Non-GAAP profit of $8.2M against a $260K operating cash outflow, and zero shares bought back under the new authorization.
Thesis: The 5nm CV75/CV72 ramp is real and automotive records are real, but GM printed 59.3% against a 59-60.5% guide while the CEO claims memory has zero gross margin impact, and cash flow went negative on IP license payments with 157 days of inventory building. The Say/Do Gap on pricing power is one quarter in. The structural driver (node transition) supports the top line; the earnings quality question is unresolved and cash is burning.
Verdict: HOLD — Conviction: MEDIUM
Catalyst: Q3 gross margin print at or above 59.5% with sequential revenue holding the $119.5M midpoint, plus first X7 design win announcement. That confirms pass-through is working and the channel partnerships translate to dollars.
Key Risk: Q4 gross margin prints below 59% or full-year revenue guide cuts on memory allocation. Either confirms the pass-through mechanism failed within one quarter of the CEO's denial.
The Tell: When asked directly how much memory cost inflation was absorbed this quarter, Fermi avoided quantifying it entirely: "the memory price does not impact our gross margin... we do not buy memory, and we do not resell memory." He reframed from a margin question to a volume question. No basis points number was given despite Liam Farr asking for exactly that.
Friction Level: MODERATE_FRICTION — Both sides agree on the beat and the 5nm ramp. Bull reads memory shortage as an architectural tailwind pulling GPU-to-AMBA design wins. Bear reads GM printing below midpoint in the same quarter the CEO denied memory impact as proof of a pass-through lag. Same facts, opposite direction on where pricing power actually sits.
2027Q2 · Sep 3, 2026
3 of 5 readings agreed3 HOLD · 1 AVOID · 1 LONGRead the full call
Cash Flow Negative While CEO Says Memory Doesn't Matter
The 5nm CV75/CV72 ramp is real and automotive records are real, but GM printed 59.3% against a 59-60.5% guide while the CEO claims memory has zero gross margin impact, and cash flow went negative on IP license payments with 157 days of inventory building. The Say/Do Gap on pricing power is one quarter in. The structural driver (node transition) supports the top line; the earnings quality question is unresolved and cash is burning.
When asked directly how much memory cost inflation was absorbed this quarter, Fermi avoided quantifying it entirely: "the memory price does not impact our gross margin... we do not buy memory, and we do not resell memory." He reframed from a margin question to a volume question. No basis points number was given despite Liam Farr asking for exactly that.
CEO stated memory price "has no impact to our gross margin" and they plan to pass costs through. GM printed 59.3%, below the 59-60.5% guide midpoint, in that same quarter. The pass-through lag is visible one print into the memory shock.
Days of inventory rose from 145 to 157 while revenue ramped 7.7% sequentially. Inventory dollars fell 4% but revenue grew faster. This reads as pre-buying against the memory scramble rather than demand pull, and 157 days is long into a Q4 the CFO could not commit to.
Revenue $108.1M beat guidance midpoint, opex came in below guide at $57.4M against $56-59M range, DSO improved from 35 to 32 days, and Q3 guided to $119.5M midpoint. The operational execution is real and tied to the 5nm node transition.
50 million edge AI SoCs shipped on the Cooper platform. New wins with Canon, Suprema, IDIS, Sepro, Moultrie, Audi/VW in-cabin, and an S&P 100 communications company all run through the same development environment.
SAM revision from 18% to 20% CAGR rests on the X7 accelerator (sampling, zero disclosed design wins) and unannounced AI SoCs. CEO himself described the competitive field as NVIDIA, Qualcomm, and ~50 startups calling it "a crowd space."
-20
CAUTIOUS""That definitely is an uncertainty" on Q4 memory, "barring for any memory impact to our revenue" qualifies the seasonality answer, and "we continue to watch for the Q4" repeats throughout. Confident on specifics ($500M partnership sizing, X7 architecture detail) but hedging concentrated entirely in the forward outlook."
Specific GM, opex, DSO, inventory, cash flow numbers disclosed. CFO gave clean balance sheet detail. But X7 has zero disclosed design wins, SAM rests on unannounced products, and no quantified memory cost impact in basis points was provided despite direct analyst request.
Q3 gross margin print at or above 59.5% with sequential revenue holding the $119.5M midpoint, plus first X7 design win announcement. That confirms pass-through is working and the channel partnerships translate to dollars.
Q4 gross margin prints below 59% or full-year revenue guide cuts on memory allocation. Either confirms the pass-through mechanism failed within one quarter of the CEO's denial.
The verdict is public. The case behind it is Pro: the forensic reading of guidance and language, the bull and bear cases, the red flags, and the play. It opens to everyone on Dec 2, 2026.