Ambarella's CEO Flags Memory Supply Crunch as Biggest Swing Factor, Defends Edge AI Thesis Against Skeptics With ADI Deal Math
Citi Global TMT Conference, September 9, 2026
Ambarella CEO Fermi Wang used a fireside chat at Citi's 2026 Global TMT Conference to address the market's most persistent question about his company: when does edge AI actually inflect into meaningful revenue. His answer was characteristically unromantic. There is no single killer application driving edge AI the way large language model training drives data center demand, and investors hoping for a clean timeline will be disappointed. But Wang used the session to lay out several data points that sharpen the investment case, including a new read on memory supply constraints that he says represents the single biggest risk to the company's fiscal 2027 guidance.
Memory Supply Chain Emerges as the Key Downside Risk
The most concrete new disclosure from the session was around memory. Ambarella has guided to 10% to 15% revenue growth for fiscal 2027, and Wang was direct about where that guidance could break. "The downside is only one, it's memory," he said. "This is well documented." What is less documented is how deep the uncertainty runs into the fourth quarter. According to Wang, memory suppliers are not even giving Ambarella's customers allocation commitments for November, let alone pricing. "None of the memory supplier can even give them commitment about allocation for November, right? They don't even know how many memory or what's the price they're going to get in November, how can they commit to us." Wang does not expect the situation to resolve this year, telling the audience "it will definitely go to next year," with no visibility on duration. This is a meaningful data point for anyone modeling Ambarella's back half, since it suggests the company's own forecast confidence deteriorates the further out the quarter.
On pricing, Wang confirmed Ambarella has no ability to absorb rising foundry costs and will pass them through. "I have no choice. I know that our customers are going to hate me for that, but we have to," he said, adding that his own suppliers have already raised prices on him. Average selling price for the company's SoCs, around $15 in fiscal 2026, will keep climbing, but Wang was careful to frame this as a mix shift rather than a margin expansion story: gross margin should stay roughly flat, with ASP growth tracking the rising die size and AI performance requirements customers are demanding.
The Hanwha Deal Signals a Broader Semi-Custom Strategy
Wang offered new context on the roughly $800 million long-term agreement with Hanwha, explaining that the process was initiated by Hanwha itself as part of a group-wide push toward a single edge AI silicon platform across its subsidiaries, including Hanwha Vision, its security camera unit that had previously developed its own chips in parallel with buying from Ambarella. The selection criteria, according to Wang, effectively narrowed the field to a handful of vendors capable of hitting power efficiency, vision processing, strong AI inference, and access to 2-nanometer or 4-nanometer process nodes simultaneously. "If you look at the combination of that, Ambarella become one of the very few candidates to talk to," he said. More significant for investors is what Wang thinks this signals about the broader market. He argued that the same dynamic playing out in data centers, where large companies want to control their own silicon roadmap, is starting to show up in edge AI, but with a twist: most large corporations lack the volume economics to justify building custom 2-nanometer chips on their own. That gap, in Wang's view, is exactly what Ambarella is positioned to fill through semi-custom arrangements, selling the same underlying design into other markets once a partner's proprietary needs are met. "I expect more and more large corporations will say, how can I control my own silicon road map. But then I think that will add to our opportunity to grow in this business," he said, suggesting Hanwha could be a template rather than a one-off.
The ADI Acquisition as an Unintentional Validation
Asked why edge AI should be believed as an inflection point now rather than in prior cycles, Wang pointed to Analog Devices' recent $1.5 billion acquisition of a small edge AI startup as an inadvertent data point in Ambarella's favor. He noted the acquired company generates little revenue and, based on public information, tops out at roughly sub-1 TOP to 1 TOP of AI performance, compared with Ambarella's silicon spanning 1 TOP to 1,000 TOPs and $400 million of annual revenue. "I really think that's indication that edge AI is happening and a lot of the companies said, I need a road map and they try to find a way to fill the gap by using acquisitions," Wang said. The comparison is self-serving but numerically pointed: a large analog player is paying a premium for capability Ambarella already has at greater scale and higher performance ceiling.
Samsung Exclusivity Becomes a Strategic Asset in a Tightening Foundry Market
Wang used the session to reinforce why Ambarella remains an outlier in staying exclusive with Samsung rather than diversifying to TSMC, a decision that looks increasingly deliberate as foundry capacity tightens broadly. "We are the only one exclusive with Samsung," he said, arguing that at Ambarella's size, splitting allocation with TSMC would leave the company behind larger customers in the queue for advanced nodes. The exclusivity, he said, extends to packaging and test partner ASE as well, and the payoff is preferential treatment when capacity gets scarce. "The overall semiconductor supply can become extremely tight, not only foundry, packaging, testing, everywhere is becoming tight. And we hope that the relationship we have with our suppliers... we can continue to get a little better treatment than others," Wang said. It is a bet that concentration beats diversification when the whole industry is capacity constrained, and it is being tested in real time given the memory-driven uncertainty Wang flagged elsewhere in the conversation.
Automotive Growth Is Entirely Western, With China a Watch-and-Learn Market
On the automotive segment, roughly 30% of revenue, Wang was unambiguous that near-term growth is coming entirely from Western OEMs and Tier 1s, not China. Chinese OEMs, he said, prefer domestic silicon suppliers, and Ambarella's continued presence in China is more about tracking application innovation than chasing near-term revenue, given how quickly Chinese automakers iterate on product. That speed differential is itself a structural problem: Wang noted Chinese OEMs turn over product lines every 18 to 24 months, while Ambarella's Western automotive design-win-to-production cycle still runs 3 to 4 years for ASIL-rated autonomous driving programs, versus 12 to 18 months for non-ASIL telematics applications. "I think as an industry, we need to start considering how to move faster," he said. On content value, a single consumer vehicle OEM design win is worth "a few hundred million dollars" to Ambarella, with L2+ representing the larger volume opportunity relative to L3, even though L3 carries higher content per vehicle.
Humanoid Robotics: Wang Pushes Back on Near-Term Hype
Asked about humanoid robotics, Wang offered a notably skeptical timeline relative to prevailing enthusiasm in the space. "I think humanoids from a technology point of view is more difficult than Level 5 autonomous driving car," he said, invoking Elon Musk's 2015 promise of Level 5 self-driving by 2016 as a cautionary reference point still unresolved a decade later. Ambarella has a handful of design wins with humanoid companies, but Wang was explicit that volume from this category will not move the financial needle in the near term, and the company's approach is to stay technically ready on both hardware and software rather than treat humanoids as a near-term growth driver.
Competitive Positioning Against NVIDIA and Qualcomm
Wang drew clear lines around Ambarella's three main competitive references. NVIDIA offers superior raw performance but suffers from well-documented power consumption drawbacks that rule it out for battery-constrained edge applications. Qualcomm is more power-efficient than GPU-based solutions but still trails Ambarella on efficiency, and Wang described Qualcomm as "the biggest competitor to us" given the greatest degree of product overlap. Synaptics, by contrast, entered edge AI only two years ago through acquisition and competes mostly in the 1 to 10 TOP range, an area of limited overlap with Ambarella's broader 1 to 1,000 TOP portfolio. The company's differentiation, in Wang's telling, rests on performance-per-watt, a video processing heritage he considers superior to any competitor, and a unified software development kit, Cooper, that lets customers port applications across Ambarella's 15-chip portfolio without significant rework, a point he considers underappreciated by investors relative to its practical value to customers building multiple product lines.