top of page

Earnings Call: ams OSRAM (AMS) | Q3 2026 — €319M Net Profit Came From Selling Businesses and the Core Still Loses Money; AI Photonics Bets on 850nm microVCSEL With No Revenue Yet

6 days ago
8 min read

1. Executive Takeaway

First, strip the divestments out of the numbers before reading this quarter. ams OSRAM is in the middle of a run of disposals: the non-optical sensor business has closed to Infineon, and the CMOS image sensor business is about to go to Indie Semiconductor. That is why the quarter shows €503M of reported EBITDA alongside -€61M of operating cash flow. Both numbers are right; they just describe different things. The underlying business shows up on the adjusted basis, and on that basis the company is still losing money.

Second, the core portfolio is still growing in a weak market, but the growth is "traditional". On a like-for-like, constant-currency basis the semiconductor core grew about 3% year over year, led by 9% in Opto Semiconductors (OS) and roughly 23% underlying growth in Industrial & Medical (I&M); Automotive held up at about 4% on content and share gains. In other words, what keeps this company running is LEDs, automotive lamps and industrial lighting, not AI.

Third, AI photonics moved from "technology demo" to "ecosystem validation", with revenue still some way off. The keyword this quarter is not a new product but three partners: Credo (the OCP workstream and the ALC demonstrator), BizLink (fiber and connector interface) and OCP (standardization). This is an LED and sensor company taking its first step into AI scale-up interconnect, and the sequence is pluggables first, near-package later.

2. Revenue and Financials

Metric (EUR)

Q3 2026

Q2 2026

Q3 2025

YoY

Revenue

€829M

€805M

€853M

-3% (QoQ +3%)

Adj. gross margin

27.4%

—

29.7%

-2.3pp

Adj. EBITDA

€135M

€136M

€166M

-19%

Adj. EBITDA margin

16.3%

16.9%

19.5%

-320bps

Adj. net result

-€35M

-€55M

€27M

Turned negative

Adj. diluted EPS

-€0.36

-€0.56

€0.27

Turned negative

IFRS net result

€319M

-€121M

-€28M

Includes disposal gain

Revenue of €829M fell 3% year over year, with about €60M of the gap coming from divested businesses in the comparison base; like-for-like at constant currency, the semiconductor core grew about 3% and the Group core including lamps grew 4%. Headline down, underlying up: the classic report card of a company slimming down.

Adjusted EBITDA was €135M at a 16.3% margin, above the 16.0% guidance midpoint and roughly flat sequentially. The 19% year-over-year decline mainly reflects the deconsolidation of divested businesses, plus FX and raw-material (precious metals) costs. Adjusted gross margin slipped from about 29.7% a year ago to about 27.4% (derived from the company's reported adjusted gross profit and revenue), so the pressure is more visible at the gross level than at EBITDA.

The line that really needs unpacking is net income. IFRS net result of €319M and EPS of €2.97 come from the gain on selling the non-optical sensor business (other operating income of €422M in the quarter). Take that out and adjusted net result is -€35M: €99M of financial expenses in the quarter wiped out the operating profit. The market expected adjusted EPS to turn slightly positive; it came in at -€0.36, which is the direct reason the shares fell.

Segment performance:

Segment

Revenue

QoQ

YoY

Adj. EBITDA margin

Opto Semiconductors (OS)

€399M

+9%

+9%

19.2%

Light Sensors & Photonics (LSP)

€227M

-12%

-16% (about -6% ex divestment)

9.1%

Lamps & Systems (L&S)

€204M

+11%

-6%

23.1%

  • OS was the cleanest piece of the quarter: volume and factory performance lifted the margin from 17.7% back to 19.2%.

  • LSP (formerly CSA, renamed after the non-optical sensor sale) earned only a 9.1% margin because the company still provides manufacturing services to Infineon after closing, at close to zero margin; management said that at H1 pricing the LSP margin would have been about 5 points higher. This is a transition cost, not a structural problem, but it will not go away until the services agreement ends.

  • L&S rose 11% sequentially with a 23.1% margin. Q3 and Q4 are the seasonal peak for the automotive lamps aftermarket anyway, and customers shifted orders after competitor First Brands went insolvent.

End markets: semiconductor revenue was €625M, with Automotive at €235M (about +4% underlying YoY), I&M at €196M (about +23% underlying, on a recovery in horticulture and professional lighting) and Consumer at €195M (-13% YoY on a weak China Android market).

Cash: free cash flow was €462M this quarter, but €569M of that was divestment proceeds from the Infineon deal; operating cash flow was -€61M, mainly due to the continued reduction of factoring and reverse factoring, a USD 25M repayment of customer prepayments and the quarterly Senior Notes interest. Cash on hand ended at €1.39B and available liquidity at about €1.9B; net debt excluding the Kulim 2 sale-and-lease-back fell to €883M, or €1.34B including it.

3. Technology and Business Highlights

3.1 AI photonics: the first scorecard for the 850nm thin-film microVCSEL

This is the only part of the quarter directly tied to AI infrastructure, and the section STT readers should care about most.

ams OSRAM's argument is simple: scale-up needs an order of magnitude more connections than scale-out, and conventional optics, which push a few lanes to ever-higher speeds, cannot carry the cost, power and complexity once multiplied across that many links. Its answer is wide-and-slow: spread bandwidth across many parallel optical channels, each running at a lower rate, with the target of "optical reach, copper-like latency, and a path toward copper-like economics". Management also drew a clear boundary: silicon photonics and indium phosphide remain important in scale-out, while wide-and-slow targets massive parallelism inside tightly coupled AI systems.

Device data disclosed this quarter:

  • 850nm thin-film microVCSEL platform, error-free 32 Gb/s NRZ operation

  • About 0.25 pJ/bit, measured at the VCSEL device level, not the full link

  • Emitter pitch below 50 micrometers

How should these numbers be read? 0.25 pJ/bit looks great, but it covers only the emitter itself, not the driver, receiver, TIA or the SerDes interface. Microsoft's optical I/O target presented at ECOC is <1 pJ/bit for the whole link, so 0.25 at the device level means "enough budget left for everything else", not "target achieved". Meanwhile, choosing 32G NRZ per lane is a clear vote for the "slow" side, consistent with the tens-of-Gb/s-per-lane direction that Credo and Microsoft have described.

3.2 Why 850nm and not 1060nm

Management gave two reasons: 850nm uses readily available gallium arsenide (GaAs), even on 8-inch wafers, and it works with cheap standard multimode fiber. This is a cost-first choice.

But the choice is contested. Lumentum's ECOC 2026 data showed 850nm arrays failing early in high-temperature reliability testing while 1060nm lasted much longer, a route debate we broke down in 850 nm Doesn't Lose on Speed, It Loses on Junction Temperature. By betting on 850nm, ams OSRAM is betting that its thin-film structure plus lateral heat removal can solve the junction-temperature challenge; its presentation stressed that in its stack "electrons enter from the bottom, photons leave from the top, and heat is managed laterally". Reliability data is the dividing line for whether 850nm gets into AI racks. This quarter showed performance, not long-term aging data.

3.3 Vertical integration ambition: from selling components to selling a photonics platform

ams OSRAM wants to be a "turnkey OE pure-play vendor": micro-emitters, photodetectors, CMOS know-how, thin-film processing, through-silicon-via (TSV) integration, advanced wafer-level packaging and test all on one industrial platform. That differs from most VCSEL makers, which sell dies, so its competitors include not only VCSEL peers but also microLED startups. As we discussed in Slow-and-Wide Won, but Not microLED, the VCSEL-versus-microLED contest inside the slow-and-wide camp has only just begun, and ams OSRAM's disclosed data puts it on the VCSEL side.

3.4 Core business: panoramic HUDs and a vivo flagship

  • Panoramic head-up displays (HUD): first platform design wins this quarter. Citing Omdia, panoramic HUD volumes are expected to grow from under 1 million units today to more than 8 million in 2031 and beyond. For ams OSRAM, this extends its automotive LED leadership into higher-value applications and raises content per vehicle.

  • vivo X500 Pro Max: one phone uses three ams OSRAM sensors: a behind-OLED spectral color sensor, a multi-zone direct Time-of-Flight (dToF) sensor and a multi-channel spectral sensor.

  • EVIYOS smart pixel headlamps added more than €50M of design wins this quarter.

Semiconductor design wins were about €1.3B this quarter and about €3.8B for the first nine months, against a full-year target of about €5B.

4. Management Outlook

Item

Q4 2026 guidance

Revenue

€750–850M (midpoint €800M, about -3% QoQ)

Adj. EBITDA margin

16.0% ± 1.5pp

FX assumption

EUR/USD 1.12

Divestment impact

About €75M vs Q4 2025

The €800M Q4 midpoint compares with €874M a year ago, about 8% lower on a reported basis, but roughly €75M of that is the divestment base effect, so underlying revenue is broadly flat. Semiconductors should see the typical seasonal decline, while automotive lamps hit their usual seasonal peak.

Reading the language:

  • The guidance uses "expect", a neutral tone; the margin range stays at ±1.5pp, the same as Q3 and not narrowed, which means management's visibility into Q4 has not improved.

  • On 2027 free cash flow, the phrase is "continue to see positive", which is a target, not a commitment. Note that most of this quarter's €462M free cash flow was disposal proceeds and operating cash flow is still negative; turning positive in 2027 depends on Simplify savings and the end of the manufacturing services agreement.

  • On AI photonics, the CEO used "early commercialization steps" and "still a lot of hard work ahead". This was the most honest line of the quarter: the technology has been demonstrated and the ecosystem is forming, but there is no revenue timeline.

  • Simplify cost program: more than €20M of run-rate savings delivered so far, against a €200M target by fiscal 2028, about one-tenth of the way.

5. Supply Chain and Customer Signals

  • Credo: ams OSRAM joined the Credo-led OCP workstream for interoperable short-reach micro-emitter optical links, and its micro-emitters featured in the Active Light Cable (ALC) demonstrator Credo showed at OCP. Credo's OCP message was Wider, Slower, Better: Optics That Can Beat Copper, and its term "micro-emitter" deliberately covers both VCSEL and microLED. ams OSRAM is one of the few light-source suppliers publicly on this platform so far. For Credo, this also fits the optical push it described as a second growth engine in its Credo Q1 FY2027 earnings call.

  • BizLink: the two companies are working on the fiber and connector interface. Management named high-core-count alignment and a scalable connector architecture as central to commercialization, which shows that one of slow-and-wide's real challenges sits in passive components, not just the light source.

  • Infineon: post-closing manufacturing services are provided at close to zero margin, directly depressing the LSP margin; when this arrangement ends will determine when the LSP margin returns to the mid-teens.

  • Indie Semiconductor: the sale of the CMOS image sensor business is expected to close soon, which is why the company extended the tender offer for its 2027 convertible bond.

  • vivo and the Android camp: flagship design wins are in hand, but overall weakness in China Android is the main reason Consumer revenue fell 13% YoY.

  • Automotive lamps aftermarket: customers shifted orders after First Brands' insolvency, a short-term share windfall whose durability remains to be seen.

To see how ams OSRAM positions its AI photonics in an industry setting, see our coverage of FIT TECH DAY 2026: Light at Scale, where ams OSRAM shared the stage with Lumentum and NTT to discuss AI data center interconnect.

6. Conclusion

ams OSRAM's quarter needs to be read on two levels. The first is an LED and sensor company slimming down: divestments inflated reported profit and cut debt, but the company is still loss-making on an adjusted basis, operating cash flow is still negative, and Simplify savings have barely started. The second is an AI option that has not yet paid out: the 850nm thin-film microVCSEL delivered device-level data, and Credo and BizLink gave it a ticket into the scale-up ecosystem, but between accumulating field hours in active optical cables and truly moving next to the compute package lie at least two hurdles: reliability validation and volume manufacturing.

Three metrics to track next quarter:

  1. Whether the LSP adjusted EBITDA margin gets back above 12%: this shows when the Infineon services wind down, and it is a precondition for positive free cash flow in 2027.

  2. Reliability data for the 850nm microVCSEL: with the 1060nm camp emphasizing high-temperature lifetime, ams OSRAM needs to publish long-term aging results to convince system makers.

  3. Whether full-year design wins reach about €5B: Q4 needs roughly another €1.2B, a leading indicator for core revenue in 2027–2028.

In one line: ams OSRAM has its ticket into AI photonics, but this quarter the bills were still paid by automotive lamps and LEDs.

This article is for technology and industry trend analysis only and does not constitute investment advice.

Comments

Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page