> ## Content Index
> Fetch the complete content index at: https://www.secondorderresearch.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# CEVA, Inc. ($CEVA): The Edge-AI Toll Booth Is Real. The Royalty Ramp Isn’t—Yet.
- URL: https://www.secondorderresearch.com/ceva-edge-ai-toll-booth-royalty-ramp/
- Published: 2026-09-02T18:06:42.000Z
- Updated: 2026-09-02T21:02:48.000Z
- Description: CEVA has a credible edge-AI and connectivity IP toll booth, but the investment case depends on whether today’s design wins convert into durable production royalties.
- Author: Second Order Research
- Tags: Full Underwriting, Public Equities

*Research Report · As of September 1, 2026 · Price: $26.23*

## Executive Research Snapshot

- **SoR Posture:** Watchlist / buy only on weakness; do not chase.
- **Structural Conviction:** Medium-high that CEVA owns valuable, durable connectivity and edge-AI IP.
- **Timing Conviction:** Low-medium because recent AI design wins have not yet become visible royalty streams.
- **Price Snapshot:** $26.23 at the September 1, 2026 close; approximately $738 million equity value.
- **Probability-Weighted Value:** $28.90 per share, or about 10% above the report price.
- **Bear / Base / Bull:** $11.20 / $29.70 / $53.40 per share.
- **Preferred Entry Zone:** $21–$23; compelling below $20 if the thesis remains intact.
- **Next Material Catalyst:** Evidence that a major NeuPro customer has reached tape-out, followed by a dated production schedule.
- **Evidence Confidence:** High on reported financials and existing shipment scale; medium on AI design-win quality; low-medium on the timing and size of future AI royalties.
- **Underwriting Status:** Preliminary watchlist. The evidence supports continued research, not a full position at the current price.

## Executive Summary

CEVA licenses silicon and software intellectual property used to connect, sense, and run inference inside edge devices. Customers generally pay an upfront license fee while designing a chip, followed by per-unit royalties if that chip enters production. In theory, this is an excellent model: high gross margins, low incremental cost per shipped unit, and royalty streams that can last for years.

The catch is latency. A license proves that CEVA won a technical evaluation; it does not prove that the customer will ship meaningful volume. Integration, tape-out, validation, qualification, and production can take 18–36 months after signing—and longer for automotive or custom silicon.

The bullish case has improved. CEVA signed ten NeuPro agreements in 2025, with AI contributing more than 20% of annual licensing revenue. Customers include Microchip, Nextchip, ALi, a leading PC manufacturer, and a major U.S. software and AI platform company using NeuPro-M in a custom-silicon program. Q2 2026 licensing revenue rose 21% to $18.2 million, while non-GAAP operating margin reached 11%.

The royalty evidence is less exciting. Q2 royalties were $10.8 million, only 1% higher year over year. CEVA-powered shipments reached 567 million units during the quarter, yet royalty dollars have not accelerated at the same rate. Product mix may explain part of the gap, but investors should not mistake shipment growth for economic proof.

At $26.23, CEVA has approximately $7.84 per share of cash and marketable securities and no traditional debt. Our probability-weighted value is $28.90, leaving only about 10% expected upside. The base case is close to priced in while the bear case remains material.

**My read:** CEVA belongs on a serious watchlist, but patience matters. The preferred entry zone is $21–$23\. Below $20, the setup becomes compelling if the thesis remains intact. Above roughly $38–$40, investors would be prepaying for a royalty ramp that has not yet been demonstrated.

### In This Report

1. Thesis and variant perception
2. Business model and conversion economics
3. Proven production versus licensed optionality
4. Counter-thesis, risks, and valuation
5. Catalysts, falsifiers, and living-research status

![](https://storage.ghost.io/c/13/2e/132e9dcb-4aeb-48fd-8b62-00d1f66e5254/content/images/2026/09/ceva-royalty-conversion-gap-1536x1024.png)

## Thesis

### The One-Sentence Thesis

CEVA’s connectivity franchise can fund a credible expansion into edge-AI IP, but the stock becomes compelling only when either the price discounts the long conversion cycle or the evidence proves that major AI licenses are becoming high-value production royalties.

### What the Market Appears to Price In

The market already recognizes that CEVA owns valuable IP and has a stronger balance sheet than its GAAP earnings suggest. It also assigns value to the company’s expanding AI customer base.

Q2 2026 revenue and adjusted earnings exceeded expectations, and management raised its full-year revenue-growth outlook to 13%–15%. Yet the stock fell sharply after the report.

My interpretation is straightforward: investors are no longer satisfied with licensing momentum alone. They want evidence of royalty conversion, sustainable profitability, and per-share value creation after dilution.

### Where I Disagree With the Bull Case

The common bull framing—“CEVA is a mini-Arm for edge AI”—is directionally useful but economically incomplete.

Both companies monetize embedded intellectual property through licenses and royalties. But CEVA has less ecosystem control, lower royalty scale, greater customer concentration, and much less evidence that its recent AI wins will generate material production economics.

The better framing is a smaller specialty toll booth with a valuable installed base and a call option on a new AI royalty layer.

That option is real. Ten NeuPro agreements and a custom-silicon engagement with a major U.S. AI platform company are meaningful technical validations. But investors should underwrite them using milestone probabilities—not capitalize the announcements as though production has already begun.

### The Decisive Question

Can CEVA convert today’s unusually strong licensing activity into a royalty stream large enough to lift revenue growth, GAAP profitability, and free cash flow per share before stock compensation and future dilution absorb the benefit?

## Business Model and Economics

CEVA sells two economically different products inside the same customer relationship.

1. **Licensing and Related Revenue** compensates CEVA for granting IP rights, customization, development tools, and support during a customer’s chip-design process. It is high-margin but can be lumpy.
2. **Royalty Revenue** is generally earned when the customer ships a chip or product containing CEVA IP. This is the compounding portion of the model because a successful design can ship for years with little incremental delivery cost.

### The Conversion Chain

**Evaluation → License → Integration → Tape-Out → Validation → Qualification → Production → Royalties**

Each arrow is a potential failure point.

- A **license** proves that CEVA won a commercial and technical evaluation.
- **Tape-out** proves that the design became physical silicon.
- **Qualification** proves that the chip works in its intended system.
- **Production** proves that an end market exists.
- **Royalties** prove that CEVA is receiving economic participation at scale.

Production commonly takes approximately 18–36 months after signing. Automotive and custom-silicon programs can take even longer. These are analytical estimates based on semiconductor-development cycles, not CEVA guidance for a specific customer.

### Why the Model Can Be Excellent

CEVA reported an 87% GAAP gross margin in Q2 2026\. More than 21 billion devices have shipped with its IP across Bluetooth, Wi-Fi, cellular IoT, sensor fusion, automotive, mobile, PC, and other applications.

When royalty revenue grows faster than operating expenses, the incremental margins can be substantial.

### Why the Model Can Disappoint

Royalty rates vary by product, customer, volume, and content. A low-value connectivity core shipping in hundreds of millions of units can generate less revenue than a higher-content AI design shipping at much lower volume.

That means “devices shipped” is not enough. We need to track:

- Total royalty revenue
- Royalty growth versus shipment growth
- Aggregate royalty revenue per shipped device
- Customer concentration
- Content growth
- Diluted share count

Using reported annual shipments and royalty revenue, aggregate royalty revenue per device declined from approximately 3.1 cents in 2021 to 2.2 cents in 2025 and about 2.0 cents in the first half of 2026.

This is a product-mix indicator—not a disclosed contractual royalty rate—but the direction matters: volume growth alone has not produced richer economics.

## Evidence

The most important distinction is between CEVA technology that is already producing royalties and newer AI agreements that remain design-stage optionality.

### Proven in Production

- **Scale:** More than 21 billion devices have shipped with CEVA IP; approximately 2.1 billion shipped in 2025.
- **Q2 Throughput:** CEVA-powered device shipments reached 567 million in Q2 2026, up 16% year over year.
- **Connectivity Franchise:** Management reported strength in Wi-Fi, Bluetooth, and cellular IoT, with cellular-IoT shipments reaching another quarterly record.
- **Automotive AI:** Existing automotive-AI programs continued to ramp. Renesas’ R-Car V4H platform in the 2026 Toyota RAV4 provides a visible production example.
- **Sensor Software:** CEVA’s MotionEngine software has shipped in more than 500 million devices.

These facts establish that CEVA is not a pre-revenue AI concept. It has production IP, established customers, and a functioning royalty engine.

### Licensed, but Not Yet Economically Proven

- **Ten NeuPro Agreements:** CEVA signed ten NeuPro agreements in 2025, with AI representing more than 20% of annual licensing revenue.
- **Microchip:** A broad NPU portfolio license expands CEVA’s addressable product set, but production units and royalty economics remain undisclosed.
- **Leading PC Manufacturer:** A major PC company selected NeuPro for a next-generation on-device AI architecture. The customer and production schedule remain undisclosed.
- **ALi and Nextchip:** These agreements validate NeuPro across consumer-video and automotive-ADAS applications, but CEVA has not separated their royalty contribution.
- **Major U.S. Software and AI Platform Company:** The July 2026 NeuPro-M custom-silicon agreement expands CEVA beyond traditional semiconductor customers. It remains a design program—not a disclosed production ramp.

### Current Financial Evidence

| Metric                    | Q2 2026 | Year-over-Year | What It Tells Us                                        |
| ------------------------- | ------- | -------------- | ------------------------------------------------------- |
| Revenue                   | $29.0M  | +13%           | Growth is accelerating                                  |
| Licensing and related     | $18.2M  | +21%           | Strongest quarter in more than three years              |
| Royalties                 | $10.8M  | +1%            | Conversion still lags the licensing enthusiasm          |
| GAAP gross margin         | 87%     | +1 point       | Attractive IP economics                                 |
| GAAP operating income     | \-$2.1M | Improved $2.4M | The business remains GAAP loss-making                   |
| Non-GAAP operating income | $3.1M   | +$2.3M         | Operating leverage is visible before stock compensation |
| Non-GAAP operating margin | 11%     | +8 points      | A useful proof point, but not yet a full-year state     |

For the first half of 2026, CEVA generated $56.1 million in revenue while recording a $7.2 million GAAP operating loss and a $7.4 million net loss.

Research and development equaled approximately 70% of revenue. Stock-based compensation was $10.5 million, or roughly 19% of revenue.

Those investments may create future value, but today’s owner economics remain weaker than the adjusted margins suggest.

### Balance Sheet and Dilution

At June 30, 2026, CEVA held:

- $44.3 million in cash
- $176.4 million in marketable securities and short-term deposits
- $220.7 million in combined cash and investments
- No traditional debt

The balance sheet meaningfully lowers financial risk.

However, CEVA had 28.14 million shares outstanding, up from 23.63 million at the end of 2024, primarily because of the late-2025 equity offering and subsequent equity issuance.

The cash matters—but so does the dilution. Progress must ultimately be measured per share.

### Evidence-Confidence Map

| Claim                                         | Evidence Confidence | Underwriting Treatment         |
| --------------------------------------------- | ------------------- | ------------------------------ |
| Connectivity IP has durable production demand | High                | Included in every scenario     |
| Licensing momentum is currently strong        | High                | Included in 2026 expectations  |
| NeuPro wins validate technical capability     | Medium-high         | Included as strategic evidence |
| Major AI programs will reach production       | Medium              | Probability-weighted           |
| AI royalties become material by 2030          | Low-medium          | Main base/bull swing factor    |
| CEVA earns Arm-like economics                 | Low                 | Not underwritten               |

## Counter-Thesis and Risks

### The Strongest Counter-Thesis

CEVA may remain a technically respected niche licensor whose most exciting announcements repeatedly fail to become material per-share cash flow.

Customers can delay programs, choose low-volume applications, negotiate low royalty rates, or replace external IP with internally developed technology. Meanwhile, CEVA must continue spending heavily on research and development to remain competitive.

In that outcome, the company continues announcing impressive design wins while the theoretical royalty leverage never fully appears.

### Principal Risks

1. **Design-Win Conversion Risk**  
A license is not a shipment forecast. CEVA’s most important new AI programs lack disclosed tape-out, qualification, production, and unit milestones.
2. **Royalty-Mix Risk**  
Aggregate royalty revenue per shipped device has trended downward as lower-value connectivity products have dominated unit volume.
3. **Customer Concentration**  
CEVA’s five largest royalty-paying customers generated 54% of Q2 2026 royalty revenue. A product-cycle change at only a few customers can overwhelm progress elsewhere.
4. **China Exposure**  
China represented 61.9% of 2025 revenue and 52% of first-half 2026 revenue. Trade restrictions, local substitution, geopolitical risk, and pricing pressure all matter.
5. **Competition and Insourcing**  
Arm, Cadence, Synopsys, proprietary accelerators, open architectures, and customer-developed IP compete for architecture control and economics.
6. **Profit Quality**  
CEVA remains GAAP loss-making. First-half stock compensation equaled approximately 19% of revenue, meaning non-GAAP profitability materially overstates current owner earnings.
7. **Dilution**  
The outstanding share count increased substantially following the 2025 capital raise. Future equity issuance could reduce the per-share payoff even if the business grows.
8. **Valuation-Duration Risk**  
The market can compress CEVA’s valuation before AI royalties arrive—particularly if licensing revenue proves lumpy or production timelines slip.

### Why the Bear Case Does Not Require Technological Failure

The bear case does not require CEVA’s technology to fail.

It requires only that high-profile licenses produce modest units, low content, delayed launches, or offsetting declines elsewhere.

That is what makes this underwriting difficult: CEVA can continue winning technically while disappointing investors economically.

## Valuation and Scenarios

### Current Valuation

At $26.23 and 28.14 million shares outstanding, CEVA’s equity value is approximately $738 million.

Subtracting $220.7 million of cash, marketable securities, and deposits gives a net-cash enterprise value of approximately $517 million before operating leases. Including $17.4 million of operating-lease liabilities produces an adjusted enterprise value near $535 million.

Management raised its 2026 revenue-growth outlook to 13%–15%. Applied to 2025 revenue of $109.6 million, that implies:

- **2026 revenue range:** $123.8–$126.0 million
- **Midpoint:** Approximately $124.9 million
- **Net-cash EV/revenue:** Approximately 4.1×
- **Lease-adjusted EV/revenue:** Approximately 4.3×

CEVA’s net cash equals approximately $7.84 per current basic share. Investors are therefore paying about $18.39 per share for the operating business.

That operating value is justified only if the licensing cycle becomes a royalty and cash-flow cycle.

### Scenario Framework

The model uses 2030 operating assumptions, an 11% discount rate, and scenario-specific terminal revenue multiples. It is an underwriting framework—not a precise forecast.

| Scenario | Probability | 2030 Revenue | Operating Margin | Free Cash Flow | Diluted Shares | Terminal EV/Sales | Present Value/Share |
| -------- | ----------- | ------------ | ---------------- | -------------- | -------------- | ----------------- | ------------------- |
| Bear     | 30%         | $135M        | 3%               | $9M            | 32.0M          | 2.5×              | $11.20              |
| Base     | 50%         | $195M        | 18%              | $36M           | 31.5M          | 6.0×              | $29.70              |
| Bull     | 20%         | $280M        | 28%              | $78M           | 31.0M          | 8.0×              | $53.40              |

The probability-weighted value is **$28.90 per share**.

Against the report price of $26.23, that represents approximately **10% expected upside**.

- **Bear return:** Approximately -57%
- **Base return:** Approximately +13%
- **Bull return:** Approximately +104%

### What Each Scenario Requires

**Bear Case:** Connectivity remains useful but mature. AI licenses contribute little royalty revenue, R&D intensity remains high, and dilution continues. The cash balance reduces financial risk but does not prevent multiple compression.

**Base Case:** A meaningful subset of NeuPro and automotive programs reaches production. Royalty growth accelerates and operating leverage emerges, but CEVA does not achieve Arm-like economics.

**Bull Case:** Multiple major AI programs ship at meaningful volume and higher content. CEVA develops a richer royalty mix, strong incremental margins, and a premium semiconductor-IP valuation.

### Entry Discipline

At **$21–$23**, the weighted value offers approximately 26%–38% upside and better compensation for the conversion risk.

Below **$20**, the balance sheet would represent close to 40% of the company’s market capitalization, creating a more asymmetric setup if the thesis remains intact.

Above approximately **$38–$40**, investors would be paying in advance for much of the base-to-bull conversion.

### What Matters Most

The largest valuation sensitivities are not small adjustments to the discount rate. They are:

1. How many major AI programs reach production
2. Royalty dollars generated per device
3. Sustainable operating margins after stock compensation
4. The diluted share count

A single high-volume, high-content custom-AI program could materially improve the bull case. A prolonged absence of production evidence would push the model toward the bear case.

![](https://storage.ghost.io/c/13/2e/132e9dcb-4aeb-48fd-8b62-00d1f66e5254/content/images/2026/09/ceva-valuation-scenarios-1536x1024.png)

## Catalysts

The most important catalysts are not additional license announcements by themselves. They are milestones showing that existing licenses are moving closer to production economics.

- A disclosed NeuPro tape-out or silicon-validation milestone from the major U.S. AI platform customer
- A dated production schedule for the leading PC manufacturer, Microchip portfolio, or another major NPU licensee
- Royalty growth beginning to exceed unit-shipment growth, indicating improving content or product mix
- Stabilization or improvement in aggregate royalty revenue per shipped device
- Full-year operating leverage progressing toward GAAP profitability rather than depending on stock-compensation exclusions
- Greater disclosure around expected units, content, production stage, or royalty timing—even if CEVA cannot identify the customer
- Additional automotive production ramps
- Broader production adoption of newer Wi-Fi, Bluetooth, and cellular-IoT standards

## Falsifiers

A good thesis needs observable reasons to change our mind.

The CEVA thesis should be reduced or abandoned if any of the following occurs without a credible offset:

1. **No tape-out evidence by year-end 2027**  
The major 2026 custom-AI program fails to produce a verifiable tape-out or equivalent silicon milestone.
2. **No production timetable by year-end 2028**  
The program still lacks a credible production schedule or initial royalty evidence.
3. **AI licensing does not become AI royalties**  
AI continues representing more than 20% of licensing revenue but fails to become a visible royalty contributor by 2028.
4. **Royalties continue trailing shipments**  
Royalty revenue persistently grows more slowly than CEVA-powered device shipments, while aggregate royalty revenue per device fails to stabilize.
5. **Scale does not produce GAAP profitability**  
Revenue exceeds $150 million without a credible path to sustained GAAP operating profit.
6. **Stock compensation and dilution remain excessive**  
Stock compensation stays above 15% of revenue while the diluted share count grows more than 3% annually.
7. **A major customer cancels, delays, insources, or replaces CEVA IP**  
This would directly weaken the expected production and royalty pipeline.
8. **CEVA raises additional equity before becoming self-funding**  
Another capital raise without a clearly value-accretive use would weaken the per-share thesis.
9. **China risk worsens materially**  
China concentration rises while export restrictions, local substitution, or pricing pressure impair customer economics.

## Living Research Status

- **Last Updated:** September 2, 2026, using market data through September 1 and financial data through June 30, 2026.
- **Company-Thesis Status:** Intact; structurally credible, but the timing remains unproven.
- **Security-Thesis Readiness:** Conditional.
- **SoR Action:** Wait for proof / buy only on meaningful weakness.
- **New Evidence:** Q2 licensing reached a three-year high, the custom-AI platform agreement expanded CEVA’s customer set, and management raised 2026 revenue-growth guidance.
- **Scenario Probabilities:** Unchanged at 30% bear / 50% base / 20% bull.
- **Probability-Weighted Value:** $28.90.
- **Preferred Entry Zone:** Unchanged at $21–$23.
- **Falsifier Status:** None triggered.
- **First Major Time-Based Test:** Tape-out evidence by year-end 2027.
- **Next Review Trigger:** Q3 2026 results, a NeuPro production milestone, or a share-price move into the preferred entry zone.

## Sources and Methodology

This report prioritizes SEC filings, company-issued financial results, and primary company disclosures. Market-expectation context is identified separately. All scenario values are Second Order Research estimates—not company guidance.

### Primary Sources

- [CEVA 2025 Form 10-K](https://www.sec.gov/Archives/edgar/data/1173489/000143774926006091/ceva20251231%5F10k.htm?ref=secondorderresearch.com), filed February 27, 2026
- [CEVA Q2 2026 Form 10-Q](https://www.sec.gov/Archives/edgar/data/1173489/000143774926026774/ceva20260630%5F10q.htm?ref=secondorderresearch.com), filed August 10, 2026
- [CEVA Q2 2026 Financial Results](https://www.ceva-ip.com/press/ceva-inc-announces-second-quarter-2026-financial-results/?ref=secondorderresearch.com), published August 10, 2026
- [Landmark Custom-AI Licensing Agreement](https://www.ceva-ip.com/press/ceva-wins-landmark-ai-licensing-deal-with-major-u-s-software-and-ai-platform-company/?ref=secondorderresearch.com), published July 6, 2026
- [CEVA’s 2025 AI Licensing Review](https://www.ceva-ip.com/press/ceva-highlights-breakthrough-year-for-ai-licensing-and-physical-ai-adoption-in-2025/?ref=secondorderresearch.com), published February 17, 2026
- [CEVA-NeuPro-M Product Page](https://www.ceva-ip.com/product/ceva-neupro-m/?ref=secondorderresearch.com), retrieved September 2, 2026

### Analytical Conventions

- Enterprise value is shown both before and after operating-lease liabilities.
- Aggregate royalty revenue per device is calculated using reported royalty revenue and shipment volume. It is a product-mix indicator—not a disclosed contractual royalty rate.
- Scenario probabilities represent underwriting judgment, not statistical forecasts.
- We do not speculate about the identity of unnamed customers.
- Future tape-out, qualification, production, and royalty timing are treated as estimates unless CEVA explicitly discloses them.
- Evidence confidence and investment readiness are evaluated separately. Strong technical evidence does not automatically make the stock attractive at every price.

### How This Research Was Produced

Second Order Research is AI-native and human-directed.

AI performed much of the evidence gathering, document review, calculation, comparison, and drafting support. The final thesis, scenario probabilities, valuation judgment, entry discipline, and decision to publish remain human judgments.

The purpose is not to remove judgment from investing. It is to give judgment better evidence.

## Disclosure

Second Order Research provides independent investment research for informational and educational purposes only.

This report is not investment advice, a recommendation, or an offer to buy or sell securities. Forecasts, probabilities, and valuations are uncertain and may be wrong. Positions and views may change as evidence changes.

Readers should perform their own due diligence and consider their financial circumstances, objectives, and risk tolerance.

##