6 Best Small-Cap Edge AI & Embedded Compute Stocks — May 2026
Edge AI is not the same as datacenter AI. Inference at the edge happens on auto SoCs, IP-licensed DSP cores, embedded perception sensors, and optical interconnect. 6 small-cap names, with the model-specific framing each requires.
Edge AI — inference running on devices outside the cloud — is structurally different from datacenter AI in customer profile, gross-margin shape, and capex intensity. The names that get cited as 'AI semis' are usually datacenter-exposed (NVIDIA, AMD, Marvell, AVGO). The edge-AI small-cap layer is a different conversation: auto SoCs, IP licensing, perception sensors, optical interconnect.
We scored every small-cap with edge-AI or embedded-compute exposure. Here are the 6 highest-scoring names, with explicit framing of where each plays in the edge stack.
Why Small-Cap Edge AI & Embedded Compute Is Different
- Edge != datacenter — both touch AI but the customer profiles, revenue cycles, and gross-margin shapes are different. Don't conflate them when sizing positions.
- IP-licensing models are capital-light — CEVA's royalty model is gross-margin-positive. Pure-fab-customer models (auto SoCs, perception) are capex-heavy with thinner gross margins.
- Auto cycle dominates a third of this list — INDI, AEVA exposure to ADAS and autonomous-driving platforms means OEM-cycle softness affects revenue directly.
- Optical interconnect at the edge is emerging — POET's optical-interposer thesis is on AI-cluster interconnect, not strictly edge inference; included here because the small-cap silicon-photonics universe overlaps the edge story.
Our scoring rewards capital efficiency, gross margin (IP-rich vs commodity), runway, and insider alignment. For semis, gross margin is the leading indicator of business model quality.
Top 6 Small-Cap Edge AI & Embedded Compute Stocks by Fundamental Score — May 2026
Get picks like these first.
Free weekly digest — top movers, score changes, new small-cap opportunities. No spam, unsubscribe anytime.
1. MaxLinear, Inc (MXL) — Score: 70.7 | Grade: SOLID
| Metric | Value | Score |
|---|---|---|
| Revenue Growth YoY | +29.7% | 48 |
| Gross Margin | 56.8% | 80 |
| Cash Runway | >36 months | 100 |
| Debt/Equity | 32.09 | 73 |
| P/S Ratio | 3.2x | 88 |
| Rule of 40 | 7.8 | 32 |
| Insider Ownership | 6.7% | 48 |
| 12m Dilution | +0.1% | 99 |
What drives the score: MaxLinear makes RF, broadband, and connectivity ICs. Recent re-orientation toward AI-infrastructure interconnect (PAM4 SerDes, optical PHY) tries to attach the company to AI capex; legacy broadband remains the volume. Profitable, modest growth.
Key risk: Broadband legacy is the volume base; AI-infrastructure re-orientation is recent and unproven at margin-impact scale. Customer concentration in handful of cable/broadband OEMs; competitive pressure from Marvell, Broadcom on PAM4 SerDes.
Market cap: $1.48B. Industry: Semiconductors.
2. CEVA, Inc. (CEVA) — Score: 57.7 | Grade: SPECULATIVE
| Metric | Value | Score |
|---|---|---|
| Revenue Growth YoY | +2.5% | 4 |
| Gross Margin | 87.1% | 100 |
| Cash Runway | 145 months | 100 |
| Debt/Equity | 4.79 | 96 |
| P/S Ratio | 5.0x | 70 |
| Rule of 40 | -7.9 | 14 |
| Insider Ownership | 3.5% | 28 |
| 12m Dilution | +16.1% | 22 |
What drives the score: CEVA licenses DSP and AI-inference IP to chip vendors (MediaTek, Renesas, Sony). Royalty + licensing model, capital-light. AI-inference exposure is real but indirect — they don't sell silicon, they sell the math.
Key risk: IP-licensing model is structurally good but customer concentration is real — top 5 licensees produce most royalty revenue. License renewals every 5-7 years can show negotiation lumpiness. Edge-AI inference share is real but indirect (the customers are the chip vendors).
Market cap: $551M. Industry: Semiconductors.
3. Identiv, Inc. (INVE) — Score: 48.0 | Grade: SPECULATIVE
| Metric | Value | Score |
|---|---|---|
| Revenue Growth YoY | -38.7% | 0 |
| Gross Margin | 1.3% | 2 |
| Cash Runway | 105 months | 100 |
| Debt/Equity | 1.31 | 99 |
| P/S Ratio | 3.6x | 84 |
| Rule of 40 | -141.7 | 0 |
| Insider Ownership | 5.9% | 44 |
| 12m Dilution | -8.5% | 100 |
What drives the score: Identiv makes RFID, IoT, and physical-access security devices. Edge-IoT exposure is the loose AI angle; the company is small, has gone through restructuring, and is not a pure AI silicon play.
Key risk: Smallest of the names here, sub-$100M market cap. Multiple business segments (RFID, IoT, security devices), AI exposure is loose. Scale and capital position make this more of a binary survival story than a fundamentals investment.
Market cap: $79M. Industry: Building Products & Equipment.
Still reading? Get the full weekly rundown.
Every Monday: fresh score changes, new names crossing 80+, and context you won't find in the screener.
4. indie Semiconductor, Inc. (INDI) — Score: 42.3 | Grade: SPECULATIVE
| Metric | Value | Score |
|---|---|---|
| Revenue Growth YoY | +0.3% | 1 |
| Gross Margin | 39.9% | 55 |
| Cash Runway | 31 months | 93 |
| Debt/Equity | 103.25 | 9 |
| P/S Ratio | 2.5x | 93 |
| Rule of 40 | -66.4 | 0 |
| Insider Ownership | 2.3% | 19 |
| 12m Dilution | +5.6% | 66 |
What drives the score: indie Semiconductor makes auto-grade SoCs and analog ICs for ADAS, in-cabin sensing, and EV systems. Heavy automotive concentration, customer wins (HD radar, vision processing) drive the bull case; auto cycle is the bear case. Lossy at current scale.
Key risk: Heavy auto exposure means cycle risk is structural. ADAS and in-cabin sensing wins are real but auto OEMs are slower-buying customers than consumer or server. Lossy at scale; runway and dilution discipline matter.
Market cap: $537M. Industry: Semiconductors.
5. Aeva Technologies, Inc. (AEVA) — Score: 41.5 | Grade: SPECULATIVE
| Metric | Value | Score |
|---|---|---|
| Revenue Growth YoY | +110.2% | 100 |
| Gross Margin | -41.8% | 0 |
| Cash Runway | 3 months | 5 |
| Debt/Equity | 3.78 | 97 |
| P/S Ratio | 55.5x | 0 |
| Rule of 40 | -1510.0 | 0 |
| Insider Ownership | 22.5% | 89 |
| 12m Dilution | +11.0% | 37 |
What drives the score: Aeva builds 4D FMCW LiDAR for autonomous driving and industrial automation. Daimler Truck program is the marquee customer relationship. Capex-heavy, lossy, share count growing through equity issuance.
Key risk: FMCW LiDAR is technically distinct from competitor approaches (Innoviz, Luminar — both troubled). Daimler Truck program is the marquee but autonomy-program timelines have repeatedly slipped industry-wide. Capex-heavy plus revenue-light is the structural shape.
Market cap: $1.00B. Industry: Software - Infrastructure.
6. POET Technologies Inc. (POET) — Score: 21.0 | Grade: HIGH RISK
| Metric | Value | Score |
|---|---|---|
| Revenue Growth YoY | -91.1% | 0 |
| Gross Margin | N/A | 0 |
| Cash Runway | 19 months | 69 |
| Debt/Equity | 35.01 | 70 |
| P/S Ratio | 1488.7x | 0 |
| Rule of 40 | -56969.1 | 0 |
| Insider Ownership | 0.3% | 3 |
| 12m Dilution | +99.6% | 0 |
What drives the score: POET Technologies makes optical interposers for AI-cluster interconnect (the 'optical engine' inside transceivers). Pre-revenue at commercial scale; design wins with hyperscalers and silicon-photonics partners (Mitsubishi, Globetech) are the catalysts.
Key risk: Pre-revenue at commercial scale. Design wins are documented but volume conversion is the hard part. Silicon-photonics adoption in AI clusters is happening (NVIDIA NVLink optical, hyperscaler co-packaged optics), but specific design-win-to-revenue transitions take 12-18 months.
Market cap: $1.14B. Industry: Semiconductors.
What these 6 stocks have in common
-
Gross margin separates the IP-light from the IP-heavy. CEVA and POET (when commercial) are royalty/licensing-heavy; INDI and AEVA are capex-heavy with thinner gross margins. Investors should match the gross-margin profile to their portfolio thesis.
-
Auto-cycle exposure is concentrated. INDI and AEVA both depend on auto OEM cycles; INVE indirectly through industrial-IoT. MXL and CEVA are diversified across consumer, communications, automotive.
-
'AI exposure' varies in directness. CEVA's DSP IP runs in inference workloads — direct. AEVA's LiDAR feeds perception models — indirect (sensor, not compute). POET's optical interposers serve AI-cluster connectivity — indirect (interconnect, not compute). Don't bundle them as identical AI exposure.
What's not on this list — and why
- Marvell Technology (MRVL) — $80B+ market cap. AI-infrastructure interconnect, optical DSPs. Out of small-cap range; the destination for some MXL bull cases.
- Lattice Semiconductor (LSCC) — $7B+ market cap. FPGA for edge inference. Out of small-cap range.
- Ambarella (AMBA) — $2B+ market cap (cyclical). Camera SoCs for ADAS, surveillance, robotics. Borderline small-cap; check current market cap before sizing.
- Hailo, Mythic, Tenstorrent — private; key edge-AI inference startups not available to public investors.
The biggest edge-AI silicon plays are mid-cap or large-cap (MRVL, LSCC, AMBA-on-the-line). The small-cap layer is more specialized, with each name representing a distinct edge-stack position rather than competing head-to-head.
How to use this data
These scores measure capital efficiency and balance-sheet survivability. For edge-AI names specifically:
- Compare gross margin trajectory — IP-licensing names should be 70%+, fab-customer names 40-55%
- For auto-exposed names (INDI, AEVA), watch OEM-program-win announcements and ADAS-software-stack mandates from Tier-1s
- For royalty names (CEVA), license-renewal cycles every 5-7 years are the lumpy revenue events
- For pre-revenue names (POET), design-win-to-revenue conversion timing is the binary
SmallCapScanner scores are calculated algorithmically based on 8 fundamental factors. They measure financial health, not future performance. See /how-it-works for the full methodology.
Track these stocks and 2,200+ others on our screener. Set alerts to know when a score crosses your threshold, or compare two tickers directly on /compare.