Ambarella (AMBA)
A vision of the future.
Rating: BUY | Price Target: $110 | Current Price: $72.18 | Upside: +53%
Old Harbor Research · May 29, 2026
For informational purposes only. Not investment advice. The author may hold a personal position in AMBA. All data sourced from publicly available filings. This piece is an update. Please read the original pitch.
Ambarella (AMBA) is a fabless semiconductor company that designs power-efficient, high-performance system-on-chip (SoC) products for edge AI applications in security, automotive, drones, and robotics. The company’s DNA in high-quality video processing at the edge has translated to an expertise in vision-based AI inference, which is, of course, highly relevant for the future of robotics. For much of this year, the stock has traded down due to a slowdown in top-line growth against a persistently unprofitable backdrop.
This is one of the many names that attracted a huge wave in 2021 and has languished since. Now that the AI technology has caught up, expanding use cases for Ambarella’s technology dramatically, and physical AI is coming into vogue, the stock may be allowed to challenge its previous heights with actual earnings supporting it.
In April, I wrote a 2-page pitch for the stock which I would like to revisit now, as I believe today’s -21% selloff after earnings is a sentiment reset after a significant run up, which creates an attractive entry point. Please take a moment to review the original pitch for details I will not cover in this brief update.
The Thesis, In Broad Strokes
Physical AI inference is one of the next major branches in the AI industry. This is not a fringe view; I take no credit for the idea. It’s simply the next logical step, largely gated by engineering challenges (which we seem to be having more success in solving every day). Jensen Huang, CEO of Nvidia, has expressed this same sentiment consistently in recent months.
Ambarella specializes in compact SoCs which can run AI models entirely locally on an extremely tight power budget, pulling 2 to 20 watts depending on the chip family and use case. For a sense of how efficient that is, datacenter server deployments would require at least several hundred watts for the combined GPU and system overhead, plus cooling and network bandwidth. So these little SoCs that can run AI models locally are exactly what is needed for edge deployments.
Imagine a drone, where a large battery is impossible. Imagine a robot, where space is limited, power needs to be directed towards movement, and battery life is paramount. Imagine a forward-deployed security system where communication with a datacenter is a vulnerability. If you can have the “brain” of the system on a chip pulling less than 20 watts while running the latest vision-AI models up to 34 billion parameters, that opens up tremendous possibilities for the rest of the system.
Indulge me as I quote myself: “$AMBA is a ticker you want to be long if there’s going to be an explosion of humanoids or otherwise agentic physical AIs in unstructured space.”
Ambarella chips aren’t going to be in every robot, but they are positioned to dominate in vision-based systems that can’t be tethered to a datacenter.
Q1 Earnings and the 21% Haircut
Overall, earnings were fine: reasonably strong revenue coming in slightly above the midpoint of guidance. The full year revenue guide was held at 10-15%, which may have disappointed the market under the circumstances of a +60% gain in the stock since the prior earnings release.
Notable Positive Developments:
Signed an $800M+, 10-year agreement with Hanwha to integrate Ambarella’s platform across its businesses in physical security, robotics, life sciences, and operational automation. This represents potential revenue (not guaranteed), but it serves as validation for the thesis and offers some increased revenue visibility.
Taped out first 2nm chip (CV8) at Samsung with production expected in H1 FY2028 (roughly a year from now). This semi-custom ASIC will serve both consumer and enterprise IoT endpoints.
Robotic design wins: “I’m very pleased to share that we now have 15-plus robotic design wins, including aerial drones, with lifetime revenue exceeding $100 million with more than 30 customers in our robotic pipeline.” (Founder/CEO, Fermi Wang)
Possible Reasons for the Selloff:
Simply an overexcited market meeting in-line results and held guidance.
Inventory days increased from 99 to 145 days, potentially indicating demand softness or supply miscalculation. However, the company claims it was done intentionally to ensure supply for customers in the context of several new product cycles.
The results, on the margin, certainly don’t warrant a 21% collapse (it wasn’t even trading down more than a few percentage points until cash open the next day). With heavy trading volume this past week (2-3x daily average), a general risk-on appetite, and the apparent involvement of financial celeb Citrini Research in the days leading up to earnings, the move could easily be explained by crowding behavior. Therefore, the move should not scare investors away from what is a healthily growing business posting in-line numbers and positive deal news.
Going Forward
While I put forward an $89 base case 12-month price target in my April pitch, I now feel comfortable elevating that to $110 on account of continued execution in product development, design wins, and a notably increased appetite for physical AI amongst investors and AI industrialists alike. This represents a +53% increase from where it currently trades.
It’s worth noting, I think, the character of the stock is such that you have to trade it as a prospective bubble stock, so to speak. The range of potential bullish outcomes is so broad that 5 years from now the company could easily be worth 5x or 10x what it is currently (or 0.5x, naturally), which also means the market could take it to similar heights in half that time if it’s picked up as an “AI bottleneck” by those who identify with such things. To chop off that potentially irrational right tail by focusing on quarterly inventory builds or margin headwinds on product mix defeats the inherent advantage in owning a stock like this.
It’s probably going to become uncomfortably expensive, but that means it’s working.
For informational purposes only. Not investment advice. The author may hold a personal position in AMBA. All financial data sourced from publicly available filings.
If you found this report to be worth your time, please consider sharing it. Thank you!