The Optical Inflection Nobody’s Pricing In

The AI infrastructure buildout is the most obvious investment theme of our generation. Everyone knows it. The problem is the obvious plays already ran.

The optical interconnect names that Wall Street finally understood are up hundreds of percent. The GPU names are priced for perfection. The power generation plays have seen massive inflows. Every obvious door has been walked through.

But there’s a category of stock that gets mispriced at every major inflection: the company that’s been quietly doing the hard engineering work for decades, sitting one product launch away from being recognized as part of the new regime, still wearing its old valuation from the old world.

This is one of those.

It’s an optical networking company. It’s been building the physical infrastructure of the internet for forty years. It has the engineering depth, the customer relationships, and now the product to compete directly in the fastest-growing segment of AI data center infrastructure.

It trades at roughly 1x 2026 revenue at 43% gross margins.

Its closest peers in this space trade at 9x and 25x.

That gap doesn’t close overnight. But when the market makes the connection, and it will, the re-rating is violent.

The full thesis is below.

---

A New Player Has Joined The Party

The AI data center buildout is creating some of the fastest wealth generation in market history. The optical interconnect layer, the wiring that holds GPU clusters together, is at the center of it.

$LITE is up 1,269% in a year. Market cap: $73 billion, trading at ~25x forward P/S on ~$2.9B in FY2026 revenue, growing ~90% YoY. $COHR is up nearly 300%. Market cap: $63 billion, trading at ~9x forward P/S on a ~$7B run rate, growing ~21% YoY. Wall Street understands the thesis: AI clusters need massive bandwidth at ultra-low power, and optical interconnect is the only solution at scale.

What the market missed is that a third player just entered the room.

$ADTN, ADTRAN Holdings, is a 40-year broadband optical veteran that just launched a product with specs that outclass first-generation LPO modules by 6–10x on power efficiency. Market cap: ~$1.34 billion. That’s roughly 1/60th of each of its two largest competitors. Same space. Differentiated product. Decades of directly applicable engineering.

---

Goldman Sachs Just Told You The Market Size, and It’s Enormous

In April 2026, Goldman Sachs published a landmark optical networking report: “The Next Mega Trend in AI Infrastructure.” The headline number is a $154 billion TAM by 2028, a 9x expansion from $15 billion today. That’s not a forecast from a market research firm. That’s Goldman’s Global Tech team, with supply chain checks across the Nvidia, Google, Amazon, Meta, and Huawei ecosystems.

The core insight from the report: the real TAM unlock isn’t scale-out (connecting racks together), it’s scale-up (connecting GPUs within and between racks). As clusters move from GB300 NVL72 to Rubin Ultra NVL576, the dollar content of optical networking per compute unit jumps from $315K to $9.4 million, a 29x increase. Goldman’s pluggable optical module TAM in scale-out alone grows from $8.3B (2026) to $20.2B (2027) to $32.4B (2028).

Critically, Goldman explicitly confirms that pluggable optical modules will co-exist with CPO through this entire cycle. CPO is for extreme short-distance, high-bandwidth applications where pluggables physically can’t go. LPO is the high-growth, commercially mature solution for the intra-rack and inter-rack scale-out that’s happening right now. LiteWave800 is squarely in the middle of the fastest-growing segment Goldman mapped.

One more data point that validates the ADTN thesis directly: Goldman cites VCSEL as the light source technology with “high energy efficiency and technology readiness” for scale-up applications, and VCSEL single-mode architecture is exactly what powers LiteWave800. ADTN didn’t stumble into the right technology. They built to spec.

---

Why Now

Data centers are hitting a wall. Every GPU added to a cluster compounds the power and thermal problem. The industry is migrating fast from DSP-based transceivers to Linear Pluggable Optics (LPO), an architecture that eliminates the digital signal processor layer entirely, cutting latency, power draw, and cost per bit simultaneously.

Goldman’s report lays the problem out plainly: “AI data centers are hitting severe thermal and power constraints per rack.” Traditional 800G pluggables require power-hungry DSPs inside the module consuming significant energy. LPO removes the DSP entirely, offloading that work to the host ASIC. The industry is adopting LPO now at 800G/1.6T for AI back-end networks, and Goldman’s GPU attach rate data makes the volume trajectory clear: Nvidia’s upcoming VR200 racks come with an optical module attach ratio of 1:4–6, doubling from GB300’s 1:2–3. More modules per GPU, higher speeds, faster upgrade cycles.

ADTN has been solving the exact engineering problems this market demands, power density, heat dissipation, signal efficiency across dense fiber environments, for four decades. A big part of that depth came through their 2022 merger with ADVA Optical Networking, a Munich-based optical transport specialist that was the global market share leader in data center interconnect per Ovum. ADVA brought metro wavelength division multiplexing, DCI, and network synchronization expertise that gave ADTRAN a full optical transport stack overnight. The carrier network and the AI data center hit the same physics wall. ADTN already has the playbook.

---

LiteWave800: A New Power Class

Launched March 10, 2026. Demoed at OFC in LA. The specs are category changing.

Spec LiteWave800 Throughput 800Gbit/s DR8 Energy efficiency 1 pJ/bit Power draw 0.8W vs. DSP transceivers 12–18x lower vs. gen-1 LPO 6–10x lower Reach 500m intra-DC Form factor OSFP, LPO MSA compliant

Single-mode VCSEL architecture, the same light source technology Goldman Sachs identifies as the superior choice for scale-up power efficiency, combined with ADTN’s proprietary low-power electronics, fully vertically integrated from chip to system. Most LPO competitors are assemblers. ADTN designs at the component layer where the efficiency gains are actually won. Plug-and-play with existing host systems. CEO Stanton confirmed on the Q1 call: not dilutive to gross margins. Scale production targets Q1 2027, right as Goldman projects the pluggable module TAM to hit $20B+ in scale-out alone.

---

The Market Is Already Responding

This isn’t a product in search of customers. The feedback loop has already started.

Breaking this morning (May 27): euNetworks, a pan-European fiber carrier serving 600+ data center customers across 17 cities, launched a new managed service called “Quantum Shield”, built entirely on ADTN’s ConnectGuard encryption and ALM optical stack. This is commercial validation of a different kind. ProRail and SUNET were end-user deployments. euNetworks is wholesaling ADTN’s technology as a branded, recurring-revenue product to its entire data center customer base. Every enterprise that buys Quantum Shield is pull-through revenue for ADTN. That’s a fundamentally different revenue shape, recurring and compounding, not one-time.

The choice wasn’t accidental. ADTN’s S-Flex is the only 400G Layer 1 BSI-approved product in the public record, making it the mandatory solution for DORA and NIS2-regulated customers across Europe. euNetworks didn’t pick ADTN from a shortlist. They picked ADTN because no one else qualifies. That certification moat, which looked like a procurement advantage on paper, just produced its first visible commercial outcome. And notably, ADTN CTO Christoph Glingener appeared on the euNetworks launch announcement directly, not a regional VP. That’s a signal: ADTN is treating this as a reference architecture they intend to replicate with other carriers. Watch for Colt, Zayo, or BT Wholesale making similar moves in the next two quarters.

On the Q1 2026 call, Stanton was asked directly about hyperscaler engagement. His answer was unambiguous:

“Hyperscalers actually did really well in the fourth quarter. They were a real positive in the quarter, and we would expect that to continue on through this year. We’ve got a fairly good backlog with some of our hyperscaler customers right now that’s building.” — Tom Stanton, Q1 2026

That backlog comment is the tell. You don’t have building backlog with hyperscalers without active engagement and commitment. These aren’t tire-kickers.

Multiple sources from the Q1 call confirm that LiteWave800 received positive market feedback at OFC in LA, with early customer reactions described as encouraging across the board. The prepared remarks went further, noting “growing strength among our cloud and hyperscaler customers” and explicitly stating management expects optical networking revenue “to build throughout the year.”

This wasn’t a one-quarter blip either. In Q4 2025, Stanton attributed optical networking’s 33% YoY growth directly to cloud providers and large enterprises, stating that “cloud providers expanding data center capacity and large enterprises upgrading their optical networks” were the core demand driver. Cloud and enterprise customers combined for 25% of total revenue in Q4 2025, up from 21% for the full year, a mix shift accelerating in the right direction.

The supply chain question from the Q1 call is also worth flagging. An analyst asked whether laser datacom chip supply was an issue. Stanton’s response:

“We do have some partners that we’re working with on this. They do know what the supply needs are right now. Depending on how aggressive that launch is, we don’t see any issues in being able to supply it as we launch it.” — Tom Stanton, Q1 2026

Sourcing is already locked in. Partners are already aligned to supply needs. This commercial launch has the supply chain infrastructure and vertical integration behind it.

---

One Product Is Just the Entry Point

LiteWave800 is the first member of a family.

On the Q1 call, an analyst asked whether the 90% power reduction engineering transfers across the broader catalog. Stanton’s answer was deliberate:

“I did call it a family. And I consider Quattro to be part of that same family, which is in our multi-mux family, which is very, very power savings as well. You’re gonna see first product is QSFP. We do have other products that are more integrated, that will be coming out over time. Different members of the family and similar application sets where this technology will actually play itself out.”

Three product tiers telegraphed: LiteWave800 (800G OSFP), a QSFP variant (next form factor, broader host compatibility), and more integrated solutions optimized further for density and power. The MicroMux Quattro is already in-market as a member of this family. ADTN’s full optical stack maps across every layer Goldman Sachs identified:

DC Layer ADTN Product GS TAM Layer Intra-rack scale-up LiteWave800 Scale-up ($106B by 2028) Inter-rack scale-out MicroMux Quattro Scale-out ($48B by 2028) Inter-DC / cloud 100/400/800G coherent pluggables + FSP 3000 Scale-across DC synchronization Precision timing solutions Infrastructure Wholesale fiber MOFN Hyperscaler capacity

When the analyst pushed on whether opportunities extend beyond data centers, Stanton closed it:

“The demand within inside the data center is worth focusing on. It is very large.”

---

The Valuation Doesn’t Add Up

• Q1 2026 revenue: $286.1M, +15.5% YoY

• Q2 2026 guidance midpoint: ~$293M

• 2026E run-rate revenue: ~$1.15–1.17B

• Market cap: ~$1.34B

• Forward P/S: ~1.15×

Mid-teens revenue growth at 1.15x sales. Non-GAAP operating margin expanded 300bps YoY to 6.9% in Q1. EPS went from $0.03 to $0.14 year-over-year. U.S. revenue alone surged 42% YoY. Target operating margins in the double digits. Earnings power is compounding fast.

None of the AI data center optionality is in the run rate. LiteWave800 doesn’t hit meaningful production until Q1 2027, right as the pluggable optical module TAM Goldman models hits $20.2B in scale-out alone. BEAD federal broadband funding is still a trickle. Stanton said it himself: “’27 starts to feel like a more material number.”

For context: $LITE trades at ~25x forward P/S. $COHR trades at ~9x, on a much larger revenue base and still growing 21% YoY. Both received $2B strategic investments from Nvidia in March 2026, a direct signal that the hyperscaler ecosystem is placing enormous bets on optical interconnect. $ADTN is entering that same ecosystem at 1.15x P/S with a product that out-powers their modules on the efficiency curve.

$HLIT is the direct playbook, a broadband peer that caught Bryan then Citrini’s attention, and caught massive momentum once the AI/data center pivot was understood. $ADTN is the same archetype.

---

What Wall Street Is Starting to See

Evercore ISI initiated at Outperform / $18 PT on April 14th, citing LiteWave800 as an upside catalyst and flagging $1.20+ annual EPS power in the medium term. The formal model uses $0.75 CY27 EPS, but Evercore puts the upside scenario at $24. Read between the lines: the $0.75 is the safe initiation anchor. The $1.20 is where they actually think this lands.

Post-Q1: B. Riley raised to $21. Rosenblatt raised to $20. Northland raised to $18. Every analyst who covers this moved higher after the print. The stock is at ~$17.

---

The Bottom Line

Goldman Sachs just published a 34-page report telling the world that optical networking is a 9x TAM expansion to $154 billion by 2028. The pluggable optical module market in scale-out alone goes from $8.3B today to $32.4B by 2028. The technology ADTN chose, VCSEL-based LPO, is what Goldman explicitly calls out as the high-efficiency, commercially-ready solution for the applications ADTN is targeting.

$ADTN is a 40-year optical engineering company that just entered the fastest-growing capex environment in networking, with a product that benchmarks the industry on power efficiency, a full-stack data center portfolio mapped across every layer Goldman Sachs identified, and a product family roadmap targeting AI infrastructure with the same power advantages displayed in LiteWave800.

The peers doing this work, $LITE and $COHR, trade at 25x and 9x forward P/S respectively. Nvidia just put $2 billion into each of them.

You can own the AI data center inflection at broadband valuations. For now.

Disclosure: Long both calls and shares.

---