The 30 second version
- Ceva licenses chip designs: wireless radios, sensor processors and NeuPro neural processors (NPUs) that run AI inside other companies’ chips.
- The NPU business sits at Liftoff: chip and platform companies are starting to pay for a licensed AI engine rather than build their own.
- Evidence: a major US software and AI platform company licensed NeuPro M in July, and licensing revenue hit a three year high of $18.2M in Q2.
- The test: Q3 results before the open on Monday November 9, which show whether licensing holds near $17M without another platform sized deal.
- It breaks if Q3 licensing falls below $14M with no new NPU license, or the full year outlook drops below $118M.
Ceva earns roughly two cents in royalty on a typical chip carrying its designs, and more than two billion of those chips ship every year. In July a major US software and AI platform company, one that controls both its own hardware and its own operating system, picked Ceva’s NeuPro M processor as the AI engine for a custom chip. It chose to license rather than build, and Ceva’s chief executive called it one of the most strategically significant AI licenses in the company’s history.
That deal sits inside second quarter licensing revenue, which reached its highest level in three years. On Monday morning, before the market opens, Ceva reports the third quarter. That is the first quarter that has to stand without the platform deal, which makes it the cleanest read yet on whether AI licensing is repeating or whether July was a one off.
What just happened
The setup runs through three dated events. On July 6 Ceva announced the platform license, without naming the customer, and said its engineers had tuned neural network optimizations for that customer’s specific workloads. On August 10 it reported Q2: revenue of $29.0M, up 13% from a year earlier, with licensing and related revenue of $18.2M, up 21%. It signed ten license agreements in the quarter, two with first time customers and two directly with device makers rather than chip companies.
Then came the quieter part. On October 14 the company set its Q3 report for November 9, before the Nasdaq open, with a call at 8:30 a.m. Eastern. On October 28 it announced a next generation NeuPro NPU portfolio and said it was expanding into AI infrastructure, a push beyond the device. We have read the headline of that one, not the detail. Early November brought connectivity news, including NSING passing 250 million Bluetooth chips shipped on Ceva technology, and a collaboration with Mobilint on its REGULUS edge AI chip.
None of that is in a reported number yet. Monday is where the July license either shows up as a pattern or stays a single headline.
The business
Ceva sells designs, never chips. A customer pays an upfront license fee to use a block of Ceva’s design, say a Bluetooth radio, a 5G modem component or an NPU, inside its own chip. Years later, when that chip ships in volume, Ceva collects a royalty on each unit. In Q2 licensing and related revenue made up 63% of the total and royalties the other 37%.
The economics are lopsided in a useful way. Cost of revenue was only $3.6M on $29.0M of sales, a gross margin of 87%. The real cost is engineering: research and development ran at $19.3M in the quarter. Once a design is built, the next license costs little to deliver and the next royalty costs almost nothing, so every extra dollar of revenue falls heavily to profit. On our arithmetic, $43M a year of royalties spread across more than two billion devices works out to about two cents a chip. An NPU inside a computing device should earn more than a radio in a fitness band, though Ceva does not disclose its rates.
Two filing details matter. Unbilled receivables fell from $29.9M at the end of 2025 to $24.4M at June 30, which tells us license revenue booked in earlier periods is being invoiced and turned into cash. And the company booked $1.8M of income tax in Q2 on a pretax loss of $1.1M, which we read as foreign withholding on royalties paid from abroad.
The competition is Arm, Cadence and Synopsys, all of which license their own NPU designs, plus the in house teams at large chip makers. Ceva wins where a buyer wants an NPU, wireless and sensing from one vendor with the engineers willing to customise.
Where it sits on our curve
Our Disruption Map puts edge AI and embedded inference at Liftoff, which matters now because device makers are choosing the AI engines for their next chips. The product we care about is NeuPro, and we place it at Liftoff too. The binding constraint is the cost of an NPU team. A company that wants AI running on the device needs an engine that fits a strict power, area and heat budget and comes with a working software stack. Building that in house takes years and a dedicated team, by our estimate, for every product generation.
The buyer’s threshold is simple: the license has to be cheaper and faster than doing it yourself, and good enough that the platform’s own software runs well on it. The hardest buyer to win owns both an operating system and its hardware, because it can afford to build. That buyer signed and paid. The furthest step for NPU specifically is a paid deployment. Ceva also credits the ramp of automotive AI programs for part of its royalty recovery, but it does not split out NPU royalties, so we do not count that as proof.
The strongest alternative explanation is less exciting. Licensing may simply be recovering from a weak 2024 and 2025, padded by one unusually large platform deal. The test that separates the two is plain: licensing at or above $16M a quarter through Q3 and Q4 without another platform sized deal, and NPU licenses from customers unrelated to the first.
There is also the bill to follow. The customer pays for a design and a share of every chip. If base models get much better and much smaller, more of them run locally and an efficient NPU becomes more useful. The real risk is the platform customer learning enough from Ceva to build the next generation itself. (Expect every call to bring guesses about who that customer is. We will not join in.)
The numbers
The turn shows in licensing first. Licensing and related revenue was $15.0M in both Q1 and Q2 of 2025, then $17.8M in Q1 2026 and $18.2M in Q2. Trailing twelve month licensing reached $69.6M, up 13%. Royalties are the weaker line: $9.2M in Q1 2025, $10.7M in Q2 2025, $9.2M again in Q1 2026 and $10.8M in Q2. Flat on a year, but up 17% from the prior quarter.
Total revenue climbed from $24.2M and $25.7M in the first half of 2025 to $27.0M and $29.0M this year, so first half sales rose 12% to $56.1M. Costs barely moved. Operating expenses excluding stock pay and acquisition items rose 3% to $22.3M, which lifted the adjusted operating margin to 11% from 3%. On a GAAP basis the company still lost $2.1M at the operating line, mostly because of $5.2M of stock compensation.
The balance sheet carries no debt and $220.7M of cash, deposits and marketable securities, against $222.0M at the end of 2025. Cash and investments slipped only $1.3M in six months. The weak spot is the share count: basic weighted shares rose from 23.9M to 28.0M in a year, an increase of 17%. That is a big step for a company with no debt, and we did not open the filing that explains it.
Ceva gives its outlook on the call, and we could not confirm its guidance figures from a primary source today. The screening desk carries a street estimate of about $125M for 2026, up 14%, and $141.6M for 2027. We could not verify either. Our own check is mechanical: $125M needs about $68.9M in the second half, against $56.1M in the first.
What the price already requires
At around $24.56 a share, a price we could not confirm this morning, and 28.0M basic shares, Ceva is worth about $688M. Take away the cash and the business itself is valued at roughly $467M, about 4.2 times the first half revenue rate. Turn that around. At 15 times adjusted operating profit, $467M needs about $31M a year of it. With 88% gross margin and operating costs held near $92M, a small rise from today’s $89M rate, that takes revenue of about $140M, which we think is reachable in 2027.
Now the cases. Downside: licensing slips back to $15M a quarter, royalties stay flat, 2027 revenue is $118M and the market pays three times sales, which gives about $20.50 a share. Central: revenue of $141M in 2027 at four times sales gives about $28, or about $26 on the 29.8M diluted share count. Upside: NPU royalties start to arrive, revenue reaches $155M and the multiple is five, which gives about $35.50. Our central case covers today’s price on either share count.
The gap is in the royalty line. The price treats Ceva as a wireless royalty company growing in the low teens, with licensing that might fade. What no published number holds is the royalty stream from an NPU inside a platform company’s own computing devices, a chip that will not ship for years but was paid for this summer. The price needs Ceva to keep what it has. We think the business is adding a new stream on top of it.
What breaks it
Written before Monday: the thesis breaks if Q3 licensing and related revenue comes in below $14M with no new NPU or AI license disclosed, or if the full year outlook falls below $118M, which would mean growth under 8%. It also breaks if the Q4 report in February 2027 arrives without a further AI license from a new customer. A deadline that slides is a failure, and we will treat it as one.
The next most likely failure is slower. Royalties could stay stuck near $10M a quarter while smartphone and consumer volumes soften, leaving licensing to carry a business the market prices on royalties. Financing is not the issue: Ceva is roughly cash neutral with $220.7M in the bank and no debt, so it reaches every test on our calendar without raising money. The danger is another equity raise used for acquisitions, which would hand the upside to new holders.
What we are watching
- November 9, 2026, Q3 results: stronger if revenue reaches $31M, licensing $17M, royalties $11M and at least one new NPU license is named. Broken if licensing is below $14M with no NPU license, or the full year view is under $118M.
- Same call: the 2026 outlook. A view consistent with $125M or more strengthens the case; anything implying under $118M breaks it.
- February 2027, Q4 report: at least one more AI or NPU license from a customer unrelated to the July platform deal, ideally the first from the AI infrastructure push announced October 28. None disclosed breaks the thesis.
- May 2027, Q1 report: royalties above $10M in a seasonally soft quarter, against $9.2M in Q1 2026. A fall below $9M would say the royalty recovery has stalled.
Not financial advice.
References
- Ceva, Inc. Ceva Wins Landmark AI Licensing Deal with Major U.S. Software and AI Platform Company. Published July 6, 2026. Accessed November 5, 2026. https://www.ceva-ip.com/press/ceva-wins-landmark-ai-licensing-deal-with-major-u-s-software-and-ai-platform-company/
- Ceva, Inc. Form 8-K, Exhibit 99.1: Ceva, Inc. Announces Second Quarter 2026 Financial Results, for the quarterly period ended June 30, 2026. US Securities and Exchange Commission. Published August 10, 2026. Accessed November 5, 2026. https://www.sec.gov/Archives/edgar/data/0001173489/000143774926026648/ex_1001319.htm
- Ceva, Inc. Ceva, Inc. Schedules Third Quarter 2026 Earnings Release and Conference Call. Published October 14, 2026. Accessed November 5, 2026. https://www.ceva-ip.com/press/ceva-inc-schedules-third-quarter-2026-earnings-release-and-conference-call/
- Ceva, Inc. Press releases index, including Ceva Unveils Next Generation NeuPro NPU Portfolio and Announces Expansion into AI Infrastructure (October 28, 2026), Ceva and Mobilint Advance Physical AI with REGULUS Edge AI SoC (November 2, 2026) and NSING Technologies Surpasses 250 Million Bluetooth Chips Shipped Powered by Ceva Technology (November 5, 2026). Accessed November 5, 2026. https://www.ceva-ip.com/press/