Ambarella Unveils X7 Accelerator and Pivots to Edge Infrastructure
New AI SoC, indirect channels, and a memory-strained supply chain redefine its growth playbook
AMBA · Earnings Call · 2026-09-03
Ambarella's fiscal Q2 report is more than another record quarter of edge AI revenue—it is the clearest signal yet of a strategic pivot from camera-centric SoCs to a broader AI platform play. With revenue of $108.1M (+13% YoY) and non-GAAP EPS of $0.18, management used the call to launch the X7, its first standalone AI accelerator, and announced two substantial partnership agreements to build an indirect sales channel. CEO Fermi Wang framed it as the company expanding beyond its traditional market:
We refer to this new AI accelerator as X7. This SoC is sampling now and expected to land initial design wins in edge infrastructure applications where it can serve as an AI coprocessor for host processors such as ARM or x86.
A New Product Line and a Bigger TAM
The X7 is not just another chip—it anchors a new category that targets on-premise inference at the edge, a market where Ambarella sees edge infrastructure as the next growth engine. Management revised its 5-year SAM to $22.9B by FY32, a 20% CAGR, with IoT now representing ~70% of the terminal year. The increase is attributed largely to the company's push into AI accelerator territory. Fermi explained the product's positioning: “We define an AI accelerator as an AI processor that is not camera specific and targets a wide variety of digital or physical modalities.” — Fermi Wang, President and CEO · 2026-09-03
This expansion is a direct response to a market shift he described earlier: "operational efficiency at the edge" using open-weight models on on-premise hardware. The company is betting that its performance-per-watt advantage, proven across 50M+ edge AI SoCs shipped, will translate into the edge infrastructure segment—even as competitors like NVIDIA and dozens of startups crowd the space.
Memory: A Headwind That's Also an Opportunity
The April quarter was marked by soaring memory situation costs, as data-center AI demand squeezes supply. Fermi acknowledged the indirect impact on customer orders: “For Q3, we are comfortable with the guidance we provide today. In Q4, we continue to talk to customers to make sure our customer will have—we can secure enough memory for a Q4 business.” — Fermi Wang, President and CEO · 2026-09-03
The memory crisis may ironically work in Ambarella's favor. Its chips require far less memory than GPU-based alternatives—the X7 only needs 4MB of memory to run a large language model within a 4-5W power envelope. Customers seeking to reduce memory costs are increasingly evaluating Ambarella's power-efficient solutions, as Fermi noted: "most, in fact, almost all the customers who come to talk to us is because our power efficiency solution and the lower cost solution than what they're using." This dynamic could accelerate the shift from GPU-driven first-gen products to Ambarella's custom silicon.
Building Scale Through Partners, Not Just Direct Sales
The most consequential announcement was the creation of an indirect sales channel, which CEO Fermi Wang flagged as critical for reaching small, fragmented markets like robotics and for supporting the more complex edge infrastructure designs. The company announced a 7-year agreement with Macnica (a technical distributor) and a partnership with CapGemini. Fermi said these could each drive $0.5B in incremental revenue over seven years, with “meaningful revenue is expected in 2 to 3 years and will grow as we introduce new products for the market.” — Fermi Wang, President and CEO · 2026-09-03
This marks a deliberate move from the company's long-standing direct-sales model. In the prior quarter, management had framed building out ISVs and channel partners as an important step: "there is significant effort building out channel partners and even system integrators as part of this effort" (Louis Gerhardy, 2026-05-28, component_hash="7475706322666742073"). Now it's turning that talk into concrete agreements, signaling a potential structural change in how Ambarella goes to market. The semi-custom chip strategy is another pillar—the first project, the 2nm CV8, is already underway and expected to generate production revenue in fiscal 2028.
The company is also leveraging these partnerships to strengthen its indirect channel, with a focus on enabling its Cooper developer platform. Both CapGemini and Macnica will use Cooper, which Fermi believes was a key factor in their selection.
Financial Validation and Risks
Ambarella's financials remain in transition. R&D expense of $58M consumed 58% of revenue in the latest quarter, a level that underscores the company's willingness to spend for future growth. Gross margins held at 58.4% (GAAP, as per fundamentals), slightly below the long-term target range of 59-62%, reflecting elevated costs and product mix. Gross margin has declined ~5pp over the past decade, and management does not expect the new business models to change the target range imminently. Free cash flow was sharply negative at -$51M in the quarter, a result of higher inventory and IP license payments—a near-term drain that could become a concern if the X7 ramp and channel expansion don't yield revenue as planned.
Momentum behind the stock is also a question. Over the past 90 days, AMBA shares have fallen 6% and remain 31.5% below their May 2026 peak, even as the company reports record results and an expanding SAM. That gap likely reflects investor skepticism over the memory-driven order environment and the time-line risk of the new partnerships.
Ambarella's latest quarter is not just a beat and raise—it's a repositioning. The company is pivoting from a camera-centric silicon vendor to a broader AI platform player, using edge infrastructure and AI accelerators to expand its TAM, while simultaneously building the ecosystem needed to serve fragmented markets. The X7, the indirect channel, and the semi-custom strategy are early, but they offer a clear answer to the question of what comes after the CV2 and CV5 ramp. As Fermi put it, "We are becoming a full-stack Physical AI platform provider." Whether the market is willing to pay for that ambition will depend on execution over the next few quarters.