安霸 2024年度报告

April 2025 Dear Shareholders: Ambarella saw substantial improvement in its business in fiscal 2025, following the industry-wide cyclicalinventory correction the prior year. We finished the year with stronger financial results and positive momentumheading into fiscal 2026. Edge AI is now clearly established as our key revenue driver, enabling us to grow andovercome the cyclical, economic and geopolitical challenges that have persisted. We exited fiscal 2025 with over70% of revenue from our edge inference AI SoCs, and we have shipped a cumulative total of 30 million edge AISoCs into various markets. Both IoT and automotive revenue grew in fiscal 2025 and, prior to the uncertainty ofthe recent global trade discussions, we anticipated continued growth in both markets in fiscal 2026. We characterize the edge AI market as nascent, despite our growth in the last year It is important to understand that a vast majority of the global AI market activity, to this point of time, hasfocused on AI processing in data centers at the core of a computing network, and this AI market has primarilyutilized graphics processing semiconductors (GPUs). Ambarella is not addressing the data center market. Ourhighly efficient processor, when implemented within the edge of a computing network, reduces latency, improvespower consumption, reduces data communication expense, and improves privacy and security. Relative to the AIcore market, the edge AI market is at a much earlier stage of development. We see a growing trend towards performing more AI processing at the edge, and we have optimized our AIprocessors for edge applications. AI models, which are executed on our chips, are becoming “smaller andsmarter”, which allows them to be practically deployed at the AI network edge. For example, recentbreakthroughs with more powerful, efficient and open-sourced reasoning models have been an exciting industrydevelopment. We expect the deployment of open-source reasoning models to enable more advanced decisionmaking and intelligence in all tiers of the AI processing hierarchy, including our target market at the edge. Whilebreakthroughs like this bring new levels of compute efficiency, the implementation of the new reasoning modelsat the edge is expected to require incremental AI compute in a power efficient manner. We believe ourinvestment strategy and product portfolio are well aligned with this development. We expect growth to be driven by expanding use-cases for edge AI and higher ASP products AI inference processors now represent a majority of both our automotive and IoT revenues and have led to ahigher blended average selling price (“ASP”). We expect continued revenue growth in our edge AI inferenceprocessor business as a result of two primary factors: (1) the expansion of edge applications that areimplementing AI functionality, and (2) a rising ASP of our AI processors. The edge AI inference marketcommenced in the enterprise market (e.g. physical security) and is now reaching consumer markets. Within theenterprise and consumer markets we see several new application verticals at the early stage of adopting edge AI.We also intend to expand our edge AI business from the endpoint (the terminal device in a network whichrepresents a vast majority of our revenue today) into the edge infrastructure market, which will allow us toaddress incremental and higher value market opportunities. We anticipate the ASP of our edge AI processors toincrease as edge AI processing requirements become more sophisticated (e.g. moving from convolutional neuralnetworks (CNN) to transformer AI networks). New products are key to growth From a product perspective, our 5nm products should represent our key growth driver in fiscal 2026, withrevenue expected to arrive in several waves. The first wave, from the growing CV5 SoC family, is underway,with initial revenue generated in fiscal 2025 and projected to grow nicely in fiscal 2026. The second wave, fromour CV7 SoC family, reached production status for the first time in the fourth quarter of fiscal 2025, and is alsoexpected to grow well in fiscal 2026. Combined, these first two waves are expected to represent more than half ofour incremental revenue in fiscal 2026. In addition to the CV5 and CV7 product waves, other new product offerings include our custom/semi-customASIC go-to-market (G2M) strategy and our N1 family of edge AI processors that will allow us to serve the edgeinfrastructure market. Our custom/semi-custom G2M offering enables customers to define SoCs utilizing someof their own IP in combination with our IP (e.g. our AI accelerators and perception). We announced our firstengagement with this new G2M offering on 2nm semiconductor manufacturing process nodes. The 2nm nodedevelopment is critical for our new product roadmap, and we expect to leverage our current experience for futureproduct architectures. For the edge infrastructure market, we introduced our N1-655 edge GenAI SoC at CES this year, wit