8 Best Small-Cap Edge AI Semiconductor Stocks — October 2026
Eight small-cap edge AI semiconductor stocks ranked on our September 27, 2026 score snapshot, with the gap between each score's trailing growth input and the latest quarterly filing spelled out per name.
Edge AI is the inference layer that runs on the device itself — in the phone, the laptop, the car, the wearable, the factory sensor — rather than in a hyperscale data center. The demand side has stopped being a forecast: major smartphone OEMs now ship AI features on their flagship phones; AI PCs with a dedicated NPU were less than 10% of new PC sales in 2025 but are expected by Research and Markets to become the majority of new PC sales by the early 2030s; and automotive is moving to sensor-heavy ADAS and cockpit architectures that process camera, radar and lidar data locally. Research and Markets (September 2026) forecasts the edge AI chip market to exceed $80 billion by 2036. The engineering constraints differ from the data-center GPU trade: at the edge the binding limits are the power budget, memory bandwidth and non-volatile storage for model weights, and the cost of moving sensor data across a vehicle or a device. Those constraints favor specialized suppliers rather than one dominant architecture, which is why the small-cap layer of edge AI is more fragmented than cloud training.
Below: eight US-listed small-caps (market cap under $2B) that ship pieces of that layer, ranked by our September 27, 2026 score snapshot. Five names carry over from the June 2026 edition (ALMU, AIP, VLN, MRAM, CEVA); GSIT, AMBQ and INDI are new; LAES (119% twelve-month dilution) and QUIK (revenue down 29.9%, HIGH RISK grade at 38.6) drop out. The thing to know before reading the ranking: the score is built on trailing revenue, and for each name below the growth input behind it differs from the latest reported quarter — in both directions. Each write-up says by how much.
Why the Small-Cap Edge AI Setup Is Different This Cycle
1. The score is behind the filings, and the gap runs both ways. A score snapshot is a trailing readout, and its growth inputs have not caught up with the June-quarter filings. For five names the filing is stronger than the input. Ambiq's input reads -4.7%; Q2 2026 net sales rose roughly 90% year-over-year. Everspin's +9.5% sits against a 42% quarter, Arteris' +22.3% against 46%, indie's +0.3% against 24% and Ceva's +2.5% against 13%. For three names it is weaker. Aeluma's +407.6% predates fiscal 2026, in which revenue fell to $4.5 million from $4.7 million. Valens' +22.1% compares with a 5.8% quarter. GSI Technology's +22.4% sits against a June quarter that was flat both sequentially and year-over-year. The ranking below is therefore a fundamentals filter on cash, margin, dilution and balance sheet, not a ranking of current momentum. Read the score as the screen and the filing as the thesis.
2. The binding constraint is power and data movement — and this cycle, supply. A training GPU is judged on data-center throughput. An edge chip is judged on energy per operation, on whether model weights survive a power cycle, and on how cheaply sensor data reaches the compute that reads it — which is why this list spans interconnect IP, NPU IP, a sub-threshold microcontroller, MRAM, sensor links, compute-in-memory and photonic sensors rather than variations of one architecture. The late-2026 issue is supply rather than demand: wafer, packaging, substrate and test capacity is tight industry-wide. Ambiq says it is shipment-constrained, with some customer preorder volumes running three to five times what those customers themselves expected, so its reported quarter reflects shipments the company says were capacity-limited.
3. Design cycles run years, and policy money is visible in the filings. Automotive design cycles in particular run years from design win to volume: Valens' four VA7000 MIPI A-PHY programs are not expected to ramp revenue before 2027, and GSI Technology's Plato tape-out is scheduled for March 2027. On the policy side, the CHIPS and Science Act put $52.7 billion behind domestic semiconductor manufacturing, R&D and workforce programs, and several names here draw on federal R&D or defense contracts. Aeluma holds a letter of intent for up to $30 million from the CHIPS R&D Office. Everspin is recognizing initial revenue under a $40 million contract with a U.S. prime contractor to establish U.S.-based MRAM manufacturing processes. Military and defense customers were 31.3% of GSI Technology's June-quarter shipments. Policy cuts the other way too: U.S. export controls restrict advanced-chip sales into China, and Arteris (Li Auto, SiEngine) and indie (vision-processor wins with OEMs in China, the pending sale of its Wuxi stake) carry that exposure.
The bear case is specific to each business model. Ceva's royalties are thin per unit and tied to consumer volumes, so a consumer-electronics downturn or a large licensee moving to in-house NPU designs hits that line quickly — and the shares trade about 32% above the snapshot market cap. Three end customers accounted for 26.3%, 24.4% and 24.3% of Ambiq's net sales in the six months ended June 30, 2026, in a business that is supply-constrained. indie carries roughly $416 million of debt against $149 million of cash and restricted cash, with a convertible tranche due in 2027. GSI Technology's Gemini bet is financed by equity raises already visible as 32% twelve-month dilution. And Aeluma's EXCELLENT grade resets as the stale growth input rolls off unless commercial, non-contract product revenue replaces it. Scores built on trailing revenue will re-rate in both directions over the next snapshots.
The Names
ALMU — Aeluma, Inc.
Score: 81.3 (EXCELLENT) | Market cap: $0.24B | Revenue YoY: +407.6% | Rule of 40: 361.7 | Cash runway: 38 months
Aeluma grows compound-semiconductor material directly on large-diameter silicon wafers — a CMOS-compatible route to short-wave-infrared photodetectors and imaging sensors for machine vision, automotive LiDAR and 3D sensing. It tops this list on score, but the +407.6% growth input predates the latest fiscal year. For the fiscal year ended June 30, 2026, revenue was $4.5 million versus $4.7 million and the net loss $9.2 million versus $3.0 million; revenue remains primarily government R&D contracts. What changed is the funding path: $56.0 million of cash at June 30, 2026, a letter of intent for up to $30 million from the CHIPS R&D Office, six new contracts worth $5.3 million including NASA and U.S. Navy awards, and qualification partnerships with Tower Semiconductor and Sumitomo Chemical Advanced Technologies. EXCELLENT here means balance sheet plus growth history at more than 45x sales; the grade resets as the stale input rolls off.
AIP — Arteris, Inc.
Score: 66.1 (SOLID) | Market cap: $1.02B | Revenue YoY: +22.3% | Rule of 40: -24.7 | Cash runway: net cash
Arteris licenses network-on-chip (NoC) interconnect IP — the on-chip fabric that moves data between CPU cores, NPUs, memory and accelerators — and says the majority of its new customer designs now incorporate some form of AI compute. The +22.3% growth input understates the run-rate: Q2 2026 revenue was $24.1 million, up 46% from $16.5 million; Annual Contract Value plus royalties exited the quarter at $99.5 million, up 44%; full-year 2026 guidance is $95.0 to $98.0 million of revenue. Enterprise computing averaged 29% of ACV plus royalties over the past four quarters — edge-plus-cloud interconnect, not a pure edge play. The negative Rule of 40 reflects GAAP losses from R&D intensity, not cash strain: $93.3 million of cash plus $28.8 million of short-term investments at June 30, 2026. Catalyst: royalty layering as 2024-2025 design wins reach production silicon. Risk: Li Auto and SiEngine sit inside U.S.-China export-control crossfire.
VLN — Valens Semiconductor Ltd.
Score: 65.6 (SOLID) | Market cap: $0.18B | Revenue YoY: +22.1% | Rule of 40: -26.4 | Cash runway: 26 months
Valens makes high-speed connectivity chips: HDBaseT for audio-video and industrial machine vision and the VA7000 family for the MIPI A-PHY standard, carrying camera, radar and LiDAR data from the sensor to the ADAS compute inside a vehicle. Here the score runs ahead of the filing: the +22.1% input is trailing, and Q2 2026 revenue was $18.1 million versus $17.1 million — above guidance, but a 5.8% rate. Automotive revenue was $5.0 million (30% of the total) versus $4.3 million, and management raised full-year 2026 guidance to $78.0 to $81.0 million, roughly 13% growth at the midpoint. Cash, cash equivalents and short-term deposits were $83.4 million at June 30, 2026, against a net loss of $8.1 million. The four automotive VA7000 design-win programs advanced toward production, with revenue expected to ramp in 2027 — that ramp is the catalyst, and the 26-month snapshot runway is the clock it runs against.
MRAM — Everspin Technologies, Inc.
Score: 59.0 (SPECULATIVE) | Market cap: $0.38B | Revenue YoY: +9.5% | Rule of 40: -2.3 | Cash runway: net cash
Everspin supplies MRAM — magnetoresistive memory that keeps its contents without power and writes at near-SRAM speed, which is what an edge device needs to hold model weights and calibration data through power cycles without the wear-out limits of flash. The +9.5% growth input lags the reported acceleration: Q2 2026 revenue was a record $18.7 million versus $13.2 million, up 42%, and gross margin was 53.9% versus 51.3%. The non-product line includes initial revenue under a $40 million contract with a U.S. prime contractor to establish U.S.-based MRAM manufacturing processes. GAAP net loss was $3.6 million, cash and cash equivalents $43.9 million at June 30, 2026, and Q3 guidance is $19.5 to $20.5 million. The catalyst is the STT-MRAM xSPI family moving from qualification into volume; the risk is that growth now leans on that one government-linked program, or that foundries' embedded-MRAM offerings pull discrete sockets on-chip.
GSIT — GSI Technology, Inc.
Score: 58.7 (SPECULATIVE) | Market cap: $0.20B | Revenue YoY: +22.4% | Rule of 40: -47.1 | Cash runway: 51 months
GSI Technology is a legacy SRAM vendor — military and defense customers were 31.3% of shipments in the quarter ended June 30, 2026 — that has bet the company on Gemini, an associative processing unit that runs search and inference inside the memory array. The SRAM base is flat: revenue for the June 2026 quarter was $6.3 million, level with the prior quarter and the year-ago quarter; the +22.4% growth input is trailing, not the current run-rate. Operating expenses rose to $8.8 million from $5.8 million to fund Gemini-II software and the Plato chip, for a net loss of $4.8 million; cash was $77.0 million after equity raises that show up as 32% twelve-month dilution in the snapshot. Plato tape-out is scheduled for March 2027. This is an option on a new architecture, financed by the SRAM business and the balance sheet, burning roughly $5 million a quarter.
CEVA — CEVA, Inc.
Score: 57.0 (SPECULATIVE) | Market cap: $0.73B | Revenue YoY: +2.5% | Rule of 40: -7.9 | Cash runway: 145 months
Ceva licenses the processor IP inside other companies' edge chips — NeuPro NPUs for on-device inference, DSPs for audio, vision and radar, and wireless connectivity cores — and collects royalties per unit: customers shipped 567 million Ceva-powered devices in Q2 2026, up 16% year-over-year. The +2.5% growth and 87% gross margin inputs describe the trailing model; the quarter accelerated. Revenue was $29.0 million, up 13%; licensing and related revenue was $18.2 million, up 21%; and management raised full-year 2026 growth guidance to 13% to 15% from 12%. Ceva held $44.3 million of cash plus $176.4 million of marketable securities and bank deposits at June 30, 2026. The strategic catalyst is the NeuPro-M NPU selection by what Ceva calls a leading global AI and computing platform company. The risk is thin per-unit royalties tied to consumer volumes, with shares about 32% above the snapshot market cap.
AMBQ — Ambiq Micro, Inc.
Score: 46.3 (SPECULATIVE) | Market cap: $1.37B | Revenue YoY: -4.7% | Rule of 40: -59.2 | Cash runway: 85 months
Ambiq designs ultra-low-power microcontrollers and SoCs on a sub-threshold transistor-operation technique that cuts energy per operation — the reason its chips sit in battery-powered wearables and hearables that need always-on sensing and local inference. The score has not caught up with a company that listed on the NYSE in July 2025: the -4.7% growth input contrasts with Q2 2026 net sales of $33.9 million, up roughly 90% year-over-year. Q3 guidance is $36.0 to $37.0 million, and the CFO said the company ended the quarter with $366.8 million of cash and no debt. The 37% twelve-month dilution in the snapshot is the IPO share issuance. The catalyst is capacity, not demand: wafer, packaging, substrate and test constraints are limiting shipments. The risk is concentration — three end customers at 26.3%, 24.4% and 24.3% of net sales in the six months ended June 30, 2026 — at 14x snapshot price-to-sales.
INDI — indie Semiconductor, Inc.
Score: 41.0 (SPECULATIVE) | Market cap: $0.77B | Revenue YoY: +0.3% | Rule of 40: -66.4 | Cash runway: 31 months
indie is an automotive-focused fabless chip company: 77 GHz radar chipsets, the iND880 vision processor for camera-based ADAS — the silicon that turns a car into an edge inference platform. The 41.0 score reflects a trailing +0.3% growth input, a 103% debt-to-equity ratio and a 31-month runway; the latest quarter differs. Q2 2026 revenue was $64.0 million, up 24%, and Q3 guidance of $67 to $73 million implies roughly 30% year-over-year growth at the midpoint. Named wins: a radar program with a Tier 1 supplier supporting Volvo, iND880 wins with OEMs in China, and Physical AI sensing wins at humanoid-robot makers Unitree and Agibot — indie's first non-automotive edge-AI vector. The balance sheet is the risk: $139.0 million of cash at June 30, 2026, against $403.7 million of long-term debt — mostly convertible notes due 2027, 2029 and 2031.
How to Think About Position Sizing
Three tiers, defined by where the revenue comes from and how far the score input sits from the filing.
IP and royalties (AIP, CEVA). Both license into other companies' chips, run gross margins the snapshot prints at 90.2% and 87%, and hold cash and investments well beyond their quarterly losses (Arteris $93.3 million plus $28.8 million of short-term investments; Ceva $44.3 million plus $176.4 million of marketable securities and bank deposits). Their risk is customer behavior — a licensee going in-house — rather than a balance sheet. This is the durable end of the list.
Product companies with reported acceleration (AMBQ, MRAM, INDI). Each reported a quarter well ahead of its score input (roughly 90%, 42% and 24%); the distinction is the balance sheet. Ambiq's CFO cites $366.8 million of cash and no debt; Everspin holds $43.9 million with a 4.8% debt-to-equity ratio in the snapshot; indie's equity sits behind roughly $416 million of debt, so size it smaller than its growth alone would justify.
Dated-catalyst options (ALMU, VLN, GSIT). The score input overstates the current run-rate for each of the three, and the catalysts carry dates: Aeluma's CHIPS letter of intent converting into a definitive award, Valens' 2027 A-PHY ramp against a 26-month snapshot runway, GSI Technology's March 2027 Plato tape-out against roughly $5 million of quarterly burn and $77.0 million of cash. Size them as options.
A practical mix: roughly 40% in the IP tier, 40% in the product tier and 20% in the options tier — product-tier weight tilted toward Ambiq and Everspin rather than indie, option-tier positions small enough that a slipped 2027 date is survivable. Adjust toward the IP tier for lower volatility, noting that Ceva already trades about 32% above its snapshot market cap, so part of its licensing upswing is already in the price.
How We Scored
Our model rates 2,200+ US small-cap stocks (market cap <$2B) across eight fundamentals: revenue growth, gross margin, cash runway, debt/equity, P/S ratio, Rule of 40, insider ownership and 12-month dilution, sector-adjusted where appropriate. Scores are refreshed regularly; this list uses the September 27, 2026 snapshot, and the write-ups flag where a trailing input has been overtaken by a filing. See the methodology and track record or browse small-cap technology stocks in the screener.
Related Coverage
- 7 Best Small-Cap Edge AI Semiconductor Stocks — June 2026
- 7 Best Small-Cap AI Infrastructure Stocks — October 2026
- 5 Best Small-Cap Quantum Computing Stocks — October 2026
- Best Small-Cap Humanoid & Service Robotics Stocks — June 2026
Data as of September 27, 2026. Updated monthly. Past performance does not guarantee future results. Not investment advice.