensive model; cash use in FY2027 has been inventory build (~$34.8M sequential draw in Q1) rather than capex.
Latest earnings
In line to slightly ahead: revenue $100.4M modestly above the ~$100M guide midpoint; non-GAAP EPS $0.11 met-to-slightly-beat consensus. Stock initially dipped post-print on decelerating growth/valuation, then rallied sharply into late June on Rosenblatt's physical-AI top-pick call ($120 PT) + a KeyBanc non-deal roadshow.
Q2 FY2027 revenue $105-111M ($108M mid), with sequential growth in both automotive and IoT; non-GAAP gross margin guided ~59.0-60.5%. The company guides only one quarter out; Street/consensus models imply ~10-15% full-year FY2027 growth (down from FY2026's +37.2%) on consumer-IoT softness.
- Revenue
- $100.4M (+16.9% YoY, -0.5% QoQ)
- GAAP / non-GAAP gross margin
- 58.4% / 59.9%
- Non-GAAP EPS
- $0.11 (non-GAAP net income $5.0M; GAAP net loss $18.1M)
- IoT mix
- ~75% of revenue (enterprise up sequentially, consumer down double digits)
- Automotive
- All-time quarterly record (commercial-vehicle led)
- WT Microelectronics
- 60.7% of revenue (Taiwan sales-rep/fulfillment partner)
- Cash + securities
- $277.8M, no debt
Growth drivers
- Edge-AI inference SoCs (the large majority of FY2026 revenue and rising) - the core secular driver, moving AI compute to cameras/vehicles/robots
- Automotive: ADAS + commercial/fleet vehicle vision, hit an all-time quarterly record in Q1 FY2027 despite seasonality, led by commercial-vehicle telematics + safety
- Enterprise/security cameras (IoT) with on-device AI analytics — enterprise up high-single-digits sequentially even as consumer IoT fell double digits
- Robotics / 'physical AI' / embodied-AI optionality — a next-gen 2nm edge-AI SoC platform positioned for robots, drones, and autonomous machines
- Advanced-node roadmap (Samsung 5nm now, 2nm next) enabling higher AI TOPS/watt vs competitors
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-03-23. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’26
Bull & bear
Ambarella is a scarce pure-play on inference-at-the-edge - the compute layer physical-AI (cameras, cars, robots) actually runs on - with defensible low-power vision IP, records in automotive, a robotics optionality kicker, and a clean net-cash balance sheet; if the edge-AI ramp compounds, the current ~$3.8B cap is early.
- Only meaningful independent, merchant edge-AI vision-SoC pure-play - a natural beneficiary (and takeout candidate) of the physical-AI/robotics buildout
- Automotive at an all-time record and structurally growing as ADAS/commercial-vehicle vision content rises
- Edge-AI inference SoCs are the large majority of revenue, with performance-per-watt advantages GPUs can't match in thermally-constrained sockets
- 5nm/2nm Samsung roadmap keeps it on the leading edge for AI TOPS; robotics/gen-AI-at-the-edge is upside not in numbers
- Net cash ~$278M, no debt, non-GAAP profitable - can fund R&D through cycles; Rosenblatt physical-AI top-pick ($120 PT, 2026-06-30) + KeyBanc interest signal Street enthusiasm into 2H2026
A sub-$400M-revenue chip designer decelerating toward ~10-15% growth, GAAP-lossmaking, ~61% dependent on a single fulfillment partner, and squeezed between Nvidia and Qualcomm - yet trading at ~9-10x sales on a robotics story not yet in the financials.
- GAAP net loss $75.9M FY2026; 'profitability' relies on adding back heavy stock-based comp
- Growth cut from +37% (FY2026) toward ~10-15% (Street FY2027) - the AI-inflection narrative is outrunning the revenue
- 60.7% revenue concentration through WT Microelectronics is a single point of channel failure
- Structurally outgunned by Nvidia (Jetson) and Qualcomm on scale, software, and ecosystem in the same edge/auto sockets
- Consumer IoT is soft and cyclical; automotive revenue lags design wins by years
- Valuation (~9-10x sales, no GAAP earnings) prices in a robotics ramp that is optionality, not yet demand; the 2026-06-30 pop was analyst-note/flow driven, not a fundamental re-rate - high downside on any miss
What it is worth
Peer EV/Sales comp + reverse-DCF sanity check (no 5-tab model). At ~$3.75-3.8B market cap ($85.80 x ~43.8M sh) - ~$3.5B EV net of ~$278M cash - on FY2026 revenue of $390.7M, AMBA trades ~9-10x trailing sales / ~8-9x FY2027E (~$430-450M at ~10-15% Street growth). That multiple prices AMBA well above its ~10-15% near-term growth on a GAAP-lossmaking base - it is an edge-AI/robotics optionality multiple, not an earnings multiple. Reverse-DCF read: the price implies the market underwrites a re-acceleration back toward 20%+ durable growth with meaningful margin expansion (robotics/physical-AI ramp materializing), not the mid-teens now modeled.
~$45-60
growth stays low-teens or slips, consumer-IoT weakness persists, competition from Nvidia/Qualcomm compresses the multiple toward auto-semi peers on a still-GAAP-lossmaking base.
~$80-90 (roughly current)
mid-teens growth holds, non-GAAP profitable, multiple ~8-10x sales - story intact but not yet inflecting.
~$120+ (Rosenblatt PT $120
2026-06-30 physical-AI top-pick call): robotics/physical-AI + automotive re-accelerate growth to 20%+, GAAP profitability inflects, multiple sustained on scarcity/takeout premium.
No P/E anchor (GAAP loss). Comps: MBLY, NXPI trade at far lower sales multiples with scale; AMBA's premium rests on pure-play edge-AI scarcity + M&A optionality. High multiple = high sensitivity to any growth/margin disappointment. Not financial advice.
SWOT
Strengths
- Differentiated low-power computer-vision SoC IP (CVflow) with strong performance-per-watt - a genuine edge over general-purpose GPUs for battery/thermally-constrained cameras and vehicles
- Deep, sticky automotive + security-camera design wins with multi-year lifecycles and high switching costs
- Non-GAAP profitable with a clean balance sheet - ~$278M net cash, zero debt
- Leverage to two durable secular tailwinds: edge-AI inference and automotive vision/ADAS
- Advanced-node access via Samsung (5nm now, 2nm next) keeps its AI-compute roadmap competitive
Weaknesses
- GAAP-unprofitable ($75.9M net loss FY2026) — profitability depends on non-GAAP add-backs (large stock-based comp)
- Severe channel concentration: WT Microelectronics is 60.7% of Q1 FY2027 revenue as the fulfillment partner
- Small scale (~$390M revenue) versus semiconductor peers - limited R&D absorption vs Nvidia/Qualcomm
- Cyclical, inventory-sensitive demand - FY2025 was a trough; consumer IoT remains soft
- Growth decelerating hard (FY2026 +37% to Q1 FY2027 +17%; Street ~10-15% full-year)
Opportunities
- Robotics / embodied 'physical AI' - a large new TAM for edge inference SoCs beyond cameras and cars
- Automotive content growth as ADAS moves down-market and into commercial/fleet vehicles
- On-device generative/agentic AI at the edge driving demand for higher-TOPS vision-language SoCs
- Enterprise security analytics upgrade cycle (AI-native cameras replacing legacy DVR/camera fleets)
- 2nm roadmap could open higher-end inference sockets currently served by Nvidia Jetson-class parts
Threats
- Nvidia (Jetson/edge) and Qualcomm pushing edge-AI/automotive compute with far larger scale and software ecosystems
- Foundry-supply and advanced-node capacity/yield risk concentrated at Samsung
- Automotive design-win losses or delays - long lag between design win and revenue
- China/geopolitics — export controls and Chinese domestic CV-SoC competition (HiSilicon, etc.) in the security-camera market
- Rich valuation (~9-10x sales) leaves little room for a growth or margin miss
Moats, dependencies & bottlenecks
Moats
Low-power computer-vision SoC architecture (CVflow) + software toolchain Moderate-to-strong Real performance-per-watt edge in constrained sockets; but Nvidia/Qualcomm attacking with scale, so durability is not assured.
high switching cost) Long qualification cycles create stickiness once designed in; the flip side is slow revenue realization.
Helps retention but far shallower than Nvidia CUDA-class moats.
Dependencies
5nm now / 2nm next) Supply / manufacturing Fabless - leading-edge capacity and yield at a single named foundry partner gate the AI roadmap.
Distribution / channel 60.7% of Q1 FY2027 revenue flows through this sales-rep/fulfillment partner serving Asian customers - concentration risk.
SoCs embed Arm cores; licensing/royalty terms are a structural dependency.
Chip design depends on the two US EDA majors.
Demand / end-market Cyclical, inventory-sensitive, multi-year design-to-revenue lag.
Advantages
- Best-in-class performance-per-watt for edge computer vision
- Independent merchant supplier - sells to anyone, unlike captive/OEM silicon
- Clean balance sheet (~$278M net cash, zero debt)
- Two secular tailwinds (edge-AI inference + automotive vision) plus robotics optionality
- Established, sticky design-win base in security and automotive
Weaknesses
- GAAP-lossmaking; profitability is non-GAAP only
- Extreme channel concentration (WT Microelectronics 60.7%)
- Decelerating growth (Q1 FY2027 +17% YoY; Street ~10-15% full-year)
- Small scale vs mega-cap edge-AI competitors
- Cyclical consumer-IoT exposure
Bottlenecks
- Advanced-node foundry capacity/yield at Samsung (5nm/2nm)
- Single-channel concentration (60.7% via WT Microelectronics)
- Long automotive design-win-to-revenue lag limits how fast bookings convert to sales
- Sub-$400M revenue scale caps R&D spend vs Nvidia/Qualcomm
- GAAP unprofitability + high stock-based comp constrain reinvestment flexibility
Top signals & trends
Top signals
Q1 FY2027 - validates the auto/ADAS content-growth thesis.
Consumer-IoT softness the main drag; company guides only one quarter out.
Keeps AI-compute roadmap leading-edge; long-term competitive.
Sentiment/flow-driven, NOT an M&A or fundamental event - raises valuation risk.
Routine-scale but worth noting immediately after the sharp rally.
Trends
Core structural driver of Ambarella's SoC demand.
Rising vision-processor content per vehicle, incl. commercial/fleet.
Positive (optionality) · Large future TAM for edge-inference SoCs; not yet material in revenue.
Competitive encroachment on Ambarella's sockets from far larger players.
Pressure in the security-camera market and geopolitical supply risk.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Named advanced-node manufacturing partner (5nm current, 2nm next-gen).
Historical foundry supplier for prior-node Ambarella SoCs.
CPU IP licensed into Ambarella SoCs (SoftBank-owned).
EDA design tools + IP.
EDA design tools + IP.
Taiwan sales-rep/fulfillment partner - 60.7% of Q1 FY2027 revenue, fulfilling to Asian OEM/ODMs.
Automotive Tier-1s integrating vision SoCs into ADAS/camera modules.
Enterprise/security-camera customer for AI-analytics cameras.
Network-security-camera maker in the enterprise IoT segment.
Jetson edge/robotics modules + automotive DRIVE - the scale threat with the dominant AI software ecosystem.
Automotive (Snapdragon Digital Chassis) + edge-AI compute; large-scale, deep OEM relationships.
ADAS/autonomy vision SoCs (EyeQ) - direct automotive-vision competitor.
Automotive + edge processing; broad auto franchise.
Taiwan video/security-camera SoC competitor in IoT/surveillance.
Automotive processing + imaging/sensing adjacencies.
Dominant Chinese security-camera SoC supplier - competitive context only, not an ownable name.