Ambarella: CV8 Two-Nanometer Chip Returns in Weeks, Production Ramp Set for Q2 2026 — With a Second LTA Quietly in Place
Bank of America Global Technology Conference, June 2, 2026 — CEO Fermi Wang lays out the most concrete timeline yet on next-generation silicon and a second long-term customer agreement
Ambarella's CEO Fermi Wang used his appearance at the Bank of America Global Technology Conference on Tuesday to deliver the clearest public roadmap yet on the company's two-nanometer CV8 chip, while also confirming the existence of a second long-term customer agreement tied to that silicon — details that had not been fully disclosed before. Speaking in a fireside chat with BofA semiconductor analyst Vivek Arya, Wang was measured but notably specific on timelines that matter to investors modeling the next leg of growth.
CV8 Is Back From the Foundry in Weeks, Production Starts Q2 Next Year
The most actionable piece of new information from the session concerns CV8, Ambarella's first two-nanometer chip. Wang confirmed the device was taped out in January and is set to arrive back at the company's offices within weeks. More importantly, he outlined a concrete production ramp schedule: "All of the indications so far is that the process will be able to ramp up in production in the second half of next year — in fact, second quarter next year." That is an earlier and more specific schedule than the market had previously been given. Wang added that a second, as-yet-unnamed customer has already signed a long-term agreement contingent on CV8, providing NRE contributions ahead of volume shipments. The details of that deal have not been disclosed, but its existence meaningfully changes the risk profile of the CV8 program.
Samsung Warning on 4-Nano and 5-Nano Supply Is Driving Inventory Build
Wang disclosed that Samsung Foundry has proactively warned Ambarella that supply at the 4-nanometer and 5-nanometer nodes will be tight in 2026 due to incremental customer additions at those process nodes. "Samsung basically informed us with good intention — hey, I know we have a partnership, we want to give you the best service, so give us your PO commitment so we can secure those slots for you." The result is a deliberate inventory build that Wang described as prudent given potential upside from customers, not a sign of demand overconfidence. He expects inventory to drift slightly lower but remain elevated relative to prior levels. This is a notable supply-chain signal for anyone tracking foundry utilization trends.
Hanwha: A $800 Million, 10-Year Deal With More Upside Than Currently Modeled
Wang provided the most granular color yet on the Hanwha Vision long-term agreement announced on the prior earnings call. He noted that Hanwha Vision currently represents roughly mid-single-digit percentage of Ambarella's total revenue — a modest starting point relative to the $800 million, 10-year headline figure. The path to monetization runs through three vectors: gaining share of wallet within Hanwha Vision's existing camera business where Ambarella currently holds less than 50% of the silicon content, moving Hanwha Vision to newer, higher-ASP chips, and then expanding into other Hanwha Group verticals — drones, robotics, retail, banking — by redeploying the same camera hardware with different software stacks and AI models. Critically, Wang said next year should begin to show a positive revenue impact, but acknowledged the deal is not yet reflected in current or near-term consensus estimates. "I don't think that's been factored in just yet," he said. That gap between deal size and modeled contribution is something analysts will need to close.
Wang also contextualized why these long-term agreements are structurally compelling for Ambarella beyond just the revenue line: "Working with a company like Hanwha really gives us not only the scale but also a better way to forecast our revenue in the future." He noted this is the second such agreement in the past 12 to 18 months, with additional engagements ongoing. The common thread is large enterprises that previously contemplated building proprietary silicon but found the economics untenable at advanced nodes. "Trying to build their own silicon, paying $100 million just to do one chip at two nanometer doesn't make sense," Wang said, pointing to the structural shift pushing customers toward differentiated merchant silicon rather than custom development.
Growth Rate Reality Check: Edge AI Is Not a Data Center Story
Wang was refreshingly direct when Arya pressed on whether Ambarella should be thought of as a 30% to 50% growth company given the AI narrative. "That's talking about data centers, but there's no edge AI market that grows that fast." He pegged his served addressable market at an 18% CAGR and acknowledged the company has not yet hit an inflection point that would allow it to outgrow that SAM sustainably. Current fiscal year guidance calls for 10% to 15% revenue growth. Street consensus for next year sits around 15%, which Wang described as reflecting reality. The honest framing here is important: Ambarella is a well-positioned edge AI platform company, but it is not a hyperscaler-adjacent story and investors pricing it as one will be disappointed in the near term.
Memory Inflation Is the Demand Wild Card
Wang identified DRAM pricing as the single most important demand constraint facing customers across both automotive and IoT segments. "Customer gives very consistent feedback — they can still buy memory, it just became extremely expensive." He said the company has not yet determined whether to pass foundry or supply chain cost increases to customers, but acknowledged that if inflationary trends continue, some form of price adjustment becomes likely. Foundry pricing itself has remained stable, but the broader supply chain has tightened incrementally. The gross margin guide of 59.75% for the current quarter is in line with the prior two periods and sits at the lower end of the long-term 59% to 62% band, with mix — not inflation — cited as the primary driver of quarterly variability.
Drones and Edge Infrastructure: The Two Near-Term IoT Catalysts
Within IoT, Wang singled out two areas as near-term revenue drivers rather than long-cycle bets. On drones, the company has accumulated 15 robot design wins cumulatively, including several drone wins, representing approximately $100 million in revenue pipeline. However, Wang was careful to temper expectations on the timeline impact of U.S. restrictions on Chinese drone manufacturers: existing FCC-approved inventory in the market is unaffected by the new rules, meaning volume displacement to U.S.-aligned suppliers like Ambarella will manifest at the next product generation rather than immediately. The second catalyst is what Wang calls "edge infrastructure" — appliances that aggregate legacy camera feeds and enable them to run generative AI models without replacing the cameras themselves. "In this hotel floor, there are maybe 20 traditional cameras. To upgrade those to be GenAI-ready, you can plug in an appliance that aggregates all the video feed, and suddenly all cameras can be enabled by GenAI." The form factor economics — 8K P60 video with AI inference at under five watts for a $30 to $40 silicon cost — are the central competitive moat Wang describes against Qualcomm and NVIDIA, whose architectures are optimized for programmable workloads rather than this specific embedded AI use case.
Cooper SDK: The Software Lock-In That Makes the Hardware Sticky
Wang spent considerable time on Ambarella's Cooper software platform, which he said is now used by nearly 100% of its customer base. The platform supports 200 model architectures — not individual models — and recently added agentic tooling to further reduce customer software development effort. The strategic value is portability across silicon generations: a customer who builds a product on one Ambarella chip can migrate to a higher- or lower-performance variant with minimal software rework. This is the primary reason Hanwha chose to commit to the long-term agreement; the company had been using Cooper for five years and viewed the maturity of the platform as a group-level solution for edge AI standardization. Wang acknowledged that scaling to robotics and other fragmented IoT verticals will require an indirect sales model — software partners, distributors, and system integrators — and suggested this is where any M&A capital would be most productively deployed, rather than technology acquisitions where he sees no obvious gaps in the current stack.