On 31 July, the Financial Times reported that NXP Semiconductors was in talks to acquire Ambarella in a deal worth more than $3bn. Ambarella shares rose 16.1% that day on 9.2m shares, against typical volume of one to two million. Strip out that one session and the stock's entire last month amounts to about 2.6%.

Ambarella designs low-power system-on-chip silicon that squeezes video compression, image processing and computer vision onto a single part, sold to the makers of dashcams, security cameras, driver-assistance modules and robots. It is, on the user's own reading of the field, effectively the only Western listed pure-play left: Mobileye is roughly 88% owned by Intel, Hailo is private, and the Chinese alternatives are largely closed to Western original-equipment makers on export-control grounds. That scarcity is the takeover logic. Automotive is expected to supply about half of NXP's projected $15.1bn of 2026 revenue, and the deal would arrive in a consolidating industry that has already seen Texas Instruments buy Silicon Labs for $7.5bn this year.

The business behind the bid

What the report did not change is the trajectory. Ambarella's year-on-year revenue growth has decelerated for four consecutive quarters — 49.9%, then 31.2%, then 20.1%, then 16.9% in the quarter ended 30 April, on revenue of $100.4m. Gross margin has held up through it, at 59.1% in that quarter against 57.6% a year before, and the GAAP operating margin improved to -18.1% from -30.1%. The company is still loss-making. Consensus has full-year revenue growth of 13.2%, to $441m.

The shares carry 15.2x trailing gross profit, up from roughly 13.1x six months ago, and 104x forward earnings. Analysts pointed out that near $90 the company was worth about $4.0bn, or 9.6x trailing revenue — more than twice NXP's own sales multiple, and that a customary 25% premium to the pre-report close of $74.09 works out to $92.61, barely above where the stock traded on the news. Talks could still collapse. Ambarella has not reported since 28 May; it guided the current quarter to $105m-$111m.

The licensor that beat and fell

CEVA sells no chips. It licenses digital signal processor cores, its NeuPro neural-network accelerators and wireless connectivity designs to chipmakers, taking an upfront fee and then a royalty on every unit its customers ship. Four hundred employees produce an 87.4% gross margin — which is why the near-identical price-to-sales ratios of the two companies, 8.0x and 8.9x, mislead, and why gross profit is the fairer yardstick.

CEVA's growth has accelerated for four straight quarters, from 4.3% to 13.1%, and on 10 August it reported quarterly revenue of $29.0m with licensing up 21% to $18.2m — its strongest licensing quarter in three years. Non-GAAP operating margin reached 11%, from 3%. Management raised full-year growth guidance to 13-15% from 12%. The shares fell about 10% anyway, and are down 31.3% from their 15 June high.

There is one genuine problem in those numbers. Customers shipped 567m royalty-bearing units, up 16%, yet royalty dollars grew 1%, to $10.8m. Bluetooth volumes fell 16%, to 295m units, and low-priced consumer connectivity carried the mix — so the average royalty per unit is eroding even as units climb. Management's answer is to sell whole subsystems rather than components: ten licensing agreements last quarter, including one with what it described as a leading global AI and computing platform company that picked its NeuPro-M accelerator for custom silicon, with 18 to 24 months to production.

On price-to-gross-profit, CEVA has cheapened to 9.2x trailing from roughly 10.8x three months ago, as trailing gross profit rose 3.3% and the stock fell 12.4%. Ambarella trades at a 66% premium on the same measure. CEVA holds $221m of cash against a $927m market value.

The third name

Much of CEVA's drop was not about CEVA. In the last week of July, chip stocks lost more than $1 trillion of market value on reports that SK Hynix was slowing memory capacity expansion; CEVA fell 20.9% in that window and Ambarella, cushioned by its own story, was flat.

indie Semiconductor, which designs automotive chips for driver assistance, in-cabin systems and connectivity, is the third edge name often grouped with them. It grew revenue 24% to $64.0m last quarter, guided the current one to roughly 30% growth, and reported gross margin down to 36.1% from 40.6% with a $37.1m net loss. Its 18.8% jump on 14 August came with no discoverable company-specific news that day — CEVA rose 8.6% alongside it while Ambarella was unchanged and Broadcom fell 6.8%. It looks like a squeeze in small edge-AI names, and indie's own vision chip competes with Ambarella rather than validating either.

The setup

Where it stands — Ambarella's month is a takeover report; CEVA's is a sector selloff layered on royalty-mix erosion. Would confirm — CEVA's royalty dollars growing in line with unit shipments in the September quarter. Would invalidate — Ambarella reporting revenue above its $105m-$111m guide with growth reaccelerating past 20%. Watch next — Ambarella's fiscal second-quarter results, due late August; CEVA guided the September quarter to $30.5m-$34.5m. Valuation — Ambarella 15.2x trailing and 14.0x forward gross profit, versus CEVA's 9.2x and 8.6x.