Five- and ten-year opportunities, evidence and failure conditions
Research cutoff 2026-10-05; market snapshot 2026-10-02. U.S.-listed selected universe; $300M to below $2B equity market cap. Ranking is subjective research priority, not a whole-market optimum, an undervaluation claim or personalized investment advice.
Research report
1. Thesis: survival before optionality
Small-cap AI investing is not simply large-cap AI investing with a larger growth rate. Customer dependence, restricted access to capital, weaker trading liquidity and less predictable execution can dominate the outcome. A successful technology may never generate attractive returns for the original common shareholders if it needs repeated expensive financing. Our top ten is a selected U.S.-listed shortlist with an explicit $300M-$2B capitalization window, not an exhaustive ranking of every small AI company. Micro-caps are excluded rather than relabeled to fill the list.
Photronics, Red Violet and Ceva lead for an established commercial base, identifiable data/IP or a less speculative manufacturing role. Viant and Simulations Plus follow for monetized decision workflows, with distinct accounting and mix risks. AOSL and Cerence are cyclical or operational turnarounds. C3.ai and BigBear.ai are higher-risk application and contract cases. POET is an early optical commercialization option, placed last and excluded from the standard return model because smooth extrapolation from a tiny revenue base would be misleading. This is a research-priority ranking, not ten equally suitable investments.
2. Direct AI exposure is not always better exposure
An indirect photomask supplier can be a better investment than a pure-play application vendor if the former earns cash at a reasonable entry value and the latter cannot monetize demand. Conversely, a profitable legacy product can conceal a declining business that adds AI branding without creating incremental value. We classify exposure by the commercial mechanism and avoid inventing a percentage of AI revenue when the issuer has not separately disclosed it. Semiconductor masks and power components are enabling layers; identity graphs, advertising decisions and drug-modeling tools embed AI; C3.ai, Cerence and BigBear.ai have more direct application exposure.
For edge IP, license wins must become royalty-bearing shipments. For identity data, permissions and regulatory acceptability must persist. For advertising, contribution after traffic acquisition costs matters more than gross billings. For drug simulation, validate recurring software growth separately from services. For automotive voice systems, distinguish ordinary royalties from fixed contracts and patent settlements. For defense vendors, distinguish funded backlog from an aspirational pipeline. These are practical links between product adoption and cash, not predictions that any particular AI architecture will dominate.
3. Financing and dilution are product risks
A cash runway is not just cash divided by the latest accounting loss. Use restricted versus unrestricted cash, operating burn, committed capex, debt maturities, working capital and realistic funding access. One favorable cash-flow quarter may reflect collections or reduced spending that cannot recur. Equity issuance, warrants and convertible instruments can make headline market capitalization incomplete. POET illustrates the problem: one quote feed implies roughly 61M shares while another dated source reports 173.04M following financing. We use the latter for classification and disclose the conflict rather than infer a bargain from the stale number.
The scenario model assumes annual issuance and capitalizes negative modeled owner cash at a 10% funding cost. This is a stress convention, not a forecast of financing terms or available debt capacity. A real issuer may be unable to fund those losses at any price, or may sell assets and change strategy. Our bear case is not a minimum loss; complete impairment remains possible for any individual equity. Survival analysis and a full capital-structure bridge therefore precede using a favorable model result as an investment argument.
4. Milestones, not a twenty-year spreadsheet
For these companies, a rolling sequence of commercial milestones is more useful than a precise twenty-year valuation. Require paid production adoption, repeat customers, demonstrable gross profit and an improving path to sustainable cash. For early hardware, require qualification and shipment acceptance, not merely memoranda or a potentially large order. For turnarounds, require durable renewal behavior and gross-margin recovery, not just a cost-cutting headline. Growth is valuable only if enough remains after reinvestment and the increase in share count.
Use a higher return hurdle for greater execution and liquidity risk, but do not confuse a spreadsheet hurdle with compensation guaranteed by the market. The screen uses 20% annual capital-gain hurdle for small-caps, versus 15% for mid-caps. The hurdle entry column is a reverse-engineered price conditional on the modeled exit, not fair value or a recommendation. Evaluate liquidity and limit-order execution, but do not select position sizes from this report alone. A diversified core and a separately risk-budgeted speculative allocation are different decisions.
Ranked shortlist and financial evidence
Editorial order emphasizes commercial evidence, defensibility, capital discipline and valuation risk. No numerical ranking score or scenario probability is claimed. Direct AI revenue shares are not inferred. Quarter revenue below is reported; revenue x 4 is a model baseline only, not TTM. Market caps use the dated quote vendor, except POET's disclosed alternate share-count calculation.
Photomasks are necessary to transfer semiconductor designs into production. Photronics offers a less fashionable, indirect route to chip complexity and design activity. It is not a leading-edge EUV monopoly and not all end markets benefit from AI. Its place near the top reflects valuation discipline and established commercial operations rather than a forecast of the fastest AI growth.
Reported evidence
Quarter revenue +2.7%; high-end IC reached 44% of IC revenue.
Commercial milestone
Higher-complexity IC designs and geographic capacity additions improve product mix.
Monitor
Track high-end IC revenue, utilization, capex, minority interests and cash location.
Thesis failure
Mask pricing falls, display demand weakens, or geopolitical limits strand overseas capacity.
Reliable entity resolution and identity data become more valuable as automated activity rises. Red Violet's attraction is an existing monetized data business rather than a speculative model launch. Differentiated data can support repeat usage, but access rights and compliance are essential parts of the moat. A rich multiple already anticipates successful growth, so business quality must be separated from entry value.
Reported evidence
Revenue +23%; GAAP net income $5M; operating cash flow $10.6M.
Commercial milestone
Identity-graph usage expands across consequential transactions and repeat customers.
Monitor
Monitor organic growth, customer count, GAAP margin, data costs and post-offering share count.
Licensing neural-processing, sensing and connectivity IP gives Ceva a capital-light way to participate in smarter devices. The decisive step is converting licenses into royalty-bearing mass production. A license signed today may generate limited royalties for years. AI IP must prove efficiency and integration advantages against internal designs and other licensors.
Reported evidence
License revenue $18.2M; royalties $10.8M; GAAP operating loss $2.1M.
Commercial milestone
New AI license programs generate royalty-bearing device shipments.
Monitor
Track license/royalty split, shipment volumes, royalty per device and GAAP operating expense.
Programmatic buying turns prediction into measurable advertiser decisions. Viant's identity, content and attention context may help smaller advertisers access connected television effectively. Gross revenue includes media-related economics, so comparing its sales multiple directly with SaaS is misleading. The Up-C structure and tax receivable obligations also make a simple common-equity valuation inadequate.
Drug development can benefit from validated simulation and informed trial design without betting on a single molecule. Simulations Plus sells tools and services into that workflow. Its AI exposure is blended with mechanistic modeling and consulting, not a pure generative-AI business. Improving software growth is necessary before a strong recurring-revenue compounder thesis is justified.
Reported evidence
Software $12.6M, flat YoY; services $9.3M; GAAP net income $3.6M.
Commercial milestone
Validated modeling products gain software renewals and broader development usage.
Monitor
Track software versus services growth, renewals, organic margins, acquisitions and customer R&D budgets.
High-current accelerators need reliable power conversion and protection. AOSL's relevant products are controllers, power stages and MOSFETs, supported by its published AI-server portfolio. Diversified consumer and industrial demand remains important. This is a cyclical turnaround candidate, not evidence that every product sale benefits from AI demand.
In-vehicle agents can benefit from embedded OEM relationships and automotive-grade integration. Cerence has a real installed base, but car cycles, fixed license agreements and debt complicate the apparent low sales multiple. A quarter with strong cash collection does not establish durable owner earnings. Evaluate connected-service growth separately from one-time licensing and patent settlements.
Reported evidence
Connected services +20%; GAAP net income $1.5M; reported FCF $19.6M.
Commercial milestone
xUI production launches translate into recurring connected services.
C3.ai offers enterprise applications with direct AI relevance, but product relevance is not proof of shareholder returns. The current case is a turnaround: improve paid deployment, delivery cost and sustainable retention. A large cash balance buys time but can disappear if deployments remain services-heavy. Its low rank reflects execution uncertainty and the gap between bookings rhetoric and durable economics.
Reported evidence
GAAP gross margin 32%; non-GAAP operating loss $36.2M; cash/securities $651.1M.
Commercial milestone
Repeatable deployments lift gross margin and retention rather than only bookings.
Monitor
Track GAAP gross margin, operating loss, cash burn, renewals and dilution; do not extrapolate one positive cash quarter.
Thesis failure
Turnaround fails, project delivery remains costly, or stock compensation dilutes holders faster than revenue grows.
Mission software, secure deployment and procurement relationships can be durable. BigBear.ai offers relevant exposure, including acquired generative-AI capabilities, but government contracting has long cycles and significant services content. Backlog is not the same as funded, high-margin cash flow. Acquisition-led growth needs careful comparison with organic expansion and the shares used to finance it.
Reported evidence
Backlog $269.6M; gross margin 32.8%; adjusted EBITDA loss $11.6M.
Commercial milestone
Funded defense/security contracts convert into higher-margin repeat deployments.
Optical integration addresses an important bandwidth and power problem, but POET remains an early commercialization bet. The potential market is far larger than current sales; that mismatch is a risk, not automatically an opportunity. Orders, qualification and volume cash revenue must be distinguished. Conflicting vendor capitalization also prevents treating a cheap-looking share price as cheap equity.
Reported evidence
Quarter revenue $0.570M; net loss $11.3M; quote cap corrected with 173.04M shares.
Commercial milestone
Customer qualification completes and repeated volume revenue becomes material.
Monitor
Monitor recognized product revenue, shipment acceptance, unit economics, dilution and the post-financing share register.
Thesis failure
Qualification fails or volume ramps stall; follow-on capital and warrants impair per-share returns. No standard CAGR valuation is published for this pre-scale case.
None of the modeled base cases clears the selected 20% annual return hurdle at snapshot prices on either horizon. This is a due-diligence shortlist, not a list of ten proven bargains.
Every forecast is a research assumption, not guidance or consensus. Capital-gain CAGR excludes dividends. Ranking does not follow modeled CAGR. A zero terminal value means model impairment, not a prediction that the issuer must fail. POET is milestone-only; no smooth return forecast is offered.
Return scenarios by company
Stock
Bear 5Y / 10Y CAGR
Base 5Y / 10Y CAGR
Bull 5Y / 10Y CAGR
Base 10Y hurdle entry USD
PLAB
-27.5% / -17.4%
10.0% / 7.2%
27.7% / 18.5%
$10.20
RDVT
-24.5% / -13.7%
3.3% / 6.0%
18.2% / 16.5%
$23.02
CEVA
-30.0% / -18.2%
1.3% / 4.5%
16.7% / 15.3%
$9.34
DSP
-100.0% / -100.0%
16.8% / 12.2%
39.0% / 25.8%
$6.26
SLP
-26.4% / -16.3%
5.2% / 5.3%
21.6% / 16.4%
$5.01
AOSL
-100.0% / -100.0%
8.5% / 6.1%
30.8% / 19.9%
$8.86
CRNC
-44.2% / -100.0%
15.6% / 9.3%
36.0% / 22.0%
$3.19
AI
-100.0% / -100.0%
-17.2% / -6.2%
1.7% / 6.9%
$0.93
BBAI
-100.0% / -100.0%
-19.7% / -8.5%
-2.2% / 4.0%
$0.18
POET
Not modeled
Not modeled
Not modeled
N/A
Capital-gain CAGR
7.2%
Modeled exit price
$63.15
Hurdle entry
$10.20
Exit funding deficit $M
0.0
Annual research forecasts USD millions
Relative year
Growth
Revenue $M
Owner cash margin
Owner cash $M
Quote-equivalent shares M
Funding deficit $M
1
7.0%
924.5
12.0%
110.9
59.0
0.0
2
7.0%
989.2
13.0%
128.6
59.3
0.0
3
7.0%
1,058.4
14.0%
148.2
59.6
0.0
4
7.0%
1,132.5
15.0%
169.9
59.9
0.0
5
7.0%
1,211.8
16.0%
193.9
60.2
0.0
6
5.0%
1,272.4
16.0%
203.6
60.5
0.0
7
5.0%
1,336.0
16.0%
213.8
60.8
0.0
8
5.0%
1,402.8
16.0%
224.5
61.1
0.0
9
5.0%
1,473.0
16.0%
235.7
61.4
0.0
10
5.0%
1,546.6
16.0%
247.5
61.7
0.0
Assumptions, not reported margins
Base model inputs
Stock
Starting revenue x4 $M
Growth Y1-5 / Y6-10
Owner cash margin Y1 → Y5
Annual dilution
Exit cash multiple
PLAB
864.0
7.0% / 5.0%
12.0% → 16.0%
0.5%
15x
RDVT
106.8
18.0% / 11.0%
17.0% → 25.0%
2.0%
22x
CEVA
116.0
15.0% / 10.0%
-5.0% → 22.0%
2.0%
22x
DSP
417.0
16.0% / 10.0%
-2.0% → 10.0%
2.0%
20x
SLP
87.6
9.0% / 7.0%
12.0% → 20.0%
1.5%
18x
AOSL
681.6
8.0% / 6.0%
-3.0% → 9.0%
2.0%
16x
CRNC
278.4
8.0% / 6.0%
3.0% → 13.0%
2.5%
16x
AI
209.6
17.0% / 11.0%
-55.0% → 14.0%
5.0%
18x
BBAI
146.8
16.0% / 10.0%
-35.0% → 12.0%
5.5%
18x
POET
N/A
N/A
N/A
N/A
N/A
Scenario adjustments
Case
Growth shift
Cash-margin shift
Dilution shift
Exit multiple scale
Bear
-8.0%
-10.0%
2.5%
0.60x
Base
0.0%
0.0%
0.0%
1.00x
Bull
5.0%
4.0%
-1.0%
1.25x
Model equations: revenue baseline = quarter sales x 4; growth is compounded annually. Owner cash = revenue x assumed after-tax, after-interest, after-reinvestment cash margin, including normal compensation with no SBC addback. Margin ramps linearly from Y1 to Y5 and stays constant thereafter. Shares = quote-equivalent initial shares x compounded annual dilution (minimum zero issuance). Dilution is a separate conservative capital-structure stress, not a precise SBC reconciliation. Deficit funding = prior funding x 1.10 + negative owner cash. Positive cash is retained but not separately valued or distributed. Exit equity = max(0, next-year positive owner cash x multiple - funding); exit price = equity / end-horizon shares. CAGR = (exit price / entry price)^(1/years) - 1. Hurdle entry discounts that exit price at 15% for mid-cap or 20% for small-cap. These are reverse-engineered thresholds, not DCF fair values.
Sources, eligibility and limitations
Sources were reviewed through October 5, 2026. Issuer releases substantiate the quarterly facts, not the forecast assumptions or editorial rankings. Prices and capitalization are frozen October 2 quote snapshots, not live quotes. The selected universe is not an index and the size cutoffs are house conventions. No issuer-wide AI revenue percentage, consensus target or guaranteed highest-ROI list is asserted.
Important valuation omissions: no beginning cash, debt, preferred shares, tax receivable agreement or minority-interest bridge; no full GAAP forecast or historical financial database; quarterly seasonality is not normalized. BOX preferred claims, DSP Up-C/TRA economics, CRNC debt, and PLAB minority/cash-location issues require separate issuer work. POET uses 173.04M vendor-reported shares x $7.79 = $1,347.9816M, overriding a stale quote capitalization of about $475.9M. That count is not a fully diluted audit. Its financing and warrants require primary-register reconciliation. Model output must not be treated as a validated undervaluation conclusion.
Eligibility exclusions at the reviewed snapshot: IOT, CLS, FORM and AEIS exceed $10B; VERI, OSS, SSTK and REKR fall below $300M. INOD, AMBA and OUST are mid-caps rather than small-caps. These exclusions are about classification, not investment quality. A cap change after publication requires reclassification.
Review checklist: validate reported periods; reconcile share classes and dilution; test a recession and lower exit multiple; confirm funding access; track paid renewals or shipment acceptance; compare with a diversified benchmark. The bull case has no assigned probability. The bear case does not cap losses. General educational research, not personalized investment advice.