Ambarella Bets Big on Edge AI Infrastructure and Samsung 2nm — But Auto Revenue Remains Elusive
Morgan Stanley Technology Conference, March 3, 2026 — CEO Fermi Wang outlines the next phase of the company's edge AI pivot
Ambarella CEO Fermi Wang used his appearance at the Morgan Stanley Technology Conference to lay out what is arguably the most consequential strategic shift in the company's history: a move from selling chips into individual edge endpoints toward enabling an entirely new category of AI-powered edge infrastructure boxes. Combined with new disclosures on 2-nanometer production timelines at Samsung, a first warehouse robotics win, and a candid reassessment of the automotive opportunity, the session offered investors considerably more signal than a typical conference appearance. The stock's recent turbulence, driven partly by an Insta360 lawsuit scare that Wang quickly dismissed as having no fundamental impact, gives additional context to what management is actually building toward.
The Edge AI Infrastructure Box: A New Revenue Category Investors Should Track
The most significant new concept Wang introduced is what Ambarella is calling its "edge infrastructure" business — a dedicated AI box that sits not in a data center but physically at the edge, aggregating multiple sensor inputs and running generative AI models against them. This is a departure from the company's traditional model of selling into individual camera chips. Wang illustrated the concept with a retail demonstration from CES: "They collected the security camera feed to this AI box at a retail store. And with the new GenAI, they can turn this security camera into operational tools that can monitor how the customer comes in, what they buy and collect the customer data. So suddenly, security camera becomes your operational tool for your efficiency."
The commercial logic is straightforward — there is a massive installed base of video infrastructure in retail, logistics, and enterprise settings that has never been monetized beyond basic security. Ambarella's proposition is that a relatively modest incremental investment in an AI box can unlock that data for operational use. The go-to-market model, however, depends heavily on third-party independent software vendors and system integrators, which introduces execution risk. Wang acknowledged this directly: "For each application, you need a software vendor to write software on that. We need to have a system integrator to integrate that box into existing infrastructure of the retail stores." Investors should watch how quickly that partner ecosystem scales.
Cooper SDK Portability Is Accelerating Customer Adoption Faster Than Expected
One of the more concrete data points from the session was the speed at which ISVs are now moving onto Ambarella's platform. Wang described how a software vendor given access to the Cooper SDK just three months before CES was able to port its application from a competitor's platform "in less than a few weeks" and was already demoing at the show. Wang was explicit that this was a deliberate five-year engineering investment: "That takes time for us to build the extra layer so to separate out the hardware layer, but keep all the software layer open so that our customer doesn't need to spend a lot of time to port." Morgan Stanley's Joe Moore noted that while Continental and Bosch showed similarly fast productization, converting that into revenue has historically taken years — a caution that applies here as well.
CV7 Design-In Activity Is Described as "Huge" Across Multiple Applications
Wang offered an unusually direct characterization of demand momentum for the CV7, Ambarella's first 4-nanometer chip. He noted that CV7 delivers 2.5 times the AI performance of CV5 — itself a 5-nanometer product — and that customer requirements for computational headroom are escalating rapidly, driven by both application complexity and the growing footprint of generative AI models at the edge. "I can say for sure that our design-in activity with CV7 is huge in terms of multiple different applications," Wang said. This stands in notable contrast to the more cautious fiscal year 2026 guidance the company issued at the start of last year, before a year that ended up delivering 37% revenue growth versus initial high-teens expectations.
Samsung 2nm: Production First Half of 2027, Capacity Already Secured Through the Year
Wang provided the clearest update yet on the company's first 2-nanometer chip, originally taped out at Samsung at the end of last year. He confirmed a production target of first half of 2027 and — more importantly — stated that capacity at Samsung has been secured not just for 2026 but also for 2027. The Samsung choice is a calculated bet, and Wang acknowledged that confidence in the process has grown materially over the past twelve months of close collaboration on yield. He also made no secret that Elon Musk's decision to put significant volume through Samsung provided meaningful external validation: "When Elon announced it, I felt so happy because I don't need to be the only one defending it anymore." With TSMC signaling capacity constraints, the Samsung relationship is increasingly a competitive differentiator on supply chain reliability.
The semi-custom ASIC model Wang discussed is directly tied to this 2-nanometer program. A customer approached Ambarella about co-funding a portion of the tape-out cost in exchange for some customization, with Ambarella retaining the right to sell the chip to non-competing customers. Wang described significant inbound interest from additional customers on similar terms since that initial arrangement became known, and indicated the company is now evaluating two to three engagements to size the ROI before committing to it as a formal business line. The key qualifying criteria are whether prospective customers want to leverage Ambarella's video processing IP, AI inference engine, low-power architecture, and 2nm process — all together. This is not a foundry-services play; it is an IP licensing and co-development model layered on top of the chip business.
Automotive: A Deliberately Narrower Focus, but a Growing Identified Opportunity
Wang's tone on automotive was notably more disciplined than in prior years. The company disclosed $13 billion in total identified automotive opportunity over the next six years at its most recent earnings — a figure Wang said has grown year over year — but he was explicit that the internal posture has shifted: "For automotive, let's focus on the project that we know is going to generate revenue. That is a huge change for us." The CV3 program consumed substantial R&D with limited commercial return, and management appears to have internalized that lesson. On the lost Volkswagen deal, Wang confirmed it came down to a competitor offering financial concessions rather than a technology shortcoming, and said the process gave Ambarella broader visibility into the OEM RFQ pipeline: "Almost all the new RFQ and RFI tend to invite us to bid." The company will not be flagging specific wins until revenue materializes, having learned from the VW experience.
Robotics: Perception Box Win Confirmed, Full Domain Controller Is the Long Game
Ambarella disclosed a warehouse robotics design win in its most recent earnings call, and Wang provided additional color at the conference. The win involves a "perception box" — the second of three robotics product tiers Wang outlined — that aggregates multiple camera feeds and other sensor inputs to perform sensor fusion and environmental perception. He was careful to distinguish this from the more ambitious domain controller concept, which would manage perception, path planning, and movement control on a single chip and represents the architecture most analogous to CV3. Wang also addressed humanoid robots directly and with appropriate skepticism: "Humanoid is more complicated than Level 4 cars. Level 4 car drives in an environment that is well controlled. Humanoid, you have no limit working environment." The implication is that near-term robotics revenue will come from industrial and logistics applications, not from the humanoid projects attracting most of the industry attention.
Fleet Management and Wearable Cameras Emerging as Underappreciated Growth Vectors
Wang singled out fleet management — citing Samsara as Ambarella's largest customer in that category — as a market experiencing rapid AI adoption that the investment community has not fully appreciated. The use case has expanded well beyond GPS tracking to include driver monitoring, vehicle condition assessment, and cargo management, all enabled by AI-capable cameras and telematics integration. He also described a quiet but growing shift in wearable cameras from law enforcement toward retail customer service environments, driven by demand for documented interactions between staff and customers. Neither market generates the headline excitement of humanoids or autonomous vehicles, but both represent near-term volume with meaningful AI content.
Insta360 Lawsuit Has No Revenue Impact
Wang addressed the Insta360 patent lawsuit that rattled the stock on earnings night with characteristic directness. He confirmed that Insta360 made a public announcement that the lawsuit has no impact on their operations, and therefore no impact on Ambarella. "They obviously have done a lot of work to work around the patent involving the product," he said. The concentration risk in portable video remains a legitimate concern Wang acknowledged separately — not because of litigation, but because relying heavily on a single customer in that category is a structural vulnerability the company is actively working to dilute by expanding into adjacent AI applications.
Ambarella, Inc. Deep Dive
The Business Model: Monetizing the Algorithmic Edge
Ambarella operates as a fabless semiconductor designer specializing in low-power, high-definition video compression, image processing, and computer vision systems-on-chip. Historically recognized as the silicon engine behind consumer action cameras and early drones, the company has executed a structural pivot toward physical artificial intelligence. Today, Ambarella generates revenue by selling highly integrated proprietary silicon and accompanying software stacks optimized for edge artificial intelligence inference. The company monetizes the algorithmic edge, where computational perception occurs directly on the device rather than in a remote data center. End markets are cleanly divided into two segments: Internet of Things, which currently accounts for approximately 75% of revenue, and Automotive, which makes up the remaining 25%. The Internet of Things segment is driven by enterprise security cameras, robotics, and industrial automation, while the automotive segment supplies advanced driver assistance systems, autonomous driving domain controllers, and commercial fleet telematics. Through its proprietary architecture, Ambarella provides the essential hardware-software combination required for machines to perceive, map, and navigate their physical environments in real time.
Customers, Competitors, and Market Share Dynamics
The customer base reflects a complete geopolitical and strategic transition. In previous years, Ambarella derived substantial volume from Chinese surveillance giants Hikvision and Dahua, which together dominate the global video surveillance market. Following stringent United States export controls, Ambarella forcefully pivoted toward Western and allied enterprise manufacturers. Today, its Internet of Things customer base is anchored by Motorola Solutions, Axis Communications, and South Korea-based Hanwha Vision. This transition culminated in late May 2026 with a landmark agreement with Hanwha Group, estimated at over $800 million in potential revenue across a ten-year span, embedding Ambarella silicon across Hanwha security, robotics, and life science divisions. In the automotive vertical, Ambarella supplies Tier-1 integrators including Continental and Bosch, as well as autonomous trucking pioneers like Kodiak Robotics and electric vehicle manufacturers such as Rivian and Lotus.
The competitive landscape is fiercely contested, pitting Ambarella against significantly larger and more capitalized semiconductor incumbents. In automotive computing, Nvidia dominates the high-end, centralized computing architectures with its premium-priced Orin and Thor platforms, holding a dominant position in compute-heavy robotaxi architectures. Qualcomm leverages its mobile supremacy to aggressively price its Snapdragon Ride platforms, pursuing mass-market Level 2 and Level 3 deployments. Mobileye relies on a vertically integrated black-box approach, bundling proprietary perception software with its EyeQ processors. Within the Internet of Things and security camera domain, Ambarella faces direct competition from Qualcomm, Novatek, and Huawei HiSilicon. Despite this intense competition, Ambarella has successfully carved out an estimated 20% market share within the premium enterprise security camera system-on-chip market, where advanced edge inference is an absolute prerequisite.
On the supply side, Ambarella operates a conventional fabless model, relying entirely on Taiwan Semiconductor Manufacturing Company for wafer fabrication. The company has migrated its advanced computing platforms to 5nm nodes and is currently mapping its next-generation architecture to 2nm geometries. Memory components and peripheral integrated circuits are sourced from standard global suppliers such as Samsung. This supply chain centralization ensures access to bleeding-edge lithography but naturally exposes the company to the systemic geopolitical risks inherent in Taiwanese semiconductor manufacturing.
The Competitive Moat: Silicon Efficiency and Architecture
Ambarella defends its market position through a distinct architectural philosophy centered on algorithm-first silicon design. Unlike general-purpose graphics processing units, which were originally designed for parallel rendering and subsequently adapted for artificial intelligence, Ambarella builds custom accelerators explicitly tailored to specific neural network workloads and perception tasks. This translates into a formidable competitive advantage: superior performance-per-watt. Power efficiency is a critical constraint at the computing edge. In electric vehicles, massive power-hungry central computers directly degrade battery range and require complex, expensive liquid cooling loops. In enterprise security and portable robotics, thermal constraints dictate strict power ceilings. Ambarella systems-on-chip routinely deliver hundreds of tera-operations per second within single-digit watt power envelopes.
Furthermore, the integration of image signal processing, artificial intelligence inference, and video encoding on a single piece of silicon reduces system complexity and bill-of-materials costs for original equipment manufacturers. Ambarella has also achieved a competitive edge in sensor fusion. Its automotive domain controllers feature native, centralized processing for high-resolution cameras, lidars, and 4D imaging radars. By processing raw, uncompressed radar data early in the perception pipeline, the architecture avoids the latency and data loss associated with traditional distributed computing, resulting in higher fidelity path planning in adverse weather conditions. This specialized optimization provides a structural moat that general-purpose computing platforms struggle to replicate without sacrificing thermal efficiency.
Industry Dynamics: Opportunities and Structural Threats
The structural shift from cloud-centric artificial intelligence to edge-native computing presents a massive secular tailwind for Ambarella. Connected devices generate petabytes of high-definition visual data daily. Transporting this raw data to centralized cloud servers incurs prohibitive bandwidth costs, introduces unacceptable latency for mission-critical applications like driving, and raises severe privacy concerns. The industry solution is to execute neural network inference locally, directly on the sensor node. As advanced driver assistance systems scale from premium features to regulatory mandates across Europe and North America, the demand for automotive-grade perception silicon is inflecting upward. The commercial vehicle telematics market is also experiencing a rapid upgrade cycle from passive recording to active artificial intelligence-driven driver monitoring and fleet management.
However, the industry dynamics present significant structural threats. The automotive design win cycle is notoriously protracted, often requiring three to five years from initial engagement to volume production, stressing cash flows during the development phase. Furthermore, original equipment manufacturers are increasingly demanding modular software-defined vehicle architectures, forcing hardware vendors to decouple their silicon from proprietary perception software. While Ambarella supports open software stacks, the commoditization of the underlying compute layer remains a constant threat. Gross margin compression is an ongoing reality; despite high average selling prices for advanced artificial intelligence chips, non-GAAP gross margins have normalized near 59.9%, reflecting the pricing pressure exerted by automotive Tier-1 suppliers and the aggressive discounting tactics of deep-pocketed competitors eager to buy market share.
Next-Generation Catalysts: Transformers and Edge GenAI
Ambarella is actively deploying new silicon architectures designed to capitalize on the emergence of generative artificial intelligence and vision-language models at the edge. The newly introduced N1 system-on-chip family represents a strategic expansion beyond traditional endpoints into edge infrastructure and on-premise computing appliances. This technology allows enterprise clients to run multi-modal large language models locally, enabling natural language querying of video feeds without cloud connectivity. Security personnel, for instance, can prompt the system to locate specific activities or individuals using free-text searches natively processed by the local hardware.
In the automotive sector, the rollout of the CV3 platform brings hardware acceleration for transformer neural networks, a crucial leap beyond traditional convolutional neural networks. Transformers enable birds-eye-view processing and superior spatial awareness, which are prerequisites for Level 3 and Level 4 autonomous driving. By delivering these new multi-modal and transformer-capable chips, Ambarella expands its addressable market from basic edge endpoints to sophisticated edge servers, robotics controllers, and centralized automotive domains, fundamentally increasing the lifetime value of its design wins and accelerating the growth of its average selling prices.
Disruptive Entrants in Automotive Computing
The lucrative nature of automotive artificial intelligence has attracted highly capitalized new entrants deploying disruptive, localized strategies. In the pivotal Asian market, domestic Chinese semiconductor designers such as Horizon Robotics and Black Sesame Technologies represent credible, aggressive threats. Horizon Robotics has successfully combined massive state-backed capital with an open software ecosystem, securing deep integration with leading electric vehicle manufacturers. These entrants benefit from domestic substitution policies and offer highly competitive pricing tailored to the rapid development cycles of the Chinese automotive industry, effectively challenging Western silicon providers for mass-market advanced driver assistance system deployments. Additionally, early-stage entrants focused on neuromorphic computing, such as SynSense, are pioneering event-based vision processing. While still nascent, neuromorphic architectures process visual data only when changes occur in the field of view, promising power efficiencies orders of magnitude better than traditional frame-based architectures, posing a long-term disruptive risk to conventional image signal processing methodologies.
Management Track Record and Capital Allocation
Under the continuous leadership of founder and Chief Executive Officer Fermi Wang, management has demonstrated a ruthless willingness to cannibalize legacy revenue streams to secure future relevance. The executive team successfully navigated the traumatic decoupling of the United States and Chinese technology sectors, replacing hundreds of millions in lost revenue from Chinese surveillance giants with high-quality, long-term enterprise agreements in North America and South Korea. This strategic foresight has yielded tangible financial results. Fiscal year 2026 concluded as a record year, generating $390.7 million in revenue, representing a 37% year-over-year growth rate. The momentum persisted into the first quarter of fiscal 2027, with revenue exceeding $100.4 million and physical artificial intelligence operations now constituting over 80% of the total business.
Management has maintained strict financial discipline throughout this capital-intensive transformation. The company has historically committed massive resources to research and development, frequently exceeding 70% of revenue on a GAAP basis, which suppresses bottom-line statutory profitability. However, management has consistently generated positive free cash flow for 15 consecutive years. The recent implementation of a $50 million share repurchase program signals executive confidence in the durability of the current growth cycle and highlights a balanced approach to capital allocation, funding multi-generational silicon node development while mitigating shareholder dilution.
The Scorecard
Ambarella has successfully completed a grueling structural transformation from a consumer video processing component supplier into a high-performance physical artificial intelligence computing provider. The company's algorithm-first architecture secures a definitive competitive advantage in performance-per-watt, a metric of absolute supremacy in edge computing environments spanning electric vehicles, robotics, and enterprise security. The recently secured $800 million decade-long agreement with Hanwha Group definitively validates Ambarella's strategic pivot away from geopolitical risk and solidifies its recurring revenue base within the premium industrial and enterprise Internet of Things sectors. Furthermore, the systematic penetration into the automotive domain via Tier-1 partnerships with Continental and Bosch positions the company to capture outsized value as advanced driver assistance systems mandate heavier local computational workloads.
Despite these formidable engineering achievements, Ambarella faces a brutally contested market environment requiring relentless capital expenditure to maintain silicon parity. The company operates in the crosshairs of semiconductor apex predators like Nvidia and Qualcomm, alongside heavily subsidized, highly aggressive Chinese domestic entrants capturing volume in the critical Asian automotive ecosystem. Consequently, gross margins face perpetual structural ceilings, while long automotive design-in cycles defer operating leverage. Ambarella remains a high-beta execution story where sustained revenue growth and non-GAAP profitability must persistently outpace the inherent commoditization risks of the broader semiconductor cycle, demanding flawless execution from management as they deploy next-generation transformer and vision-language model capabilities to the edge.