AMBA · Ambarella, Inc.

Current Thesis

Both legs that carried this name since 2026-07-31 have now been tested, and only the operating one survived. Q2 FY27 landed 2026-09-03 after the close: revenue $108.1M, +13.2% YoY and +7.7% sequentially, above the $105–111M guide midpoint and above the $107.75M consensus, with non-GAAP EPS $0.18 against $0.17. The margin line did not hold — non-GAAP gross margin printed 59.3%, under the company's own 59.0–60.5% guide and down from 60.5% a year earlier, on a GAAP net loss of $6.7M. Nor did the forward framing: management said memory prices do not hit Ambarella's cost line because it neither buys nor resells memory, but described constrained customer purchasing power as a slight revenue impact in Q2 and Q3 and called Q4 uncertain because of memory constraints. Three target cuts followed on 2026-09-04 — Roth to $65 (Neutral), B. of A. to $70 (Neutral), Stifel to $90 (Buy).

The tape did the rest. The 2026-09-04 close of $62.89 sits 34.2% below the 52-week high of $95.51 with RSI(14) at 15.5, the lowest reading in this coverage window against 23.5 on 2026-08-28 and 69.9 on 2026-08-14. The $63.50 weekly-close level flagged in the 2026-08-29 note went on that same Friday. So the narrative is dead in the shape it traded since the Financial Times report — the premium that lifted the shares roughly 18% to $87.32 on 2026-07-31 has fully round-tripped, the June swing shelf broke on a weekly close, and the sell-side marked targets down into the flush rather than up. What is left is a growing edge-AI silicon supplier, a washed-out chart, an undated and unresolved M&A option, and no company-dated event until the Q3 FY27 print in early December.

Bullish and bearish views on Ambarella, Inc.

The model's bull view on Ambarella, Inc. (AMBA), in brief: The print beat on both lines it was set against: $108.1M revenue and $0.18 non-GAAP EPS on 2026-09-03, versus $107.75M and $0.17 consensus. The bear view: The margin miss was against the company's own guidance, not just consensus: 59.3% non-GAAP versus a 59.0–60.5% guide, GAAP 57.7% versus 58.9% a year earlier. Both cases follow in full.

Bull Case

• The print beat on both lines it was set against: $108.1M revenue and $0.18 non-GAAP EPS on 2026-09-03, versus $107.75M and $0.17 consensus. Automotive revenue set a record on commercial-vehicle adoption; management attributed growth to the 5nm CV75 and CV72 SoCs.
• The Q3 guide did not walk anything back: $115–124M for the quarter ending 2026-10-31, a $119.5M midpoint against $119.31M consensus, with non-GAAP gross margin guided 59.0–60.0% and opex $56.5–59.5M.
• Balance sheet carries the loss: cash and marketable securities of $272.3M at 2026-07-31 against Q2 free cash flow of -$7.1M and capex of $6.8M. There is no financing clock attached to the GAAP loss.
• New dated commercial anchors: the Capgemini engagement for engineering and systems integration on edge and physical-AI technologies was announced 2026-09-03; on the call, Capgemini plus Macnica were framed as up to $0.5B of incremental revenue over seven years. This sits on top of the Hanwha agreement disclosed 2026-05-28 (more than $800M over 10+ years).
• Product cadence continued through the drawdown: the X7 edge-inference accelerator was introduced alongside the print, specified at 4–5 watts and 4MB of on-chip memory — a direct response to the same memory scarcity that is squeezing customers' budgets.
• Third-party targets remain above spot after the cuts: the lowest of the three 2026-09-04 marks, Roth's $65, is still above the $62.89 close; Stifel kept a Buy at $90.

Bear Case

• The margin miss was against the company's own guidance, not just consensus: 59.3% non-GAAP versus a 59.0–60.5% guide, GAAP 57.7% versus 58.9% a year earlier.
• Q4 was explicitly left uncertain on the 2026-09-03 call. That quarter is guided at the ~2026-12-02 (est.) print, which is the next moment anything gets resolved.
• Concentration: WT Microelectronics accounted for 60.2% of Q2 revenue and Acuto 11%. A distributor's ordering pattern can move a quarter without any change in end demand.
• The M&A option has gone quiet. Neither Ambarella nor NXP has confirmed or denied the 2026-07-31 FT report of talks above $3 billion, and the earnings call produced no disclosed update. The market cap was $3.10B at the 2026-08-28 close of $70.63; the shares have closed lower since, so the reported floor no longer implies the premium it did on report day.
• Insider record is one-directional: roughly 93,390 shares sold across three months to 2026-07-09 with zero purchases, including 16,250 shares by the CEO on 2026-07-01 under a 10b5-1 plan adopted 2025-10-08. No open-market purchase has been reported since.
• The Section 232 drone tariffs took effect 2026-09-03 for most rates, and no Ambarella revenue line reflects them. The onshoring linkage remains inference about design wins that have not been disclosed.

Setup & Price Structure

The 2026-09-04 close of $62.89 completes a full round trip of the 2026-07-31 gap and then some: the shares traded roughly 18% higher to $87.32 on the report, closed $82.04 on 2026-08-14 with RSI(14) at 69.9, $70.63 on 2026-08-28 at RSI 23.5, and now $62.89 at RSI 15.5. A three-month price change of -8.5% understates the path — most of the damage is concentrated in the five weeks since the FT headline and the two sessions around the print.

Positioning evidence, stated as observables rather than a verdict: three price-target reductions on a single day (2026-09-04) from firms holding two Neutrals and one Buy; a reported 5.6% regular-session decline on 2026-09-03 extended after hours; RSI(14) at 15.5, an extreme reading that has no directional meaning on its own but marks how one-sided the exit has been; no insider purchase on record against 93,390 shares sold through 2026-07-09; a distributor supplying 60.2% of revenue; and a ~90-day gap to the next scheduled company event. The prior structure is gone — the $72 shelf was forfeited in August and the June swing area near $63.50 went on the 2026-09-04 weekly close. The base has not formed, so there is nothing here to chase; the first thing that would evidence repair is a weekly close back above $70, which is where B. of A.'s reduced target and the 2026-08-28 close both sit.

Catalyst Calendar (next 30 days)

• 2026-10-31 — Q3 FY27 quarter ends. The period into which management guided $115–124M and flagged customer memory constraints.
• ~2026-12-02 (est.) — Q3 FY27 results and call. First disclosure of whether the memory squeeze cut revenue, and the first Q4 guide after management called that quarter uncertain.
• 2027-02-09 — deferred tranche of Section 232 drone-component tariffs takes effect.

Elapsed catalysts

• No company-dated event falls inside the window. The 2026-09-03 print and call are past; the Section 232 drone rates went effective 2026-09-03; no investor conference appearance is listed for the window. (passed 8d ago)
• Undated, open — any statement or follow-up reporting on the NXP talks first reported 2026-07-31. The largest unscheduled variable in the situation. (passed 42d ago)

What Would Change Our Mind

The structural break already happened: the $63.50 weekly-close level published on 2026-08-29 went at $62.89 on 2026-09-04, which is why this refresh treats the takeover-premium leg as failed rather than pending. From here the thesis that the post-print flush is exhaustion rather than the first leg of a de-rating breaks on a weekly close below $57, roughly 9% under the 2026-09-04 close, with no scheduled company event between now and early December to arrest it. On the fundamental side, a Q3 FY27 result under the $115M guide floor, or non-GAAP gross margin printing below 59.0% against the 59.0–60.0% guide, would confirm that the memory cycle is reaching the reported numbers rather than only the commentary. In the other direction, a weekly close above $70 with the Q3 guide intact would evidence that the December print is being marked as a re-rate rather than a further cut. A confirmed end to the NXP discussions removes the option that the last five weeks have been slowly discounting; a signed deal near the reported figure resolves the whole situation at once.

Correlation Notes

No published theme cluster currently carries this name, so this is a single-name setup and the thesis does not lean on a group move. Three external linkages are observable. First, NXPI: the counterparty to the 2026-07-31 report, which fell more than 3% that session — its price behaviour around any deal news is the cleanest read on whether the market believes a transaction exists. Second, the memory complex: management's own explanation for constrained customer budgets ties Ambarella's Q4 to DRAM pricing even though it buys no memory itself, so memory-vendor pricing commentary is a leading input to the December guide. Third, drone and defence policy: the Section 232 rates effective 2026-09-03 and 2027-02-09 create a US-onshoring channel for vision silicon that has produced no disclosed Ambarella design win to date, so the correlation is thematic until a filing or call names it.