ams OSRAM isn't an AR glasses play - it's a slow-and-wide microLED chokepoint at $1.6B

ams OSRAM just reported Q2/26 at the high end of guidance: €805M revenue, 16.9% adjusted EBITDA margin, +13% YoY growth in the semiconductor core.

The headline everyone will quote is the microLED update for AR smart glasses. "Advancing toward mass-production readiness," management says.

That's not the story. The story is buried two layers deeper.

So here's what most people miss:

ams OSRAM isn't just a microLED display supplier. They're one of the very few companies on earth that has actually industrialized microLED arrays at scale.

The product is called EVIYOS - 25,600 individually addressable microLED emitters, each half the width of a human hair, paired with a CMOS driver chip in a compact package. It won the German Future Prize. It's already in production vehicles for automotive matrix lighting.

That manufacturing capability now maps to two completely separate AI infrastructure demand curves.

Demand curve #1: "Slow and wide" AI optical interconnects

Microsoft made its position public at OFC 2025. Their Corporate VP of Silicon argued that chasing 400G+ per lane is a trap. The sweet spot is 200G per lane, scaled wider with hundreds of parallel channels.

Microsoft's research arm published the MOSAIC paper proposing exactly this architecture - using microLEDs.

Then in March 2026, ams OSRAM showed up at OFC 2026 and demonstrated the exact technology Microsoft's architecture requires:

• Cut-off frequency >1 GHz
• Energy consumption below 2 pJ/bit
• Per-lane data rates of ≥3.0 Gbit/s
• Bit-error-rate below 10⁻¹⁵ over a 10-meter link
• Singulated microLEDs diced from the wafer, mounted on substrate, each coupling to its own fiber-optic channel

That spec sheet is not coincidence. It's a direct answer to the slow-and-wide architecture.

In Q1/26, ams OSRAM signed a "development agreement with a leading AI data center infrastructure partner" to commercialize Digital Photonics for optical interconnects. They didn't name the customer.

The mapping is obvious. Microsoft is the only hyperscaler publicly championing slow-and-wide with microLEDs. Google is going fast-and-narrow at 400G. AWS is playing with hollow core fiber. Meta is doing co-packaged optics at scale.

My guess: that development agreement is Microsoft.

Demand curve #2: AR smart glasses (the headline)

This is what the press release focuses on, and it's real - just not the whole picture.

ams OSRAM estimates €50–100 of content per AR smart glasses device, depending on volume and lifecycle. MicroLED-array based RGB light engines are hitting "key development milestones" toward mass production.

Meta is targeting microLED-based AR glasses for a late-2027 launch, replacing the LCoS displays in their current Ray-Ban Display. Sources indicate Meta may be co-developing microLED technology with ams OSRAM, and a recent job posting has Meta hiring an advanced manufacturing engineer specifically to drive the ramp-up of RGB microLED chips at the company's key external partner.

Apple has been working on microLED for watches since before 2022, with ams OSRAM named as a supply chain partner. That program has slipped repeatedly, but the qualification pipeline is real.

The AR glasses demand curve is longer-dated - 2027+ for volume. But it's not vaporware. It's a qualification timeline.

The chokepoint test

So the question becomes: is microLED manufacturing actually a chokepoint, or just another crowded layer?

Here's why it leans chokepoint:

• EVIYOS is the world's first high-volume automotive micro-emitter array. That means ams OSRAM has solved the yield, epitaxy stack, and CMOS integration problems that have kept every other microLED company in prototype mode.
• Samsung Display, JBD, and a handful of others have demo'd microLED panels, but most are stuck on transfer processes, yield issues, and red-emitter efficiency. ams OSRAM's EVIYOS uses a different architecture entirely - monolithic arrays that avoid the transfer bottleneck.
• For optical interconnects, the fabrication process is the same base epitaxy stack, just singulated instead of monolithic. That's an enormous advantage: the capex is already sunk.
• The slow-and-wide architecture needs massively parallel microLED emitters. Nobody else has demonstrated this at scale.

The monopoly question: ams OSRAM likely isn't the ONLY microLED producer on earth. But for the specific applications that matter here - high-reliability automotive-grade microLED arrays that can be adapted to optical interconnects and AR light engines - they're the only one with proven volume production.

The overhang

The financials are messy. Q2 net result was -€121M. The non-adjusted EBITDA margin sits at 14.2% vs. the adjusted 16.9%. There's a mountain of transformation costs, share-based compensation, and M&A-related charges.

The debt situation requires attention:

• €1B of new 7.25% senior notes, placed in May. Saves ~€40M in annual interest.
• A planned €120–150M tender offer to buy back 2027 convertible bonds and 2029 senior notes, triggered by the Infineon sale closing July 1st.
• Management sees a path to positive free cash flow in FY27, but that's excluding divestment proceeds and net interest.

The divestments - non-optical sensors to Infineon, CMOS image sensors to Indie Semiconductors, specialty lamps to Ushio - are sharpening the company toward Digital Photonics. Effective July 1st, ams OSRAM established dedicated Digital Photonics business lines.

That's a deliberate strategic pivot, not a random portfolio shuffle.

The valuation question

$1.6B market cap. Revenue of €805M in Q2, so roughly €3.2B annualized run-rate. P/S around 0.5x.

The stock is losing money on a GAAP basis. The forward P/E is negative. Traditional metrics say "distressed turnaround."

But the traditional metrics are measuring the legacy business being sold off - non-optical sensors, CMOS image sensors, specialty lamps. The Digital Photonics portfolio carries €2.5B in design wins for H1/26 alone, with €1.6B of those in Q2.

The question isn't whether the current earnings power justifies $1.6B. The question is whether the microLED exposure to slow-and-wide interconnects and AR displays is priced in.

At $1.6B, with a development agreement already signed with a leading data center player and EVIYOS manufacturing capacity that nobody else has proven at volume - the microLED story feels underpriced.

Risks

• The AI photonics customer identity is unconfirmed. Microsoft is the strongest map given MOSAIC alignment, but it's not public.
• Slow-and-wide is one hyperscaler's bet against the industry consensus of fast-and-narrow. If Microsoft doesn't win the architecture war, the interconnect demand curve shrinks dramatically.
• AR glasses mass production for microLED keeps slipping. Meta's timeline is 2027+. Apple's has been "next year" since 2022.
• Debt load is real. €1B in senior notes at 7.25% is expensive financing. The path to positive FCF in FY27 is management's projection, not a fact.
• Design wins don't equal revenue. €2.5B in pipeline is meaningful, but the conversion rate and timeline matter more than the headline.

TLDR

Everyone reading the Q2 press release is thinking "AR smart glasses."

The actual setup is: ams OSRAM at $1.6B controls one of the only industrialized microLED fabrication lines on earth. That same EVIYOS process feeds two demand curves - AR displays (2027+) and slow-and-wide AI optical interconnects (development agreement signed, customer likely Microsoft given MOSAIC architecture alignment).

The microLED manufacturing capability IS the chokepoint. Most "microLED companies" are still in prototype mode. ams OSRAM has 25,600 individually addressable emitters running in production vehicles.

Debt is the overhang. Divestments are the cleanup. Digital Photonics is the pivot.

I personally think the slow-and-wide interconnect angle is the underappreciated catalyst. If Microsoft standardizes around microLED-based parallel channels, ams OSRAM's positioning shifts from "European semiconductor turnaround" to "hyperscaler infrastructure supplier." That's a completely different multiple.

Speculative mapping on the customer. Structural thesis on the manufacturing chokepoint. Two things are different.