--- title: "What makes Ambarella claim to be the power efficiency expert in the AI era?" type: "Topics" locale: "en" url: "https://longbridge.com/en/topics/40669112.md" description: "A buddy asked about $Ambarella(AMBA.US) — it closed at .72, up 4.48% on May 11th EST, and only gained 7% for the week. It doesn't look like a blockbuster stock like MXL, but it's always mentioned in semiconductor sector discussions. Today, I'm going to lay all its cards on the table and talk about it. First, a question: What's the most expensive resource in the AI era? Most people would say computing power, which is why NVDA has skyrocketed. But let me look at it from another angle—the flip side of computing power is energy consumption. An H100 at full load draws 700 watts..." datetime: "2026-05-12T09:33:33.000Z" locales: - [en](https://longbridge.com/en/topics/40669112.md) - [zh-CN](https://longbridge.com/zh-CN/topics/40669112.md) - [zh-HK](https://longbridge.com/zh-HK/topics/40669112.md) author: "[沪上老徐](https://longbridge.com/en/profiles/26419342.md)" --- # What makes Ambarella claim to be the power efficiency expert in the AI era? A buddy asked about $Ambarella(AMBA.US) (Ambarella) this stock — it closed at .72, up 4.48% on May 11th ET, and only gained 7% for the week. It doesn't look like a blockbuster like MXL, but it's always mentioned in semiconductor sector discussions. Today, I'm laying all its cards on the table to talk about it. First, a question: What is the most expensive resource in the AI era? Most people would say computing power, which is why NVDA has skyrocketed. But let me look at it from another angle — **the flip side of computing power is energy consumption**. A single H100 at full load draws 700 watts; an AI data center's annual electricity bill can burn tens of millions of dollars. That's why peripheral "picks-and-shovels" stocks like Constellation Energy (nuclear) and Vertiv (liquid cooling) rose recently. But have you thought about another type of company — **not saving power for GPUs in data centers, but saving power for AI chips inside cameras, robots, cars, and security devices**. This is what Ambarella does, technically called "edge AI." What is edge AI? For example, a camera at a residential gate needs to recognize a license plate. The traditional method is to stream the video to the cloud, where a model runs to identify the plate number and sends it back to the gate — this process requires transmitting gigabytes of video, consumes significant bandwidth, and has latency. Edge AI is about performing this recognition task directly on the chip inside the camera, locally computing and only sending a "yes/no" result back to the cloud. Bandwidth is saved, latency is eliminated, but it demands that the small chip have **low enough power consumption** — because many devices are battery-powered, like dashcams, drones, and wearables. What Ambarella does is make this kind of "low-power, high-performance" edge AI SoC. For FY2026 (ending January 2026), full-year revenue was 0.7M, up 37.2% YoY; of that, **80% of revenue came from edge AI (HAI platform), which grew 50% YoY**. Cumulatively, HAI revenue has reached $1 billion, with 42 million SoCs shipped and over 370 customers in mass production — this is a product matrix, not a single hit. What does its edge AI chip cover? Official statement: compute range of 1-500 effective TOPS, capable of running models from tens of thousands of parameters (typical CNN) up to 34 billion parameters (the largest edge SoC available on the market today), with the next generation pushing to 100 billion parameters. In other words, it's not competing in NVDA's market; it's targeting the niche of "the largest model that can fit on the edge." In terms of business structure, IoT accounts for 78%, within that security cameras are about 45%, portable video (action cameras, dashcams, drones) also about 45%, and the remaining 10% is "green shoots" like access control, wearables, robots, and enterprise video conferencing — these are seed businesses, not contributing much this year but a source of mid-to-long-term imagination. Automotive accounts for 22%, of which 95% is L1-L2 ADAS / telematics, and 5% is L2+ to L4 advanced autonomous driving (clients include commercial truck AV companies like Kodiak and Aurora). I think the most interesting thing about AMBA isn't the current numbers; it's the technology roadmap: **The first 4nm CV7 has been taped out, contributing revenue by the end of FY26; the first 2nm gate-all-around has also been taped out** — it's one of the few small companies among TSMC's first batch of 2nm customers. A small company with only $400 million in revenue and still losing money (FY26 net loss .87M) dares to use 2nm, indicating its downstream customers can afford the unit price of 2nm. But this stock isn't without issues. I'll list two risks, more worth watching than the stock price: **First, customer concentration. A single distributor, WT Microelectronics, accounted for 73.1% of Q4 revenue and 69.7% of full-year revenue**. If anything changes in this relationship, revenue would take a hit that quarter. **Second, FY27 guidance is only 10-15% — a clear deceleration from FY26's +37%**. Management explains it as "we always give conservative guidance; last year we said 10-15% and actually delivered 37%," but the market won't buy that unconditionally; it will look at the Q1 actual numbers. Ambarella is about to report FY27 Q1 this week (specific date to be announced), and that's the critical moment. My take on AMBA: **It's not the fastest runner in the AI cycle, but it's one of the few small-cap stocks in this wave that makes sense across all three dimensions: computing power, power efficiency, and customer diversity**. I won't chase it at the current .72 level in the short term; I'll wait for the FY27 Q1 earnings report — if growth can actually reach the upper end of 15-20%, then consider entering in batches; if it comes in below 12%, then FY27 growth expectations will take a hit, possibly offering an opportunity below $70. A reminder for those who "see AI and want to charge in": AMBA isn't a blockbuster that gives you a 5x return in 1 month (refer to MXL); it's a hidden champion that gives you a steady double in 12 months — provided the Q1 earnings don't disappoint. ### Related Stocks - [AMBA.US](https://longbridge.com/en/quote/AMBA.US.md) - [AUR.US](https://longbridge.com/en/quote/AUR.US.md) - [MXL.US](https://longbridge.com/en/quote/MXL.US.md) - [AUROW.US](https://longbridge.com/en/quote/AUROW.US.md) - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [NVD.DE](https://longbridge.com/en/quote/NVD.DE.md)