42 million chips deployed. Earnings Thursday May 28. Binary event.

Edge AI is the next leg of the compute buildout. AMBA is the only US-listed pure-play.

Wall Street has spent two years pricing the AI trade through data center silicon: NVIDIA for GPUs, Broadcom for custom ASICs, AMD for accelerators. The next leg of AI compute is happening at the edge — vision systems in cars, robots in warehouses, drones overhead, security cameras everywhere — and that requires a fundamentally different chip architecture. Ambarella is the only US-listed company built from the ground up for this exact use case, with 42 million AI SoC units already in the field, automotive design wins through Continental, and a 2nm chip taped out in March 2026. The setup is a real growth story trading at a chip-cycle valuation. The catalyst is Thursday's earnings print — May 28 after market close.

Not every AI workload can run in a data center. Some have to run on the device itself.

Every conversation about AI in 2026 starts with NVIDIA and ends in a data center. That's where models are trained. That's where ChatGPT, Claude, and Gemini run. But there's a second AI compute market that's structurally different: workloads that have to run on the device itself, in real-time, often without internet. Your car can't wait 200ms for a cloud round-trip to decide whether to brake. A warehouse robot can't pause to ask AWS what it sees. A drone can't stream 4K video to a server while flying a mission. These workloads need AI chips that live on the edge — and those chips have completely different requirements than data center GPUs.

The Edge AI Chip Problem

Data center GPUs like the NVIDIA H100 cost $30,000, draw 700 watts of power, and need active cooling. You cannot put a 700-watt heater in a Tesla. You cannot put a 700-watt heater in a Roomba. Edge AI chips need to deliver real AI inference performance while:

• Running on a tiny power budget (often under 10 watts total)
• Operating at -40°C to +105°C without a fan
• Costing $10-100 per chip, not $30,000
• Processing multiple camera/sensor streams simultaneously
• Doing it all with single-digit milliseconds of latency

This is an extremely different problem than data center training. The chip architecture, the manufacturing process, the software stack — all different.

What Makes Ambarella Specifically Different

AMBA designs custom SoCs (System-on-Chip) where the AI accelerator, image signal processor, video codec, and CPU all live on one piece of silicon. Their entire product family is purpose-built for this category — they don't try to compete with NVIDIA in data centers and never have. The pitch to customers is simple: "you can use NVIDIA Jetson and get 10x the power consumption, or you can use AMBA and design a product that fits in a phone."

Three things give AMBA structural advantages in this category:

• CVflow architecture — purpose-built for vision AI, more efficient per watt than general-purpose GPUs at vision workloads
• Integrated ISP — handles raw camera data on-chip, eliminating need for separate processing
• Software stack — Cooper Developer Platform supports the entire popular ML model ecosystem (PyTorch, TensorFlow, etc.) compiled to run efficiently on AMBA silicon

The Product Generation Roadmap

The N1 Wild Card

The newest product line — N1 — gets the least attention but matters the most for the bull thesis. N1 chips target edge AI infrastructure: AI vision boxes, perception gateways, on-prem inference appliances. The first proof point: a warehouse robotics customer using N1-655 to run a fleet of perception hubs. This moves AMBA from "security camera silicon vendor" to "physical AI platform company." One enterprise customer is a proof point. Two or three named customers turns N1 into a credible edge infrastructure revenue line that justifies a higher multiple. Watch for N1 customer disclosures on Thursday's call.

Customer wins are stacking. But the insiders are selling, not buying.

AMBA's validation comes in two forms — and they're sending mixed signals. On the product side, customer wins are accumulating across automotive, robotics, drones, and edge infrastructure. Analyst price targets average $114.90 versus a $75 spot price. On the insider side, the picture is uglier: 23 sales by insiders over the past six months, zero purchases. This is the bear flag every reader should know about going into the trade, and it's the reason this thesis is sized as a tactical earnings play rather than a high-conviction long.

Tier 1 — Customer & Partnership Validation

Tier 2 — Analyst Coverage

Coverage has been consistently bullish on the structural edge AI thesis. The reduction in average PT from $115 to $114.90 is essentially flat — the Street has not lost conviction on the long-term setup. The variable is execution: can AMBA deliver the Q1 FY27 print Thursday at the high end of guidance, and what does the call say about CV7 customer pipeline + N1 expansion?

Tier 3 — The Insider Selling Bear Flag

Here is the side of the trade nobody pumping this stock is going to tell you about. Over the past six months, AMBA insiders have made 23 open-market transactions — all sales, zero purchases:

• CEO Feng-Ming Wang sold approximately 82,009 shares across 8 separate transactions, total proceeds ~$4.83M.
• Multiple senior officers including the Senior VP and other named executives also selling.
• Zero offsetting purchases by any insider.

What this likely is: Insider selling at chip companies is often part of 10b5-1 plans — pre-scheduled automatic sales that don't reflect timing-specific bearishness. CEOs with significant share grants regularly diversify via these plans. What this could be: Genuine insider read that the next 6-12 months are less attractive than the stock currently prices. Best practice with this signal: not a reason to short, but absolutely a reason to size the trade tactically rather than aggressively long. The "trust the insiders" reading of a stock works both ways.

The Read

The customer validation is real. The analyst PT spread (+52%) is genuine. The product roadmap (CV7, CV3, N1, 2nm) is concrete and progressing. But the insider selling is a flag worth respecting — especially heading into a binary earnings event. This is not the stock to bet the farm on. It is the stock to take a defined-risk earnings position on, with the structural multi-quarter thesis as the back-end optionality if the print delivers.

22 years of vision silicon. Repositioned from "security cameras" to "physical AI."

Ambarella, Inc. (NASDAQ: AMBA). Founded 2004 by Feng-Ming "Fermi" Wang (still CEO) and Les Kohn (still CTO). Headquartered in Santa Clara, CA. ~915 employees. The company started as a video compression silicon vendor — early growth came from action cameras (GoPro was a huge customer) and IP security cameras. Beginning around 2018-2020, AMBA executed a deliberate strategic shift away from low-margin video silicon toward AI-enabled SoCs for vision applications. That pivot is now visibly showing up in the numbers: edge AI SoCs are 80% of revenue. Market cap ~$3.2B. ~42M shares outstanding.

Four Product Lines

The Pivot From "Cameras" To "Physical AI"

Five years ago, AMBA was primarily known for compression silicon and IP security cameras. Today, "edge endpoint and edge infrastructure" applications — automotive, robotics, drones, edge AI inference appliances — drive the growth. Q4 FY26 was the first full quarter of production revenue from aerial drones. The 3rd-generation chips (CV72/CV75 at 5nm) already reached high single-digit percentage of total revenue in Q4 FY26 — proof that the higher-ASP mix shift is working. The story is moving from "we sell to security camera OEMs" to "we sell to anything that needs to see and decide in real-time."

The Software / Platform Layer

What most retail investors miss: AMBA is not just a chip vendor. The Cooper Developer Platform is a full software stack that lets customers compile and deploy popular ML models (PyTorch, TensorFlow, etc.) onto AMBA silicon efficiently. This makes the chip stickier — once a customer trains their perception model and tunes it for AMBA's hardware, switching costs become real. Bulls call this "full-stack edge AI platform." Bears call it "marketing for a chip company." The reality is in the middle, but the platform pitch is real enough to win attention at analyst days.

The CV3-AD Automotive Bet

The CV3-AD platform is AMBA's biggest single strategic bet. It's a family of SoCs designed for ADAS (Advanced Driver Assistance Systems) and autonomous driving applications. The flagship CV3-AD685 has 16-core Cortex-A78AE architecture optimized for safety-critical workloads. Continental — one of the largest tier-1 auto suppliers globally — uses CV3-AD as a core platform. Continental spun off its automotive electronics business as Aumovio in 2025, and CV3-AD continues into that entity. Auto design wins are slow to materialize but extremely sticky once they ship — a chip designed into a 2027 model year vehicle ships for the 5-7 year production run of that platform.

The 2nm Roadmap Signal

In March 2026, AMBA taped out its first 2nm gate-all-around (GAA) chip. This is meaningful for two reasons. First, very few companies tape out at 2nm — the cost is enormous (>$200M per design) and only the biggest customers justify the spend. AMBA committing to 2nm means they have customers willing to fund it. Second, gate-all-around is the next-generation transistor architecture replacing FinFET — pioneered by Samsung and being adopted by TSMC for the 2nm node. AMBA is leveraging the same leading-edge process tech as Apple, NVIDIA, AMD. That is not "small chip company" tape-out behavior. That's a structural bet that edge AI silicon needs the same process leadership as data center silicon.

FY26 revenue +37% YoY. FY27 guide is the question Thursday answers.

FY26 Full-Year Results (Fiscal Year Ended Jan 2026)

Q1 FY27 Guidance (For Thursday's Print)

The Decel Question Is The Trade

This is the single most important number in this report. FY26 grew +37%. FY27 is guided to +10-15%. That is a meaningful deceleration. The two competing interpretations:

Bull interpretation: Conservative guide. Continental CV3-AD ramps in H2. CV7 starts contributing revenue in Q4 FY27. N1 customers add through the year. Management is sandbagging so they can beat-and-raise across the year. Final FY27 lands at +20-25% growth.

Bear interpretation: The pull-in cycle into Q4 FY26 was unusually strong (drone customers ramping, third-gen chip launches), and FY27 is the comp-year challenge. Edge AI demand is real but not as fast as 2025. Final FY27 lands at +10% — and the multi-quarter rerate that bulls expect doesn't happen until FY28.

The Thursday print resolves this. If Q1 prints at the high end ($103M) and management raises full-year guide on the call — bull case wins. If Q1 prints mid-range and full-year guide stays at +10-15% — the bear interpretation gets validated and the stock likely falls.

The Valuation Math

At $75.49 spot, ~$3.2B market cap on FY27 revenue ~$440M (midpoint guide), AMBA trades at ~7x forward revenue. For a small-cap chip company growing in the 10-37% range with edge AI exposure and automotive design wins, this is not statistically expensive — comparable AI chip names trade higher:

• NVDA: ~25x revenue (different category, but the benchmark)
• AVGO: ~17x revenue (custom ASIC + networking)
• QCOM: ~5x revenue (mature, lower growth)
• MBLY: ~7x revenue (direct auto comp)

AMBA at 7x revenue is roughly in line with Mobileye, which is the closest pure-play comp. The multiple-expansion thesis depends on AMBA proving it can sustain 20%+ growth into FY28 — not the +10-15% currently guided for FY27.

Analyst Coverage Detail

The PT has been stable in the $114-115 range for multiple quarters. The Street has not gotten more bearish despite the modest FY27 growth guide. That's actually a tell: the analyst community is willing to look past the FY27 decel because the longer-term thesis is intact. The setup is "do you trust the long-term story or do you trade the short-term execution print."

NVIDIA owns data center. Mobileye owns highway autonomy. Where does AMBA fit?

Edge AI silicon is a fragmented market with very different competitors in each sub-vertical. AMBA does not compete with NVIDIA in data centers — that's a different category. AMBA competes with Mobileye in automotive vision, with Qualcomm in connected vehicles, with Horizon Robotics in China, and with NVIDIA's Jetson family in edge robotics. Each battle has a different winner, and AMBA is genuinely competitive in some categories while structurally disadvantaged in others.

The Three Battles AMBA Is Fighting

AMBA vs Mobileye (MBLY) — The Direct Auto Comp

Mobileye is the giant in this category. ~$20B revenue base, hundreds of millions of EyeQ chips shipped, the dominant supplier for L2 ADAS globally. Mobileye is to automotive vision what NVIDIA is to data center training. AMBA's CV3-AD is positioning to compete in the next-generation L2+/L3 segment where Mobileye's EyeQ6 dominates. Different positioning: Mobileye's pitch is "full-stack solution — chip, software, mapping, all bundled." AMBA's pitch is "open platform — bring your own software stack, we provide best-in-class silicon and developer tools." OEMs choosing AMBA over Mobileye are typically prioritizing software flexibility over end-to-end integration. Continental's choice was the marquee win for this approach. AMBA could realistically capture 5-15% of the ADAS chip market by 2030 even with Mobileye dominant — but that's enough to materially grow revenue from the current base.

AMBA vs NVIDIA Jetson — The Robotics Battle

NVIDIA Jetson (Orin, Thor, AGX) is the default platform for robotics development. The advantages: CUDA ecosystem, software maturity, brand recognition, generous power budget for performance. The disadvantage: Jetson AGX Thor draws 100+ watts. For battery-powered drones, mobile robots, or any system where power matters, that's a non-starter. AMBA wins where power matters more than absolute compute. A drone running on a 30-minute battery cannot afford a 100W chip. A warehouse robot picking 24/7 can't afford the heat. This is where N1 fits — same software flexibility as Jetson, dramatically lower power. The trade-off is less raw AI compute per chip, but for many use cases that's fine.

The China Threat — Horizon Robotics

This is the most underappreciated risk to the AMBA thesis. Horizon Robotics — a Chinese ADAS chip company — controls roughly 49% of the Chinese self-driving chip market, exceeded NVIDIA in NoA (Navigate on Autopilot) installations in 2023, IPO'd on the HKEX in 2024, and is backed by Volkswagen with a $2.3B joint venture. For the Chinese auto market, Horizon is structurally favored due to local sourcing pressure and government policy. AMBA is essentially locked out of meaningful share in China automotive — and China is the largest single auto market globally. This is a real ceiling on the AMBA bull case that bulls tend to gloss over.

AMBA vs Qualcomm Snapdragon — The Cockpit Battle

Qualcomm has been aggressively expanding into automotive via Snapdragon Digital Cockpit platforms. Their pitch: "we already have the modem and connectivity, we can integrate vision and ADAS into the same SoC stack." This is a real competitive threat for AMBA in the future-vehicle category where ADAS and infotainment converge. But QCOM's edge AI vision performance per watt remains behind dedicated solutions like Mobileye and AMBA. The bet is whether OEMs prefer "one chip from QCOM that does everything mediocrely" or "specialized chips from multiple vendors that do specific tasks excellently."

The Bottom Line On Competition

AMBA is not a category-killer in any sub-segment. They're a credible #2 or #3 in multiple high-growth categories. The bull case is that "credible #3 across 5 fast-growing edge AI categories" can compound into a meaningful business at this market cap. The bear case is that being #3 in every category eventually loses to whoever's #1 in each. The honest read: AMBA has enough differentiation to win specific design slots, but is not a dominant platform. The math works at $3.2B market cap; it might not work at $10B.

Real bull case. Real bear case. Sized as a tactical earnings play, not a high-conviction long.

Binary print Thursday May 28. Two-way move likely 10-15%.

The Thursday Binary — How To Trade It

Earnings are Thursday May 28 after market close. The implied move from current options pricing is approximately 10-15%, meaning the market is pricing a $65-87 range post-print. Three scenarios are realistic:

• Beat + Raise: Q1 prints $101-103M (high end), FY27 guide gets nudged up to +15-20%, Continental or N1 customer disclosed. Stock gaps to $85-95. Probability: ~30%.
• In-Line + No Raise: Q1 prints $99-101M (midpoint), FY27 guide reaffirmed at +10-15%, no new customer disclosure. Stock holds $73-80. Probability: ~45%.
• Miss / Negative Tone: Q1 prints below $97M or management cautious on FY27. Stock drops to $60-68. Probability: ~25%.

Position Structure — Three Approaches

Approach 1 — Wait For The Print (Recommended). Don't trade into binary earnings without an edge. Wait for Thursday's print, then add equity in the $73-78 range if the in-line scenario plays out, or chase modestly on a clean beat-and-raise.

Approach 2 — Defined Risk Through Earnings. Buy June 6 or June 20 expiry call spreads (e.g. $80/$95 call debit spread) to define max loss into the print. This caps downside to the premium paid while capturing upside if the print lands well. Best risk-reward for actually trading the catalyst.

Approach 3 — Equity + LEAPs. If conviction is on the multi-quarter thesis rather than the print specifically, equity at current levels with Jan 2027 $90-100 LEAPs lets you survive a one-quarter miss while keeping the back-half thesis alive. Higher conviction setup.

Why Defined Risk Matters Here

Three factors specifically argue for defined-risk structure on this trade:

• Insider selling overhang: 23 sales / 0 purchases over six months. Insiders may know something the market doesn't yet.
• FY27 decel guide: +10-15% vs +37% prior year is a real concern. Bears can validly point to this.
• Small-cap chip volatility: AMBA has historically gapped 15-25% on earnings prints. Binary outcome distribution is wide.

This is not the trade to put on with naked stock at full sizing into Thursday. It's the trade to either wait out, define-risk through, or position for the longer-term thesis with sleeve sizing.

Catalyst Calendar

Smallest cap of the AI chip pure-plays. Real growth, real product, real volatility.

The Three Comps That Matter

AMBA vs MBLY: Direct automotive vision comp. Both at ~7x EV/Revenue. MBLY is the dominant L2 supplier with hundreds of millions of EyeQ chips shipped; AMBA is the up-and-coming L2+/L3 player. MBLY trades at the same multiple as AMBA despite being 6x larger and profitable — which actually argues that AMBA could re-rate higher relative to MBLY if growth re-accelerates beyond the +10-15% FY27 guide. The auto vision category supports two multiples: dominant share (MBLY) and high growth (AMBA). Both can win at different price points. AMBA vs SYNA: John already holds SYNA. The two names operate in adjacent edge processor markets but with different end customers — SYNA is heavier in connectivity / IoT consumer endpoints, AMBA is heavier in vision / industrial / auto. The interesting cross-read: SYNA is profitable, growing at 8%, trades at 3x revenue. AMBA is unprofitable, growing at 37%, trades at 7x revenue. The market is pricing AMBA's growth premium correctly — but if the growth decel is real, AMBA could compress toward SYNA's 3x multiple ($35-40 stock). AMBA vs NVDA: Not really competitors except in robotics (NVIDIA Jetson vs AMBA N1). NVDA is the giant; AMBA is the specialist where Jetson's power profile doesn't fit. Reading NVIDIA's edge AI commentary on their quarterly calls is the best macro tell for AMBA — when NVIDIA talks up Jetson and edge, the secular thesis is supported. When NVIDIA goes quiet on edge, it's a tell.

Wait for Thursday's print. Don't fight a binary catalyst with naked stock.

AMBA is a real edge AI chip company with real product traction, real customer wins, and real growth — but it is also two days away from a binary earnings event with insider selling overhang and a deceleration in guided growth. The smart move is not to chase into Thursday. Wait for the print, watch the call for Continental CV3 specifics, CV7 customer disclosures, and N1 expansion commentary, then enter on the reaction. If the print delivers and guidance gets raised, the bull case is real and entry in the $80s is acceptable for a multi-quarter setup. If the print misses or guide stays soft, the stock retraces toward $65 and that's a more attractive entry for the longer thesis. Defined-risk options structures (call debit spreads) are the right way to play it for traders who want to express a directional view through the binary event. Equity bulls should wait. Options traders can structure tactically. Either way, don't size aggressively into a print you can't predict.