CEVA Inc

Phase A — Understand the business

Lens 1 · Company Overview

Model. CEVA is a two-line IP business: (1) Licensing & related revenue — upfront fees for access to IP cores, platforms, software and tools, plus customization/support; recognized mostly at a point in time on delivery, some over time on customization. (2) Royalty revenue — recurring per-unit or %-of-ASP royalties on licensee shipments, recognized in the quarter the licensee ships (often on CEVA's estimate, trued-up later). FY2025 split: licensing $63.6M (58.0% of revenue), royalty $46.0M (42.0%). This is the classic IP "razor-and-blades on someone else's razor" model — CEVA never touches silicon, fabs nothing, carries no inventory, and books ~87% gross margin.

What it actually sells, by pillar (the company's own "Connect / Sense / Infer" framing):

• Connect (the crown jewel): Ceva-Waves Bluetooth, Wi-Fi (4→7), UWB, multi-radio (Links), cellular-IoT (DragonFly NB-IoT), and 5G (XC20 DSP, PentaG2 UE, PentaG-RAN L1 PHY). CEVA claims #1 worldwide wireless-connectivity IP share at 68% (2024); ~30% of Bluetooth IoT devices, ~50% of non-Apple TWS earbuds, ~40% of cellular-IoT devices, and ~30% of global 5G RAN DSP run on its IP. Treat the 68%/50%/40% figures as company self-reported market claims, not audited.
• Sense: SensPro AI/sensor-fusion DSPs, MotionEngine (IMU fusion — the ex-Hillcrest Labs asset), RealSpace (spatial audio — ex-VisiSonics), BX1/BX2 audio/control DSPs. >500M devices to date on its sensing/audio software.
• Infer (the option): NeuPro-Nano (embedded NPU for MCUs/wearables) and NeuPro-M (high-perf NPU for generative/transformer/ADAS), with the NeuPro Studio SDK. This is the newest and the reason the stock re-rated.

Customers. Hundreds of licensees; "seven of the world's top-10 MCU vendors." Named licensees include Broadcom, Qualcomm, NXP, STMicro, Samsung, Sony, Renesas, Microchip, MediaTek, Nordic, onsemi, Panasonic, Realtek, Unisoc, Intel. Concentration is the headline risk: UNISOC = 15% of total revenue (FY2025 and FY2024), and just two royalty payers = 39% of total royalty revenue (FY2025). Top-5 royalty payers = 57% of royalty revenue.

Suppliers. CEVA has essentially no COGS supply chain (it's IP) — its "inputs" are engineers and EDA tools. It carries a $11.2M design-tool purchase obligation (Synopsys/Cadence-class EDA licenses). See Lens 2.

Contract structure. No take-or-pay; no long-term minimums disclosed. Licensing is lumpy deal-by-deal (54 deals signed FY2025). Royalty is the recurring annuity but lags device cycles. Deferred revenue is tiny ($3.5M) — this is not a SaaS-style ratable book; it's point-in-time IP delivery plus a trailing royalty stream.

Lens 2 · Supply Chain

CEVA sits at the very top of the semiconductor value chain — it is an input, not a link that touches physical goods. The "supply chain" that matters for CEVA is the demand chain: whose silicon carries its IP to volume.

Map (upstream inputs → CEVA → path to end-device):

• Upstream inputs into CEVA: (a) scarce RF/DSP/NPU design talent — 311 R&D staff, mostly Israel (363 of 424 total employees are EME); (b) EDA tools — Synopsys and Cadence design-tool licenses ($11.2M committed); (c) wireless standards bodies (Bluetooth SIG, Wi-Fi Alliance, 3GPP, FiRa/UWB) whose specs CEVA implements first-to-market; (d) the Israeli Innovation Authority and French CIR / UK R&D credits that subsidize R&D (IIA royalty-back at 3–3.5% of sales on grant-funded products).
• CEVA → licensee (the actual customers): semiconductor houses and OEMs who bake the IP into ASICs/ASSPs/SoCs — named: Broadcom, Qualcomm, NXP, STMicroelectronics, Samsung, MediaTek, Renesas, Microchip, Nordic Semiconductor, onsemi, Realtek, Sony, Unisoc, Intel.
• Licensee → foundry → OEM → end device: those SoCs are fabbed (TSMC/Samsung/GF etc. — not disclosed by CEVA, it's the licensee's foundry), then land in TWS earbuds, wearables, smartphones, Wi-Fi routers, NB-IoT meters, ADAS modules, AI PCs.

Chokepoints / single-source dependencies:

1. UNISOC is the single most important node — 15% of revenue, a Chinese baseband/handset SoC vendor. CEVA's cellular-IoT and handset-DSP royalty stream leans on it. This is both a concentration risk and a China-exposure amplifier.
2. China as a geography = 62% of FY2025 revenue ($67.9M of $109.6M), up from 49% in FY2024. APAC overall = 77%. This is a genuine single-country dependency the company flags explicitly as elevating its risk profile.
3. Israel R&D concentration — most core wireless/DSP IP is developed in Israel; the 10-K carries an extensive Middle-East-instability risk factor and notes reservist call-ups. Operations "largely unaffected" through the Oct-2025 Hamas ceasefire but volatile.
4. EDA tool dependency — CEVA cannot design without Synopsys/Cadence tooling, i.e. it is structurally downstream of the very EDA giants (SNPS/CDNS) that also compete with it in IP.

Names or it didn't happen — done. The chain is unusually clean on the cost side (no fabs, no wafers, no materials except the Palladium/Russia sanction note CEVA flags as an industry-wide, not CEVA-specific, risk) and unusually concentrated on the demand side (UNISOC + China).

Lens 3 · Competitive Advantages (moats)

Where the moat is real (Connect):

• Standards-first scale + design-win lock-in. Once a licensee designs CEVA's Bluetooth/Wi-Fi PHY/MAC into an SoC, switching means re-spinning silicon — "significant cost, time, effort and risk," per the 10-K's own design-win language. 20B+ cumulative devices since 2003 and 2B/year today is a deep installed base. The claimed 68% connectivity-IP share, if roughly right, is a durable oligopoly position built on being first to each new standard (Wi-Fi 7, BT 6/7, UWB, 5G-Advanced, NTN).
• Breadth as switching cost. 12 customers licensed multiple CEVA technologies in 2025; combo (BT+Wi-Fi) deals raise value-capture per design and deepen entanglement. The "Physical AI fabric" pitch is really a bundling strategy to sell Sense+Infer into the vast Connect account base.
• Talent scarcity. RF/PHY/DSP/NPU design is genuinely hard to build in-house; CEVA's pitch is "we amortize a scarce team across the whole industry." Credible for connectivity.

Where the moat is thin or contested (Infer / Sense):

• NPU IP is a knife-fight, not a monopoly. The 10-K itself lists the field: Arm (Ethos), Cadence (Tensilica AI/Vision DSP), Synopsys/GlobalFoundries, Imagination, VeriSilicon, Andes, SiFive, Expedera, Quadric, Cambricon, plus NVIDIA's open-source NVDLA. Arm holds ~24% of NPU IP share with the transformer-native Ethos-U85. CEVA is a credible edge-AI IP vendor, not the leader here.
• RISC-V is the structural threat CEVA names on itself. The 10-K explicitly warns that "increasing adoption of open-source processor architectures, including RISC-V-based ecosystems, may intensify pricing pressure and increase customer alternatives to licensable IP solutions". RISC-V hit ~25% processor penetration by Jan 2026 and is targeting edge-AI inference — the exact fastest-growing segment CEVA is chasing.

Bargaining power (who needs whom more):

• Over customers: moderate and asymmetric. In Connect, CEVA has real leverage — it's often the only production-proven path to a new standard on time. In AI, the customer has options (build on RISC-V, license Arm/Cadence, or design in-house), so CEVA is the price-taker.
• Over suppliers: weak. It depends on Synopsys/Cadence EDA and on scarce labor whose cost is FX-exposed (NIS/EUR); a 10% USD weakening adds ~$5.2M to opex.
• Customer consolidation cuts against it — the 10-K flags that semiconductor consolidation raises customers' negotiating leverage and deepens CEVA's dependence on a few names.

Verdict on the moat: a wide, durable moat in wireless-connectivity IP wrapped around a narrow, contested option in edge-AI NPU IP. The bull thesis rests on migrating the Connect moat's account base into Sense/Infer before RISC-V/Arm commoditize the AI layer.

Lens 4 · Segments

CEVA reports ONE operating segment (IP licensing; the CODM — the CEO — manages on consolidated net income). There is no product-line EBITDA to break out. The meaningful disaggregations are (a) by revenue TYPE, (b) by USE-CASE, and (c) by GEOGRAPHY.

(a) By revenue type ($M):

Trend: licensing is the growth engine (record licensing quarters); royalty decelerated in FY2025 (−2%) on smartphone softness + memory shortage capping unit shipments, despite record 2.1B devices shipped (+6% units). The tension: units up, royalty dollars down — mix shifted to lower-ASP categories (Bluetooth −2% units, handsets −18% units, offset by Wi-Fi +48% and cellular-IoT +42%).

(b) By use-case (% of revenue):

This is the single most important trend line in the whole dossier: Sense & Infer went 16%→25% of revenue in one year — the diversification-away-from-pure-connectivity thesis is visibly working. AI alone was >20% of licensing revenue in FY2025 (10 NPU deals).

(c) By geography ($M / % of total):

The China share jumped 49%→62% in a single year — accelerating, and the wrong direction for geopolitical risk. EME collapsed (−51%) on a lumpy FY2024 infrastructure deal not repeating.

Q1 2026 use-case split: Connect $19.8M, with AI-related licensing again >20% of licensing revenue — the mix shift is holding into 2026.

---

Phase B — Measure performance

Lens 5 · Earnings Result

Latest print: Q1 2026 (reported 2026-05-11):

vs consensus: revenue $27.0M beat the ~$26.6M Street estimate; non-GAAP EPS $0.04 beat the ~$0.018 estimate. Management called it "our strongest licensing quarter in three years".

The GAAP-vs-non-GAAP fork is the whole story of this name. On GAAP, CEVA loses money every quarter and every year. The bridge is almost entirely stock-based comp: FY2025 GAAP net loss −$10.6M vs non-GAAP net income +$10.8M, non-GAAP diluted EPS $0.42. FY2025 SBC was $19.8M — larger than the entire non-GAAP profit. So "profitability" here means "profitable if you ignore that ~18% of revenue is paid to employees in stock."

Drivers: licensing acceleration (Wi-Fi 7 combo deals, first full Bluetooth-7 integrated solution, PentaG-NTN satellite modem intro, AI/automotive NPU deals). Royalty flat — the annuity isn't yet inflecting despite record units.

Guidance/outlook: management guided FY2026 revenue growth to the top of an 8–12% range and is targeting 40–50% non-GAAP profit growth. That implies operating leverage finally showing up in the model as licensing scales on a fixed R&D base.

Balance-sheet flags — clean. $215.7M–$222.0M cash + deposits + marketable securities, no financial debt (the ~$16.9M "total debt" in screeners is operating-lease liability, not borrowings). But note: FY2025 operating cash flow was negative $3.4M — CEVA burned a little cash from operations, and has now done so two years running (FY2024 −$3.5M). The cash pile grew only because of a $63.3M November-2025 follow-on equity raise (3.45M shares), not operations. Trade receivables jumped to $49.4M from $37.2M (+33%) — outrunning the +2.5% revenue growth, a yellow flag on collection timing (see Lens 10).

Market reaction — the tell. Despite the Q1 beat, the stock declined on the print, and it had fallen ~9.8% on the day CEVA touted its full-year AI-licensing results (intraday −17.3% at the trough). The market is pricing perfection and punishing anything short of a royalty acceleration.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts on the research-layer shelf (transcripts/ empty). Reconstructed from filings' MD&A + web transcript coverage; labeled accordingly.

What management is focused on (consistent across FY2025 → Q1 2026): "Physical AI" as the unifying narrative; the licensing→royalty "flywheel"; NPU/AI licensing as the growth vector; connectivity (Wi-Fi 7, BT High-Data-Throughput, cellular-IoT records) as the base; a stated $125M aggregate lifetime-royalty-potential from FY2025's signed deals.

Tone shift over the last ~4 quarters:

• Q3 2025: revenue rose but EPS missed — cautious tape reaction.
• Q4 2025: record Q4 revenue ($31.3M), strongest royalty quarter in 4+ years, non-GAAP EPS $0.18 (+71% YoY) — stock rose, most upbeat tone of the year.
• Q1 2026: beat on revenue and non-GAAP EPS, "strongest licensing quarter in three years," but stock declined — the market wanted royalty inflection, got licensing.

Recurring phrases: "Physical AI," "connect/sense/infer," "flywheel," "record" (Wi-Fi/cellular-IoT shipments), "NeuPro." New in FY2025-26: "Physical AI" wholesale replaced the older "Smart Edge / Edge AI" framing — a deliberate narrative upgrade to ride the robotics/embodied-AI theme. What they've stopped emphasizing: the standalone 5G-RAN infrastructure story (advanced-5G licensing was a headwind in FY2025 as the big FY2024 infra deal didn't repeat).

Net: management sentiment is genuinely more confident (backed by real licensing wins and the PC-OEM NPU marquee deal), but the stock's sentiment is skeptical — a classic "prove the royalty" standoff.

Lens 7 · Comps

CEVA's true peer set is the semiconductor-IP / EDA licensing group, not broad semis. Multiples are `` with source/date; where I cannot source a clean figure I write n/a rather than fabricate.

5-yr avg ROE: n/a cleanly for the peer set, and misleading for CEVA anyway (GAAP net loss → negative ROE despite positive non-GAAP income; equity base just inflated by the follow-on).

Read: CEVA at ~8x EV/Sales sits far below Arm (~38x) and roughly in line with SNPS/CDNS on sales but at a lower absolute multiple than the mega-cap IP names — which is the bull's "cheapest way to own the IP-royalty theme" pitch (Needham's exact framing). The bear read: 8x sales and ~66x forward non-GAAP EPS is a rich multiple for a company growing revenue ~2.5% (FY2025) / ~11% (Q1'26) with a GAAP loss, no dividend, and 62% China exposure. The multiple is an AI-optionality multiple, not a fundamentals multiple.

Lens 8 · Stock-Price Catalysts

CEVA is a small-cap ($1.1–1.3B) that moves violently on single events. Pattern from the last ~18 months:

• 52-week range ~$17.02 → ~$51.60 — a near-3x move off the low. The stock has +85.8% over the trailing 52 weeks. This is not a sleepy IP name; it re-rated hard on the AI-NPU narrative.
• NPU/AI licensing wins = the up-catalyst. The Microchip full-NeuPro-portfolio license and the "strategic NeuPro NPU agreement with one of the world's leading PC OEMs" (FY2025) were the marquee re-rating events.
• Earnings reactions are bimodal and royalty-sensitive. Q4 2025 (royalty record) → stock up; Q3 2025 (EPS miss) and Q1 2026 (beat but licensing-led, not royalty-led) → stock down/soft. The −17.3% intraday trough on the AI-results day shows how twitchy it is.
• The Nov-2025 follow-on ($63.3M, 3.45M shares at ~$21-area) was a dilution event but funded from a position of narrative strength.

What the market actually reacts to: (1) net-new AI/NPU design wins (up), (2) royalty acceleration or the lack of it (the current swing factor), (3) China/tariff headlines (down-risk given 62% exposure), (4) any Israel-conflict escalation (idiosyncratic down-risk). Macro/rate moves matter less than for most semis because there's no fab capex cycle — it's a design-win and royalty-timing story.

---

Phase C — Judge people & books

Lens 9 · Management

CEO — Amir Panush (age 52, CEO since Jan 2023, board since Feb 2024). Pedigree is a strong fit: ex-InvenSense/TDK (CEO & GM of TDK's MEMS Sensors business post-acquisition), before that strategy/corp-dev at InvenSense, and Qualcomm (Sr Director, IoE/IoT client business) and Atheros (strategic marketing/partnerships) and Texas Instruments. Haas MBA, Technion CS (Cum Laude). This is a sensors-+-wireless-+-IoT operator — precisely the "Physical AI / connect-sense-infer" convergence CEVA is now selling. Track record on his watch: he inherited a stalling business (FY2023 revenue $97.4M, GAAP loss −$11.9M) and has grown revenue to $109.6M and swung non-GAAP income to +$10.8M / +20% YoY, while repositioning the story from "DSP IP" to "Physical AI" and landing the first material NPU wins. Real, if still early, turnaround signal.

CFO — Yaniv Arieli (age 57, CFO since May 2005 — 20+ year tenure). CPA, ex-PwC (Kesselman). Deep continuity and a clean audit history (EY/Kost Forer as auditor since 1999, unqualified opinions, effective ICFR). COO Michael Boukaya (since 2019, ex-DSP Group, wireless BU) and CCO Gweltaz Toquet (since 2023, 20+ yrs at CEVA, ran APAC sales from Hong Kong) round out a technically deep, long-tenured bench.

Skin in the game / comp. Insider ownership is modest (no insider-transactions.csv on the shelf; not a founder-controlled company — CEVA is a 2002 spin-merger of DSP Group's IP division + Parthus). Comp is heavily equity/PSU-driven with rigorous, disclosed performance gates: FY2025 PSUs vested on (a) a Board-approved license-revenue target, (b) TSR vs the S&P Semiconductors Select Industry index, and (c) TSR vs the Russell 2000 — with over-achievement kickers up to 110%. That's a genuinely well-aligned, index-relative comp design (better than most small-caps). The flip side: $19.8M SBC on $109.6M revenue is a heavy dilution cost that keeps GAAP in the red and grew the share count.

Capital allocation. (1) R&D-first — R&D is 68% of revenue, the explicit strategic priority. (2) Buybacks — modest and consistent (340k shares / $7.2M in FY2025 at ~$21 avg; 375k / $8.5M in FY2024) — but the FY2025 buyback ($7.2M) was dwarfed by the FY2025 follow-on raise ($63.3M) and the SBC ($19.8M), so net share count rose sharply (23.6M → 27.5M shares out). (3) Bolt-on M&A — Hillcrest Labs (MotionEngine), VisiSonics (RealSpace/spatial audio), a small Greek company; divested Intrinsix (ASIC design services) — a disciplined "buy IP/algorithms, shed services" pattern. Grade: capital-allocation intent is sound (invest in IP, stay asset-light), but the equity-issuance + SBC combination means shareholders are funding the growth via dilution while the buyback is largely cosmetic.

Red flags on people: none material. No related-party dealing, no promotional CEO behavior beyond the (industry-standard) "Physical AI" narrative inflation, no strategy whiplash. The main governance cost is dilution, not integrity.

Lens 10 · Forensic Red Flags

Forensic pass across the three statements.

• Revenue recognition — the auditor's Critical Audit Matter. EY flagged rev-rec (identifying distinct performance obligations + the residual method for allocating transaction price to IP licenses) as the sole CAM. This is inherent to IP licensing, not a specific problem, but it's the judgement-heavy line. Additional softness: royalty revenue is frequently recognized on management's estimate of licensee shipments before actual reports arrive, trued-up later. Estimated (not yet reported) royalties are an area to watch for later reversals, though history shows no material restatements.
• Receivables outrunning revenue — the clearest yellow flag. Trade receivables grew +33% (to $49.4M from $37.2M) on +2.5% revenue in FY2025. DSO is stretching (receivables now ~45% of annual revenue), concentrated in APAC/China. The allowance for credit losses was actually cut ($2.6M→$0.3M) even as receivables ballooned — i.e. the balance grew and the reserve shrank. Given 62% China exposure and first-time APAC customers the 10-K itself flags for uncertain creditworthiness, this is the single line most worth monitoring. (Q1 2026 receivables held ~flat at $48.9M, so it's stretched but not still-worsening.)
• Cash flow vs earnings — negative, and worse than the loss suggests on a "quality" basis. Operating cash flow was −$3.4M in FY2025 (−$3.5M FY2024). The GAAP loss (−$10.6M) is "rescued" toward breakeven cash only by $19.8M non-cash SBC + $4.3M D&A; working capital (the receivables build) was a $15.2M drag. So on a cash basis CEVA is roughly breakeven-to-slightly-negative — the "record year" did not generate operating cash. The cash balance grew purely on the equity raise.
• SBC flattering non-GAAP — material. $19.8M SBC (FY2025) is the entire delta between the −$10.6M GAAP loss and the +$10.8M non-GAAP profit. Anyone valuing CEVA on non-GAAP EPS (~66x forward) is capitalizing a number that adds back a real, recurring, dilutive expense. $25.8M of unvested RSU/PSU comp remains to be recognized over ~1.5 yrs — SBC stays elevated.
• Goodwill/intangibles — low risk. Goodwill $58.3M (flat, no impairment), intangibles net just $1.0M and amortizing to near-zero. Goodwill is ~15% of total assets but well-covered by cash; no impairment triggers evident.
• Leases — modest, and a jump worth noting. Operating-lease ROU assets jumped $5.8M→$17.5M (a new Israel lease) — bigger fixed footprint, but immaterial to the thesis.
• Tax — a small negative signal. CEVA booked a $1.0M valuation allowance against its French subsidiary's deferred tax assets in FY2025 (uncertainty over future French taxable income). Minor, but a tacit admission one geography isn't generating profit. Overall tax is oddly high vs pretax loss (FY2025 tax $5.9M on a −$4.7M pretax loss) because of withholding taxes and GILTI/Sec-174 dynamics on international R&D — structurally raises the effective drag.

Regulatory findings (required sub-section):

• SEC: Read regulatory/regulatory-findings.md (generated 2026-07-06). Zero SEC Litigation Releases and zero AAERs naming CEVA Inc in the 2021-07-06 → 2026-07-06 window (SEC EDGAR EFTS, LR + AAER).
• 10-K Item 3 (Legal Proceedings): CEVA's own disclosure — only ordinary-course claims with a ~$300,000 provision; "not a party to any other legal proceedings" with a material adverse potential. Immaterial.
• Non-SEC enforcement (web search): No material FTC/DOJ/FDA/CFPB action against CEVA Inc (the IP company) found. Important disambiguation: search results surfacing a "$1.5M California wage-and-hour settlement" and Brazil/Italy labor disputes belong to CEVA Logistics — an unrelated freight-forwarding company that happens to share the name — not CEVA Inc the semiconductor-IP firm. Do not attribute those to this issuer.
• Conclusion: No material regulatory or legal findings for CEVA Inc — verified via SEC EDGAR EFTS (LR, AAER), web search, and 10-K Item 3 as of 2026-07-06. Auditor issued unqualified opinions on both financials and ICFR (EY, auditor since 1999).

Forensic verdict: clean on integrity/enforcement; the quality-of-earnings flags are the story — (1) non-GAAP profit that vanishes under GAAP because of SBC, (2) negative operating cash flow two years running, and (3) a China-heavy receivables build with a shrinking reserve. None is fraud-shaped; all three matter for how you value the "profitability" management is selling.

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

Built bottom-up from FY2025 actuals + FY2026 guidance. All outputs `` with arithmetic shown; inputs labeled. No forecast.ts create in watchlist mode (per skill).

Anchors:

• FY2025 actual revenue $109.6M; non-GAAP EPS $0.42; non-GAAP NI $10.8M; ~27.5M shares out (growing).
• Guidance: FY2026 revenue growth at top of 8–12% → ~12%; targeting 40–50% non-GAAP profit growth.
• Q1 2026 revenue +11.5% YoY, licensing +18.5%, royalty flat.
• Share count creep: ~+3-4% annually from SBC net of buybacks.

Revenue FY2026 / FY2027 / FY2028:

• Base: FY26 ~$123M (+12%, matches consensus ~$123M ); FY27 ~$137M (+11%, royalty finally inflecting as Wi-Fi 7 + NPU deals ramp); FY28 ~$152M (+11%).
• Bull: FY26 ~$126M; FY27 ~$146M (+16%, royalty acceleration + PC-NPU volume); FY28 ~$170M (+16%).
• Bear: FY26 ~$118M (+8%, China/handset drag); FY27 ~$124M (+5%); FY28 ~$130M (+5%) — royalty stays stuck, RISC-V bites licensing pricing.

Non-GAAP EPS FY2026 / FY2027 / FY2028, applying operating leverage on a ~flat-to-slowly-growing R&D base (the 40-50% profit-growth guide implies strong incremental margins):

• Base: FY26 ~$0.52 (consensus ~$0.51-0.53; = FY25 $0.42 × ~1.45 profit growth ÷ mild dilution); FY27 ~$0.72 (revenue +11% on high-incremental-margin licensing → ~40% NI growth); FY28 ~$0.92.
• Bull: FY26 ~$0.58; FY27 ~$0.95; FY28 ~$1.40 (royalty inflection is the swing — royalty is ~87% incremental margin).
• Bear: FY26 ~$0.45; FY27 ~$0.50; FY28 ~$0.55 (leverage stalls without royalty growth).

GAAP will likely stay near breakeven / small loss through FY2026-27 because SBC (~$20M+, ~$25.8M unvested over 1.5 yrs ) roughly equals or exceeds non-GAAP NI. This is a company that is "profitable" only ex-SBC for the foreseeable future.

What the price implies: at ~$40 and ~$0.52 base FY26 non-GAAP EPS, the forward P/E is ~66-77x. To justify $40 on a ~25x-out-year multiple you need roughly FY28 non-GAAP EPS ≥ ~$1.40-1.60 discounted back — i.e. the bull royalty-inflection case, not the base. Needham's own $55 PT is explicitly 70x CY27 EPS — a nosebleed multiple that only makes sense if you believe the AI-royalty flywheel compounds for years.

(If committing a tracked forecast later: "CEVA FY2026 non-GAAP diluted EPS ≥ $0.50," p≈0.62, resolves 2026-12-31 — deferred here per watchlist rules.)

Lens 12 · Bull vs Bear

Bull case. CEVA is the lowest-priced pure-play way to own the semiconductor-IP-royalty theme that has made Arm a mega-cap — at ~8x sales vs Arm's ~38x. The Connect moat (68% connectivity-IP share, 20B+ devices, first-to-standard on Wi-Fi 7 / BT 7 / UWB / 5G-Advanced / NTN) is a durable, high-margin annuity that the market under-appreciates because royalty has been temporarily capped by smartphone/memory softness. Layered on top is a real, inflecting edge-AI option: Sense & Infer went 16%→25% of revenue in one year, AI is >20% of licensing, 10 NPU deals in FY2025 including a top-PC-OEM and full-Microchip-portfolio win. The $125M signed-deal lifetime-royalty backlog is ~2.7x the current annual royalty base. As Wi-Fi 7 and NPU designs move from license to volume, royalty inflects, ~87%-margin dollars drop through a fixed R&D base, and non-GAAP EPS compounds 40%+ (management's guide). CEO Panush's InvenSense/Qualcomm sensors-and-wireless pedigree is ideally suited to the Physical-AI convergence. $222M net cash, no debt, and index-relative PSU comp keep the house in order.

Bear case (three things that could permanently impair or de-rate):

1. The AI moat may be a mirage — RISC-V + Arm commoditize the exact layer CEVA is re-rating on. CEVA is ~a top-5 NPU-IP vendor, not the leader (Arm ~24% share, transformer-native Ethos-U85). RISC-V is at 25% processor penetration and targeting edge-AI inference. CEVA's own 10-K names open-source RISC-V as a pricing-pressure and substitution threat. If NPU IP prices erode, the growth engine and the multiple both deflate.
2. China = 62% of revenue and rising, into a worsening tariff/decoupling backdrop, with UNISOC (15%) as the single point of failure. Any China export-control escalation, forced-localization push ("domestic over foreign IP"), or Taiwan-tension shock hits the majority of the revenue base at once. This is the highest-probability permanent-impairment scenario.
3. The "profitability" is a non-GAAP artifact and the price is a growth multiple. GAAP loss every year; operating cash flow negative two years running; the cash pile grew via dilution, not operations; $19.8M SBC exceeds non-GAAP NI. At ~66x forward non-GAAP EPS with ~2.5% FY2025 revenue growth, the stock has priced the bull royalty-inflection case as the base case.

Pre-mortem (it's Jan 2028 and the thesis broke — what happened?): Royalty never inflected — the record device units kept translating into flat royalty dollars because volume grew in low-ASP Bluetooth/cellular-IoT while high-ASP handset/5G kept shrinking. Meanwhile a couple of the marquee NPU "wins" stayed at the low-volume evaluation stage and never ramped to production royalty. A China export-control tightening in 2027 knocked out a chunk of the UNISOC-linked and APAC royalty stream. With growth back to single digits and still no GAAP profit, the ~8x-sales / 66x-EPS multiple compressed to ~4x sales, and the stock round-tripped toward the low-$20s — right back to where the November-2025 follow-on priced.

Are multiples too high? For the base case, yes — ~66x forward non-GAAP EPS and ~8x sales require the bull royalty-inflection to be the outcome, not the upside. The valuation is an option premium on edge-AI, underpinned by a genuinely good but low-growth connectivity annuity.

Contrarian view (what the market is refusing to see): The bulls are buying the AI/NPU story, but the thing that will actually move the P&L is boring, high-margin Wi-Fi 7 + cellular-IoT royalty on the existing connectivity base — the AI deals are still too small to matter to royalty for 2-3 years. So the stock could work for reasons different from the reason people own it (connectivity royalty inflection, not AI), OR the AI premium could deflate before AI royalty ever shows up. The mispricing is in which leg carries the thesis, and the market has the timing backwards.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case as a skeptical short:

• Structural break in how it makes money: CEVA's revenue is 58% lumpy, non-recurring licensing that must be re-won every quarter (54 deals/yr just to grow 2.5%) and 42% royalty that has declined despite record device volumes. That's a treadmill: it runs hard on licensing to stand still on total revenue, and the "recurring" leg is shrinking. Strip the AI narrative and you have a low-single-digit grower dressed as a growth stock.
• Revenue concentration + what happens if it shifts: UNISOC 15% of total; two royalty payers 39% of royalty; top-5 royalty payers 57%; China 62% of geography. Lose UNISOC (a Chinese vendor squarely in the export-control crosshairs) and you lose ~15% of revenue and a chunk of royalty overnight. This is a portfolio with a fuse in it.
• Why the moat is weaker than bulls think: the wide moat is in connectivity — a maturing, price-competitive category where in-house teams at NXP/STMicro/Qualcomm already build their own radios and only license CEVA "for some applications." The growth moat (NPU/AI) is the contested one, against Arm, Cadence, Synopsys, Imagination, VeriSilicon, SiFive, and free RISC-V. Bulls are extrapolating connectivity durability onto an AI business that doesn't have it.
• Most dangerous competitor bulls underestimate: RISC-V (the ecosystem, not a company). Free, open, momentum-heavy (25% penetration), and aimed at exactly the edge-AI-inference segment CEVA calls its growth market. It attacks CEVA on price from below while Arm attacks on performance/ecosystem from above. CEVA is squeezed in the middle.
• Worst capital-allocation / accounting choices: raising $63.3M in equity in Nov-2025 while the buyback ($7.2M) and $19.8M SBC quietly diluted holders; recognizing royalty on management estimates; cutting the credit-loss allowance $2.6M→$0.3M while receivables grew 33% into China — that combination flatters both revenue quality and the balance sheet at the same time.
• Assumptions that must hold for ~$40: (a) royalty inflects to double-digit growth within ~2 years, (b) AI/NPU deals ramp to production royalty (not just licenses), (c) China revenue is not impaired by export controls/localization, (d) the ~66x forward multiple doesn't compress. All four, simultaneously.
• If growth disappoints 20-30%: FY27 revenue ~$110M instead of ~$137M, non-GAAP EPS back toward ~$0.40-0.45, GAAP still a loss → the multiple re-rates from ~66x toward the low-20s and the stock halves to the low-$20s (the follow-on price).
• Single scenario that permanently impairs: a hard China semiconductor decoupling / export-control regime that forces Chinese SoC vendors (UNISOC et al.) onto domestic IP — permanently removing the majority of CEVA's revenue base and the fastest-growing part of it. Plausibility: moderate and rising, and it's the one risk CEVA cannot engineer around because it's geopolitical, not technical.

Lens 14 · Management Questions (ordered by information value)

1. Royalty is the whole debate: what specifically inflects royalty-dollar growth from ~flat to double-digit, and in which quarter — is it Wi-Fi 7, NPU production ramps, or cellular-IoT, and what's the ASP trajectory of each? (The single fact that most changes the thesis.)
2. Of the 10 FY2025 NPU licensing deals (incl. the PC-OEM and Microchip wins), how many are contracted to reach production royalty, at what unit volumes, and when does the first material NPU royalty dollar hit the P&L?
3. Given 62% China revenue and 15% from UNISOC, what is your concrete contingency if US export controls or Chinese forced-localization removes Chinese SoC vendors from your addressable market — how much revenue is at risk and how fast could it go?
4. On RISC-V: where are you already losing (or discounting) deals to open-source cores, and what stops NPU-IP pricing from eroding the way you concede in your own risk factors?
5. Non-GAAP profit ($10.8M) is entirely SBC add-back ($19.8M). What is the multi-year path to GAAP profitability and positive operating cash flow, and why should investors value the business on non-GAAP EPS?
6. Operating cash flow was negative two years running and the cash build came from a follow-on. What is the FY2026 operating-cash-flow plan, and was the $63.3M raise for M&A, runway, or optics?
7. Receivables grew 33% on 2.5% revenue while you cut the credit-loss allowance to $0.3M — walk through the collections risk in APAC/China and why the reserve went down as exposure went up.
8. What is the realistic conversion rate and timing of the $125M "aggregate lifetime royalty potential" from FY2025 signed deals into recognized royalty revenue?
9. Arm's Ethos-U85 is transformer-native; where does NeuPro-M win or lose against it head-to-head on transformer/generative edge workloads, and on what axis (TOPS/W, toolchain, price)?
10. How do you defend and grow the 68% connectivity-IP share as NXP/STMicro/Qualcomm keep building radios in-house and only license you "for some applications"?
11. What is the steady-state SBC-as-%-of-revenue you're managing to, and how do you stop dilution from offsetting shareholder returns?
12. Capital allocation: with $222M net cash and a stock the Street calls a Strong Buy, why the $7.2M buyback vs a $63.3M raise — what's the M&A pipeline and hurdle rate?
13. What is the through-cycle target operating model (licensing/royalty mix, gross margin, non-GAAP operating margin) at ~$150M and ~$200M revenue?
14. On the $1.0M French deferred-tax valuation allowance — is France structurally unprofitable, and does the R&D footprint need rebalancing away from high-cost/low-return geographies?
15. Israel R&D concentration: what is the business-continuity exposure if the Middle-East situation re-escalates, and how much of the core wireless roadmap is single-sited there?

---