·Dashboard·Research·Work·Archive
/
← wiki
ticker networkingoptical5gai-infra updated 2026-06-04

Nokia (NOK)

Identity

Field Detail
Full name Nokia Oyj (Nokia Corporation)
Ticker NYSE: NOK / HEL: NOKIA
Sector Communications Equipment / Network Infrastructure
HQ Espoo, Finland
Founded 1865 (telecom ops from 1960s)
Employees 72,000-77,000 (target after restructuring, from 86,000)
CEO Justin Hotard (since April 1, 2025)
CFO Marco Wiren
Website nokia.com

Nokia started as a Finnish pulp mill, became the world's dominant mobile phone maker, watched Apple obliterate that business, and emerged as the telecom infrastructure company it is today — bolstered by the 2016 Alcatel-Lucent acquisition that brought Bell Labs (9 Nobel Prizes) and a full optical/IP/fixed-line portfolio.

FY2025 revenue: EUR 19.9B (+3.5% YoY). But that blended number is misleading. There are really two Nokias inside this company.

Thesis

Nokia is an AI optical infrastructure company wearing a telco disguise, and the disguise is starting to slip.

The investment case hinges on a structural insight: Nokia's fastest-growing business — optical networking, up 22% in FY2025 — is being valued at a deep discount because it's bundled with a flat-to-declining mobile networks business that still represents 39% of revenue. Strip away the telco overhang, and Nokia looks like an AI infrastructure company trading at 1.5x EV/Revenue versus Ciena's 6-7x.

Two converging megatrends drive the thesis:

  1. AI datacenter buildout — hyperscaler capex is projected to exceed $600B in 2026, ~75% tied to AI. All that compute needs optical pipes. AI clusters spanning multiple data centers require all-to-all GPU communication at terabit speeds, and only coherent optical networking can deliver that bandwidth. Nokia's optical orders were up 40% YoY, DC switching orders surged 150%.

  2. AI-native RAN — the long-term transformation of mobile networks into AI compute platforms. NVIDIA's $1B strategic investment (Oct 2025, 2.9% stake) validates this vision, but it's a 2028+ revenue story. Don't buy the stock for this alone.

The Infinera Acquisition ($2.3B)

This is the strategic pivot that makes the thesis work. Closed February 2025, it gave Nokia:

  • The only in-house InP PIC fab owned by any network equipment OEM (Sunnyvale, CA)
  • Hyperscaler customer relationships Nokia badly lacked
  • A second DSP design team for faster roadmap execution (PSE-7 + ICE-8)
  • ~20% global optical market share — roughly equal to Ciena, behind only Huawei

Synergy target: EUR 200M in net comparable operating profit by 2027.

The NVIDIA Validation

NVIDIA invested $1B for a 2.9% stake in October 2025. Nokia committed its entire baseband roadmap to NVIDIA platforms — an effectively irreversible decision. Co-developing AI-RAN products using NVIDIA CUDA/ARC-Pro. T-Mobile anchoring trials in 2026, commercial products targeted 2027.

NVIDIA doesn't write billion-dollar checks for marketing. They want Nokia to build AI-RAN products on NVIDIA platforms and integrate Nokia's optical/switching tech into NVIDIA's AI infrastructure stack.

The Google Deal

On March 2, 2026, SemiAnalysis reportedly published that Nokia won a ~$1.5-2B optical equipment contract at Google. Shares jumped ~7%. Neither party confirmed officially, but:

  • Nokia had announced its "first 800G pluggable award from a US hyperscaler" in Q2 2025
  • Jefferies called it "additional evidence of Nokia gaining market share in optics, notably against Ciena"
  • Ciena's own CSO acknowledged Nokia as the primary rival: "We're really looking at us and Nokia with the combination of Infinera"

A deal this size would be a multi-year framework agreement (3-5 years), transformational for the Optical Networks division.

Sum-of-Parts Valuation

The conglomerate discount is where the opportunity lives:

Segment Revenue Multiple Implied Value
Network Infrastructure EUR 7.6B (+22%) 3.0x EV/Rev (50% Ciena discount) ~$24B
Mobile Infrastructure EUR 11.4B 0.8-1.0x EV/Rev ~$10-12B
Nokia Technologies EUR 1.4B @ ~78% OP margin 10-12x operating profit ~$13-16B
Total implied equity $50-64B ($9.35-$11.75/share)

Current price ~$8.24. Current enterprise value ~$46B. The gap exists because the market still categorizes Nokia as a "telco equipment vendor" rather than pricing each business at what a buyer would actually pay.

Conviction: Medium

The optical story is compelling, but the stock has already re-rated 62% in the past year and trades above the $7.01 consensus target. Execution on Infinera integration and PSE-7/ICE-8 delivery is the key variable. Entry at $6.50-7.00 on a pullback creates much better risk/reward.

Business

Nokia operates across four legacy segments (restructured to two effective January 2026). Each tells a fundamentally different story.

Segment Overview (FY2025, old structure)

Segment Revenue (EUR M) % Total YoY Change Operating Margin
Network Infrastructure 7,986 40% +22.5% ~11.7% (improving)
Mobile Networks 7,806 39% -4.3% ~5.3%
Cloud & Network Services 2,606 13% +0.7% 8.2%
Nokia Technologies 1,501 8% -22.2%* ~78.5%
Total 19,889 100% +3.5% ~10.2%

*Technologies decline is optical — FY2024 had EUR 400M+ in catch-up licensing payments inflating the comparison.

New two-segment structure (effective Jan 1, 2026): Network Infrastructure (EUR 7.6B) and Mobile Infrastructure (EUR 11.4B), plus Portfolio Businesses (EUR 845M). The logic is to separate the growth engine from the mature business and make the optical story visible to investors.

Geographic Revenue Mix (FY2024)

Region % of Revenue
Europe ~33%
North America ~28%
Asia-Pacific ~24%
Rest of Americas ~5%
MEA / Other ~10%

Network Infrastructure — The Crown Jewel

EUR 7.6B revenue, +22.5%, the growth engine.

Three sub-segments: Optical Networks (primary growth driver, +15-19% quarterly in 2025), IP Networks (FP5-powered routing, +24% in Q4 2024), and Fixed Networks.

By Q2 2025, NI overtook Mobile Networks as Nokia's largest segment for the first time ever. AI & Cloud customers accounted for 29% of Optical Networks revenue by Q3 2025, growing ~1 percentage point per quarter. Hyperscalers represented 5% of total group sales (~EUR 225M/quarter).

Nokia targets 10-12% net sales CAGR for combined Optical + IP Networks through 2028, with operating margins expanding from ~10% to 13-17%.

Optical Networking Deep-Dive

This is the core of the growth thesis. Understanding coherent optics is essential for evaluating whether the Infinera acquisition was worth $2.3B and whether Nokia can compete with Ciena.

How fiber optic communication works:

Fiber optic communication encodes data as pulses of light through thin glass fibers. Light travels through a glass core surrounded by cladding with a slightly lower refractive index, bouncing via total internal reflection for kilometers with minimal loss.

Property Copper Wire Fiber Optic
Bandwidth ~10 Gbps max 100+ Tbps per fiber
Distance ~100m before degradation 100+ km between amplifiers
Interference Susceptible to EMI Immune

Single-mode fiber (~9um core) allows only one light path, extending reach to hundreds of km. This runs the internet's backbone.

Wavelength Division Multiplexing (WDM):

A single fiber carries many different wavelengths (colors) of light simultaneously, each with an independent data stream. DWDM (Dense WDM) uses tight 100/50 GHz channel spacing and supports 80-96+ channels. A single fiber pair: 96 wavelengths x 800 Gbps = ~76.8 Tbps. The industry is within 2-3x of the theoretical Shannon limit.

Coherent optical technology — the game changer:

The old way (direct detection): Vary laser brightness — on = 1, off = 0. Topped out at ~10 Gbps per wavelength. Morse code with photons.

The coherent breakthrough: Modulate the amplitude, phase, and polarization of light simultaneously across both polarization states. At the receiver, a local oscillator laser "beats" against the incoming signal, recovering the full electric field. Instead of 1 bit per symbol, coherent systems pack 4, 6, 8, or even 12 bits per symbol. The spectral efficiency improvement is dramatic.

Inside a coherent optical engine:

A coherent engine comprises three tightly integrated building blocks:

  1. A digital ASIC (the DSP — digital signal processor)
  2. Analog electronics (DACs and ADCs)
  3. Photonics (lasers, modulators, receivers — often as a TROSA)
TRANSMIT SIDE                              RECEIVE SIDE
+--------------+                            +--------------+
| Tunable      |---> CW light               | Local        |
| Laser        |                            | Oscillator   |---> Reference
+--------------+                            | Laser        |     light
      |                                     +------+-------+
      v                                            |
+--------------+                            +------v-------+
| IQ Modulator | <-- DAC <-- Tx DSP         | 90 deg       |
| (4 MZMs for  |                            | Optical      |
| dual-pol)    |---> Modulated light --->   | Hybrid       |
+--------------+     through fiber          +------+-------+
                                                   |
                                            +------v-------+
                                            | Balanced     |
                                            | Photodetects |---> ADC ---> Rx DSP
                                            | (4 pairs)    |
                                            +--------------+

Key components explained:

  1. Tunable laser — generates continuous-wave light at a precisely controlled wavelength across the C-band (~1530-1565nm). Also used as local oscillator on receive side.

  2. IQ Modulator — nested Mach-Zehnder interferometers. A single MZM splits light into two arms, applies different phase shifts, recombines via interference. An IQ modulator nests two MZMs: one for In-phase, one for Quadrature, with 90-degree offset. For dual-polarization: four MZMs total (I and Q for each polarization state). Maps any point on the IQ constellation diagram onto the optical field.

  3. DSP ASIC — the brains. Among the most complex chips outside GPUs. Compensates chromatic dispersion, polarization mode dispersion, laser phase noise, fiber nonlinearities, and applies FEC. On transmit: modulation encoding, pulse shaping, pre-compensation, probabilistic constellation shaping. On receive: CD compensation (can compensate thousands of km), polarization demultiplexing, PMD compensation, clock recovery, carrier recovery, nonlinearity compensation, FEC decoding, real-time performance monitoring. This is why coherent optics couldn't exist until ~2008 — CMOS had to catch up.

  4. DACs/ADCs — convert between digital (DSP) and analog (photonic) domains. Modern DACs at 130+ Gbaud sample rates in 5nm CMOS.

  5. Balanced coherent receivers — optical hybrid mixes incoming signal with local oscillator. Four balanced photodetector pairs convert optical signals to electrical currents. "Balanced" detection cancels common-mode noise, doubling signal amplitude while rejecting DC and noise.

Modulation formats — the reach vs. capacity tradeoff:

Format Bits/Symbol (dual-pol) Capacity Typical Reach Use Case
PM-QPSK 4 1x baseline >3,000 km Subsea, ultra-long-haul
PM-8QAM 6 1.5x ~1,500 km Long-haul
PM-16QAM 8 2x ~800 km Regional/metro-core
PM-32QAM 10 2.5x ~300 km Metro
PM-64QAM 12 3x <120 km DCI

Going from PM-QPSK to PM-16QAM doubles capacity but requires 4x the OSNR (6 dB). Going from PM-16QAM to PM-64QAM adds only 50% more capacity but requires another significant OSNR jump. Diminishing returns hit hard at higher orders.

Probabilistic Constellation Shaping (PCS): Instead of using all constellation points with equal probability, PCS biases transmission toward lower-power inner points. Creates a Gaussian-like distribution approaching the Shannon capacity limit. Nokia pioneered commercial PCS (now 3rd generation). Enables fine-grained rate adaptation, ~1 dB sensitivity gain, and reduced fiber nonlinearity penalty.

Baud rate progression (the Moore's Law of optics):

Era Baud Rate CMOS Node Per-Wavelength
~2010 28 Gbaud 40nm 100G
~2015 35 Gbaud 28nm 200G
~2018 56-64 Gbaud 16nm 400G
~2023 96-100 Gbaud 7nm 800G
~2024 130 Gbaud 5nm 1.2T
~2026+ 200 Gbaud 3nm 1.6T

Ciena's WaveLogic 6 Extreme shipped the industry's first 1.6T at 200 Gbaud in late 2024 — 72 customers. Nokia's PSE-7/ICE-8 won't ship until late 2026 at earliest.

Soft-decision FEC: The receiver assigns probabilities to each bit rather than hard 1/0 decisions. Feeds iterative LDPC decoding. Provides ~3 dB coding gain — roughly 2x noise tolerance, meaning 2x longer reach or higher modulation. Modern FEC overhead: 15-25% of total bit rate.

Nokia's Coherent Engine Portfolio

PSE-6s (Nokia, current flagship):

  • 5nm FinFET CMOS
  • 130 Gbaud, 1.2 Tb/s per wavelength
  • 2.4 Tb/s in dual-chip superchannel (industry first)
  • 40% power/bit reduction vs. prior gen
  • 94% reduction in W/Gbps vs. first-gen coherent
  • 800GbE over 2,000 km reach
  • 3rd-gen Probabilistic Constellation Shaping
  • Integrated with Nokia CSTAR silicon photonics (3rd gen)

ICE7 (Infinera, current flagship):

  • 5nm CMOS DSP, 148 Gbaud, 1.2 Tb/s
  • 800G over 3,000 km — 3x+ reach of prior gen at this speed
  • 60% power/bit savings, 30% cost/bit reduction vs. ICE6
  • Custom DSP tightly integrated with Infinera's InP PIC TROSA
  • Covers 85%+ of network links at 800G

PSE-7/ICE-8 (in development):

  • 200 Gbaud, 1.6T target — the response to Ciena WL6e
  • Both in active development, long-term goal is to converge
  • Nokia also targeting 3.2T for future generations

ICE-X 800G ZR/ZR+ (pluggable):

  • DCI growth product
  • Volume ramp: thousands (2025) -> >100K (2026) -> >1M (2030)

ICE-D (intra-DC):

  • Specialized for AI/ML data center interconnect
  • 1.6T, 75% power/bit reduction vs. existing solutions
  • $2.2B+ market by 2027

CSTAR Silicon Photonics:

  • Nokia's in-house SiP platform (3rd gen) — Coherent Silicon Transmitter And Receiver
  • Sold internally and externally to third-party pluggable DCO makers
  • Variants: CSTAR-200+ (up to 34 Gbaud) and CSTAR-400 (up to 69 Gbaud)

Infinera's Indium Phosphide — The Crown Jewel Technology

This is the single most important technology asset from the $2.3B acquisition.

Why InP matters (fundamental physics):

  1. Native light generation. InP generates, amplifies, modulates, and detects light on the same chip. Silicon cannot generate light — needs external lasers. Fundamental physics advantage.

  2. 10x modulation efficiency. InP's electro-optic effect is up to 10x more efficient than silicon's plasma dispersion. More compact, lower-power modulators.

  3. Monolithic integration. Infinera integrates ALL photonic functions — laser, modulator, SOA (semiconductor optical amplifier), photodetector, multiplexer — onto a single InP chip. No wire bonds, no coupling losses.

  4. On-chip amplifiers. InP PICs include SOAs, impossible in silicon. Compensates on-chip losses, extends reach.

The downside: Smaller wafers (3-4" vs. 12" silicon), less mature fab ecosystem, higher cost per chip. For volume pluggable applications, silicon photonics wins on cost. For high-performance coherent engines, InP is superior.

Infinera's Sunnyvale fab is now Nokia's — the only in-house InP PIC fab owned by a network equipment OEM. Only Huawei has comparable vertical integration.

Infinera Platform Products

GX Series (Compact Modular Platform):

  • Next-generation open optical transport
  • Modular chassis supporting 100G to 1.2T per wavelength (with ICE7)
  • Super C+L band support (35% more fiber capacity)
  • Multi-haul: metro, regional, long-haul in one platform
  • Open architecture, programmable, automation-ready

XTM Series:

  • Edge/metro transport platform
  • Supports pluggable 400G coherent optics (QSFP-DD and CFP2 form factors)
  • Compact footprint for access/aggregation

The Optical Equipment Stack

Site A                      Intermediate           Site B
+----------+   ~80km fiber  +---------+   ~80km   +----------+
| Router    |               |  ROADM  |           | Router    |
|    |      |               | (add/   |           |    ^      |
|Transponder|--->[EDFA]--->  | drop/   |--->[EDFA]--|Transponder|
|(Coherent) |               | pass)   |           |(Coherent) |
|  ROADM    |               +---------+           |  ROADM    |
+----------+                                       +----------+
  <----- 96 wavelengths x 800G = ~76.8 Tbps ----->
  • Transponders — convert client data to DWDM wavelengths using coherent engines
  • ROADMs (Reconfigurable Optical Add-Drop Multiplexers) — traffic cops, remotely directing wavelengths
  • EDFAs (Erbium-Doped Fiber Amplifiers) — boost all wavelengths every 80-100 km

The 400ZR Pluggable Revolution

Hyperscalers asked the OIF in 2016 if coherent could be shrunk into pluggable modules fitting directly into router faceplates. The answer: 400ZR.

  • 400ZR: 400G in QSFP-DD/OSFP pluggable, point-to-point DWDM, up to 120 km reach, DP-16QAM. Plug into a router and go.
  • OpenZR+ (400ZR+): Multiple modulation formats, reach up to 1,400 km, rates from 100G-400G. Starting to cannibalize embedded coherent territory.
  • 800ZR: OIF released IA October 2024. Doubles to 800G via ~118 Gbaud and 4nm DSP. Volume production 2025-2026.
  • 1600ZR: OIF standardization launched 2024. Target: commercial ~2027-2028. Will leverage 3nm or 2nm CMOS.

The trend: Pluggable is eating embedded from the bottom up. By 2023, 400G pluggables became the most widely adopted coherent technology in history. Over 70% of coherent ports are now pluggable. Each generation reaches further and carries more data. Embedded shrinks to the most demanding applications.

Data Center Interconnect (DCI) — Why It Matters

DCI is the high-speed optical plumbing connecting geographically distributed data centers into a unified computing fabric. As hyperscalers build AI training clusters spanning multiple facilities, DCI has become the critical infrastructure bottleneck.

Three tiers:

  • Intra-DC (<2 km): servers, GPUs, switches. Market >$16B in 2025, growing 60%+ annually. AI racks need 16-36x more fiber than traditional cloud racks.
  • Metro DCI (<100 km): 400ZR/800ZR pluggables. Captures 60% of DCI revenue. Exploding because hyperscalers are building "scale-across" architectures linking adjacent facilities into single virtual AI compute clusters.
  • Long-haul DCI (hundreds-thousands of km): Sophisticated optical line systems. Dell'Oro projects 12% five-year CAGR.

Total optical transport equipment market: $16B in 2025 (Dell'Oro), growing ~10%/yr. DCI sub-segment growing 40-50% annually. Broader DCI market estimated at $15-16B in 2025, growing 11-15% CAGR to $25-42B by 2030-2032.

IP Routing

Nokia's FP5 network processor — 6.0 Tb/s per chip, 75% power reduction versus FP4 — powers the 7750 Service Router family. Won hyperscaler deployments at CoreWeave, Apple, CoreSite, and Microsoft. First with 800GE interfaces.

SR Linux — Nokia's network operating system, co-developed with Apple. Modern microservices architecture designed specifically for hyperscaler operators who previously had to strip vendor software and write their own. Genuinely differentiated.

Nokia is one of very few Western vendors offering world-class capabilities across optical transport, IP routing, AND data center switching. Ciena lacks IP routing. Arista has no optical. Cisco has deprioritized standalone optical transport. Only Nokia can sell a hyperscaler a complete WAN-to-data-center networking stack.

Mobile Networks — The Albatross

EUR 7.8B revenue, -4.3%, the drag.

The global RAN market is ~$35B and has been essentially flat since 2024 after steep declines from the 2021 peak. Dell'Oro expects flat through 2026 with a 1% CAGR through 2030. The next real capex wave won't arrive until 6G approaches late in the decade.

The margin profile compounds the problem: 5.3% operating margin in FY2024, less than half NI's 11.7%. In the first nine months of 2025, the segment posted a EUR 64M operating loss on EUR 5.3B revenue.

The telco drag dilutes Nokia's story four ways:

  1. Revenue gravity — still ~57% of revenue under new structure (Mobile Infrastructure includes Core + Tech/patents)
  2. Margin dilution — blending 5% margin with 10-12% and 78% produces mediocre ~10% group margin
  3. Cyclicality — 78.5% of 2024 revenue from CSPs, a structurally flat customer base
  4. Narrative contamination — investors categorize Nokia as "telco vendor," suppressing the multiple

The decline has been brutal — Mobile Networks fell from EUR 10.7B (2022) to EUR 7.7B (2024), a 27% contraction. India alone collapsed 41% in 2024, accounting for 7pp of Nokia's 9% group revenue decline that year. The RAN market lost roughly $9B in revenue between 2021 and 2024 (Dell'Oro).

Silver lining: stabilization. Q4 2024 showed Mobile Networks declining just -2% CC — dramatic improvement from -39% in Q1 2024. Nokia targets 6-9% operating margins by 2026 via aggressive cost restructuring (72,000-77,000 employee target versus 86,000 at start). AT&T's $14B award to Ericsson was a blow, but position with other major operators remains intact. A RAN recovery remains a potential upside surprise, but no analyst is modeling one.

Product family: AirScale — Habrok (Massive MIMO), Doksuri (AI-era, 30% more efficient), Tuuli (cost-efficient). ~15-17% global RAN share. Lost AT&T $14B to Ericsson; won T-Mobile extension.

AI-RAN Architecture

AI replaces static, rule-based network configurations with dynamic, data-driven optimization.

RAN components (modern 5G cell site):

YOUR PHONE                                                    INTERNET
   |                                                              |
   | (radio waves)                                                |
   v                                                              |
+----------+     fronthaul     +----------+   midhaul   +------+  |  backhaul
| ANTENNA  |  <============>  |    DU    |  <=======>  |  CU  |  | <========>  CORE
| + Radio  |   (fiber, eCPRI) | Distrib. |  (Ethernet) | Cent.|  |  (fiber)    NETWORK
|  Unit    |                  |   Unit   |             | Unit |  |
+----------+                  +----------+             +------+
 Cell tower                   Near site               Central
  • Radio Unit (RU): On tower. Up to 64 antenna elements (Massive MIMO), power amplifiers, DAC/ADC.
  • Distributed Unit (DU): Real-time processing (~1ms latency). Channel coding, beamforming, scheduling. Where NVIDIA GPUs want to replace ASICs.
  • Centralized Unit (CU): Non-real-time. Connection management, handover, security. Often virtualized on standard servers.

Massive MIMO and beamforming: MIMO uses multiple antennas to send multiple data streams simultaneously. Massive MIMO uses 64+ antenna elements, increasing spectral efficiency 3-5x over 4G. Beamforming steers radio beams toward specific users — spotlight vs. floodlight. Those large rectangular panels on modern cell towers are massive MIMO active antenna units.

The signal chain — what happens when you stream a video (downlink):

  1. IP packets arrive at CU from core network
  2. CU handles connection management, PDCP processing (header compression, ciphering)
  3. DU schedules which resource blocks (time slots x frequency chunks) to use
  4. DU performs channel coding, modulation, beamforming weight calculation
  5. Digital baseband samples sent via fronthaul to RU
  6. RU converts digital to analog RF via DACs
  7. Power amplifiers boost signal
  8. Antenna elements transmit shaped radio waves toward your phone

Key asymmetry: tower downlink power is 50-100x your phone's uplink power (~20-40W vs. ~0.2W). This is why upload is always slower than download.

Open RAN vs. Traditional:

Traditional RAN = single-vendor proprietary black box. All-in-one RU + DU + CU with proprietary interfaces. Vendor lock-in: buy Ericsson radios, need Ericsson baseband, Ericsson software, Ericsson management. No mixing allowed. Highly optimized but creates oligopoly pricing power.

Open RAN (O-RAN Alliance, founded 2018, 946 technical documents) disaggregates with standardized interfaces:

  1. Disaggregation: Split into O-RU, O-DU, O-CU from different vendors
  2. Open interfaces: Standardized protocols (fronthaul, midhaul, E2, A1, O1)
  3. Virtualization: DU/CU on standard COTS servers instead of proprietary hardware
  4. Intelligence: New component — the RAN Intelligent Controller (RIC) — uses AI/ML to optimize
  5. Cloud-native: Containerized RAN software on Kubernetes

Open RAN market: $3.98B in 2025 (~11% of $35B RAN market), growing to ~$25B by 2031. But much of "Open RAN" spending still goes to traditional vendors (Nokia, Ericsson, Samsung) who support O-RAN interfaces. Pure-play vendors (Mavenir, Parallel Wireless) haven't demonstrated scale. Nokia and Ericsson are co-opting O-RAN into their integrated products, blunting the disruption narrative.

O-RAN Architecture:

+------------------------------------------------------------------+
|                    SERVICE MANAGEMENT &                            |
|                    ORCHESTRATION (SMO)                             |
|  +-----------------------------------------------------------+   |
|  |                   NON-RT RIC                               |   |
|  |   +--------+  +--------+  +--------+  +--------+          |   |
|  |   | rApp 1 |  | rApp 2 |  | rApp 3 |  | rApp N |          |   |
|  |   | Energy |  | Policy |  | SLA    |  |  ...   |          |   |
|  |   | Saving |  | Mgmt   |  | Assure |  |        |          |   |
|  |   +--------+  +--------+  +--------+  +--------+          |   |
|  +-----------------------------------------------------------+   |
|           | A1 interface (policies, ML models)                    |
+-----------|------------------------------------------------------+
            v
+---------------------------+
|       NEAR-RT RIC         |
| +------+ +------+ +-----+|
| |xApp 1| |xApp 2| |xApp ||
| |Traffic| |Beam  | |Inter||   E2 interface
| |Steer | |Optim | |ference|  (real-time control)
| +------+ +------+ +-----+|
+-----------|---------|-----+
            v         v
  +--------+  +--------+  +--------+
  |  O-CU  |  |  O-CU  |  |  O-CU  |    Higher-layer processing
  +--------+  +--------+  +--------+
       |           |           |        Midhaul (F1 interface)
  +--------+  +--------+  +--------+
  |  O-DU  |  |  O-DU  |  |  O-DU  |    Real-time L1/L2
  +--------+  +--------+  +--------+
       |           |           |        Fronthaul (Open FH, split 7-2x)
  +--------+  +--------+  +--------+
  |  O-RU  |  |  O-RU  |  |  O-RU  |    Radio + antenna
  +--------+  +--------+  +--------+

Key O-RAN interfaces:

  • A1: Non-RT RIC -> Near-RT RIC (policies, ML model updates) — timescale >1 second
  • E2: Near-RT RIC -> O-CU/O-DU (real-time control and monitoring) — timescale 10ms to 1s
  • O1: SMO -> all RAN components (management, configuration, fault/performance data)
  • Open Fronthaul: O-DU -> O-RU (split 7-2x, eCPRI protocol)
  • F1: O-CU -> O-DU (3GPP-defined)

Key distinction: xApps directly control RAN functions (adjusting beam weights, scheduler parameters). rApps influence behavior indirectly through policies ("prioritize energy savings midnight to 6am" or "maintain >50 Mbps per user in this slice").

AI in RAN — What It Actually Does

1. Network Energy Efficiency

RAN accounts for ~80% of a mobile network's total energy consumption. Traffic varies dramatically — 2am might be 10% of noon traffic, yet radios burn power 24/7.

  • Traditional approach: fixed time-based schedules ("turn off small cells midnight to 6am")
  • AI approach: ML models predict traffic per-cell, per-hour based on history, events, weather, day of week. Proactively sleep cells before traffic drops, wake before it surges.
  • Advanced: reinforcement learning agents continuously optimize while maintaining coverage/QoS
  • Results: Nokia ~15% additional savings network-wide; with aggressive AI, average 25% reduction. Industry pilots: 6-10% conservative, 20-25% aggressive.
  • Other: adaptive MIMO layer shutdown, dynamic power amplifier bias, predictive cooling, symbol-level micro-sleep during empty OFDM symbols

2. Beamforming and MIMO Optimization

Traditional beamforming (codebook-based, grid-of-beams) uses fixed patterns and exhaustive search — O(N^2) or worse, slow to adapt, suboptimal in real environments.

AI improvements:

  • ML-based beam prediction: Predicts optimal beam pair from user position, velocity, partial channel measurements. Achieves 99% of exhaustive search at fraction of compute.
  • Real-time beam tracking: Deep learning predicts future beam directions, enables proactive switching before signal degrades.
  • Environment-aware beamforming: Neural networks learn specific propagation of each site (reflections, blockages) — something static codebooks can't do.
  • Multi-user MIMO scheduling: AI optimizes which users to serve simultaneously and with which configuration.
  • Results: SINR/throughput up to 400% in mmWave. Energy efficiency up to 5.3 Mbps/W with ML vs. 0.9 without (~6x improvement).

3. Spectrum Management

  • Dynamic spectrum sharing (DSS): AI optimizes 4G/5G split in real-time. Multi-agent RL achieves >85% throughput with convergence in 24ms.
  • Interference detection: AI detects subtle patterns (external interference, PIM, hardware degradation) invisible to traditional alarms.
  • Carrier aggregation optimization: AI dynamically selects which carriers to aggregate per-user, per-moment.

4. Predictive Maintenance

  • AI analyzes streams from towers, fiber, switching to predict failures before they happen
  • Detects: power amplifier drift, antenna VSWR changes, fiber attenuation, temperature anomalies
  • 60%+ reduction in unnecessary truck rolls, 30-40% better first-time fix rates
  • For 10,000 dispatches/month at $300/roll, a 60% reduction = $1.8M/month savings

Nokia's AI-RAN Software Stack

MantaRay SON (Self-Organizing Networks):

  • Nokia's flagship AI optimization platform — 120+ operator customers
  • Industry leader in RAN automation 10 consecutive years (TERAL Research)
  • Cognitive SON enables TM Forum Level 4 autonomy (highly autonomous, cannot be reached without AI)
  • Functions: self-configuration, self-optimization, self-healing
  • Intelligent beamforming, multi-vendor support (deployed in NTT DOCOMO's multi-vendor 5G)

MantaRay SMO (Service Management and Orchestration):

  • Full O-RAN-compliant SMO framework
  • Non-RT RIC with 30+ Nokia rApps at launch
  • MantaRay AutoPilot: AI-powered orchestration engine
  • Only SMO meeting TM Forum Level 4 automation specs
  • Open ecosystem — Ericsson recently joined Nokia's SMO Marketplace

MantaRay RIC (Near-RT):

  • O-RAN-compliant near-real-time RIC
  • Advanced Traffic Steering xApp: ML-based dynamic traffic distribution
  • Acquired Juniper's RIC technology and 45-person team (Oct 2025)
  • Previous: Nokia and AT&T successfully trialed O-RAN RIC with native E2 on commercial 5G

Nokia AVA Platform: Broader network analytics — data collection, ML model training/deployment, feeds MantaRay.

Nokia AirScale Product Family

AirScale Macro Radios: Comprehensive portfolio for all spectrum bands and technologies (4G/5G). Massive MIMO variants (32T32R, 64T64R). O-RAN compliant fronthaul.

AirScale Habrok: Massive MIMO radios. Nokia's performance flagship.

AirScale Doksuri (new, AI-era): Launched 2025-2026, engineered for AI-RAN. 30% improvement in power efficiency vs. prior gen, 25% lighter. "AI-ready" — designed for integration with NVIDIA-powered baseband.

AirScale Tuuli: Energy-optimized, compact outdoor baseband. Cost-efficient 5G upgrade path. Traditional ASIC-based (not GPU).

AirScale Cloud RAN (anyRAN): Virtualized DU/CU as containers on COTS servers. Container-as-a-Service layer. vDU L1 acceleration via RAN SmartNIC (in-line) or GPU (NVIDIA path). Partners: Cisco, HPE, Microsoft, AWS, Dell. Live: du (UAE) first commercial 5G SA Cloud RAN in MEA; Orange evaluating on hybrid AWS.

NVIDIA's AI-RAN Vision — The Big Bet

Nokia committed its entire baseband roadmap to NVIDIA platforms — an effectively irreversible strategic decision.

NVIDIA Aerial Platform:

  • cuPHY: CUDA library for L1 physical layer — channel coding (LDPC, Polar codes), modulation/demodulation, MIMO processing, channel estimation — all on GPU
  • cuMAC: GPU-accelerated L2 MAC scheduler
  • cuBB: Complete GPU-accelerated 5G/6G signal processing pipeline
  • All processing stays within GPU high-bandwidth memory for minimal latency

NVIDIA ARC-Pro (Aerial RAN Computer Pro): Latest hardware platform purpose-built for AI-RAN. Nokia designing base stations based on ARC-Pro. Integration with Dell PowerEdge servers.

The dual-use proposition: Peak traffic -> GPU handles RAN. Low traffic -> same GPU runs edge AI inference (computer vision, gen AI, analytics). Claims 2-3x capacity utilization improvement. Eliminates siloed infrastructure.

Deployment progress (March 2026):

  • T-Mobile: Tested GPU-accelerated AI-RAN at Seattle Innovation Centre. Nokia AirScale Massive MIMO (3.7 GHz) on NVIDIA GPU supported video streaming, gen AI queries, AI video captioning.
  • BT, Elisa, NTT DOCOMO, Vodafone: All engaging with AI-RAN powered by NVIDIA Aerial
  • Indosat (Indonesia): Southeast Asia's first AI-RAN L3 5G call using Nokia AirScale + NVIDIA GPUs
  • MWC 2026: Nokia demonstrated functional GPU-accelerated AI-RAN tests

Timeline: Field trials 2026 -> first commercial products 2027 -> 6G-ready architecture beyond.

What's Deployed Today vs. Roadmap

Shipping and generating real value (2026):

  • MantaRay SON with Cognitive SON (Level 4 autonomy) — 120+ customers
  • MantaRay SMO with non-RT RIC and 30+ rApps
  • MantaRay RIC (near-RT) — trials and early deployments
  • AirScale radios including Doksuri — standard product
  • Cloud RAN (anyRAN) — commercial deployments underway
  • AI-driven energy savings, beamforming optimization, traffic steering

On roadmap (2026-2028):

  • NVIDIA GPU-based baseband (trials 2026, commercial 2027)
  • Full AI-RAN with converged RAN + AI inference on same GPU
  • 6G-ready AI-native architecture
  • Expanded xApp/rApp ecosystem with third-party developers
  • Agentic AI for autonomous network operations

Aspirational / marketing-heavy:

  • "AI-native 6G" — pure roadmap, 2030+ reality
  • Converged RAN + AI inference on shared GPU — concept proven, business case unproven at scale
  • Edge AI revenue from operators — unclear if operators want this or know what to run

Revenue impact timeline:

  • 2025-2026: AI-RAN is a cost story, not revenue. Differentiates RAN equipment but no incremental AI-specific revenue.
  • 2027-2028: If GPU products ship, could drive RAN sales uplift. Real value is market share gain, not new category.
  • 2029+: If converged model works at scale, Nokia gets both RAN revenue AND edge AI platform. Transformative but speculative.

The honest assessment: SON and energy savings are deployed and real (120+ customers). GPU-RAN is functional in labs and limited field tests. But the full converged vision is 2028+, with genuine risk that Ericsson's custom silicon proves more power-efficient. A GPU burning 300-400W vs. an ASIC at 50W for the same RAN processing is a hard sell to operators obsessed with energy costs and site power budgets. 65% of telcos admit their AI initiatives haven't delivered expected value. The NVIDIA $1B validates the strategic direction, but NVIDIA also has obvious commercial motivation to see GPUs deployed everywhere.

Nokia vs. Ericsson — fundamentally different bets:

Nokia Ericsson
Strategy NVIDIA GPUs — general-purpose compute Custom ASICs with AI accelerators
Bet RAN as distributed AI compute platform AI embedded in purpose-built RAN
Partner NVIDIA ($1B investment) AWS (no equivalent deal)
SON platform MantaRay — 120+ customers, 10yr leader Competitive but behind in analyst rankings
Risk GPU power/cost overhead Cloud-native shift makes custom silicon obsolete

The market decides over the next 3-5 years. Notable: despite competing, in March 2026 Ericsson and Nokia announced a landmark collaboration on autonomous networks — Ericsson joined Nokia's SMO Marketplace, Nokia joined Ericsson's rApp Ecosystem.

End Markets & TAM

Market Size (2025) Growth Nokia Position
Optical transport $16B ~10%/yr #3 globally, #2 Western DCI
DCI sub-segment $15-16B 40-50%/yr Primary growth driver
RAN ~$35B Flat (1% CAGR to 2030) #3 globally, ~15-17% share
Hyperscaler capex >$600B (2026) +36% YoY Indirect — sells the optical/IP pipes
Patent licensing ~$15B global pool Stable Top 3 licensor

Secular tailwinds: AI datacenter buildout (optical pipes for GPU clusters), Huawei Western exclusion (durable oligopoly), 6G cycle (2028-2030 catalyst for RAN recovery).

Secular headwinds: Telco capex structurally flat, RAN in 4-5 year doldrums, pluggable optics commoditization pressure.

Value Chain Position

[Semiconductor/DSP] -> [Optical Components] -> [Engines] -> [Equipment OEMs] -> [Telcos/Cloud]
       * Nokia DSP                              * Nokia           * Nokia

Nokia is vertically integrated across three layers in optical. Designs own chipsets (ReefShark) in RAN. Owns Infinera's InP PIC fab. Key suppliers: TSMC (DSP fabrication), NVIDIA (GPU partnership). Key customers: T-Mobile, Vodafone, Deutsche Telekom (telco); Google, Microsoft, Meta, CoreWeave (hyperscaler — growing fast via Infinera relationships).

Nokia Bell Labs in Murray Hill, NJ remains one of the world's premier industrial research labs. Infinera's San Jose InP fab is the crown jewel manufacturing asset.

Nokia divested Alcatel Submarine Networks to the French state for EUR 350M (closed Jan 2025) and took full ownership of Nokia Shanghai-Bell (Chinese JV) for ~EUR 500M.

Nokia Technologies — The Hidden Gem

EUR 1.3-1.5B recurring, ~90%+ gross margin, ~78% operating margin.

Patent licensing across 5G, WiFi, video, multimedia. Over 20,000 patent families. Apple, Samsung, Huawei, OPPO, vivo, Honor, Xiaomi all signed to multi-year deals. The smartphone renewal cycle is complete — no renewal cliff for several years. Expanding into automotive, consumer electronics, IoT licensing.

FY2025 decline (-22.2%) is optical — FY2024 had EUR 400M+ one-time catch-up payments. Underlying run-rate is stable.

At 10-12x operating profit, this business alone is worth EUR 12-15B — roughly a quarter of Nokia's current enterprise value. As a standalone entity, it would command a premium multiple.

Portfolio Businesses

EUR 845M revenue, EUR 90M operating loss. Non-core: FWA customer premises equipment, site implementation, enterprise campus edge. Earmarked for potential divestiture in 2026 — modest but real value unlocking.

Competitive Position

Optical Market (2025 Estimates)

Vendor Global Ranking Share Est. DCI Ranking Key Trend
Huawei #1 ~25-28% Not a factor in Western DCI Dominant in China/emerging
Ciena #2 ~22-25% #1 Gained ~3pp in 2025; US dominance
Nokia #3 ~18-20% #2 Gained ~2pp post-Infinera
ZTE #4 ~10-12% Minor Share declining
Cisco #5 ~6-8% #3 Gained ~1pp; pluggable focus

Total market: $16B (2025) -> ~$17.6B (2026).

Ciena — Nokia's Primary Optical Rival

The uncomfortable truth: Ciena is winning the technology race right now.

Specification Nokia PSE-6s Ciena WaveLogic 6e
CMOS process 5nm 3nm
Baud rate 130 GBaud 200 GBaud
Max single-carrier rate 1.2 Tb/s 1.6 Tb/s
800G unregenerated reach ~2,000 km 4,000-5,000 km
Live 1.6T deployment Not yet Aug 2024 (Arelion, 470 km)
Customers on latest gen — 72

Ciena holds an estimated 12-18 month technology lead in high-end 1.6T optics by jumping to 3nm CMOS. They hold nearly 50% U.S. optical market share and gained 3pp of global share in 2025 alone.

Ciena financials (March 2026):

  • Revenue: $4.77B FY2025 (+19%), TTM $5.12B (+27%)
  • FY2026 guidance: $5.7-6.1B (20%+ growth)
  • Market cap: ~$51B
  • EV/Revenue: 10.1x | EV/EBITDA: 90.6x | Forward P/E: 55x
  • Gross margin: 42.0% | Net margin: 4.5%
  • Added to S&P 500 February 2026

Ciena's strengths vs. Nokia: Clear tech lead at high end, dominant in hyperscaler DCI, pure-play optical focus.

Ciena's weaknesses vs. Nokia: Extreme valuation (priced for perfection), thin margins (4.5% net), less geographic diversity, no RAN or broader network portfolio to bundle.

Nokia counters with: dual-chip 2.4T superchannel architecture, InP PIC vertical integration (Ciena is fabless for photonics), broader portfolio (optical + IP + switching), and PSE-7/ICE-8 in development targeting 1.6T parity.

The thesis doesn't require Nokia to "beat" Ciena. It requires Nokia to be competitive enough to capture share in a market that's structurally undersupplied. Ciena entered 2026 "essentially sold out" — both companies are benefiting from a market growing faster than either can supply.

Infinera vs. Ciena vs. Cisco/Acacia — technology comparison:

Dimension Infinera (Nokia) Ciena Acacia (Cisco)
Photonics platform InP PIC (monolithic) Hybrid (merchant SiP + own DSP) Silicon Photonics
Latest DSP ICE7 (148 Gbaud, 5nm) WL6e (200 Gbaud, 3nm) Acacia CIM 8 (TBD)
Max per-wavelength 1.2T (ICE7), 2.4T (dual PSE-6s) 1.6T (first, single-carrier) ~800G shipping
Key advantage Vertical integration, InP, monolithic DSP lead, first to 1.6T Silicon economics, Cisco distribution
Fab ownership Own InP fab (San Jose) Fabless (TSMC for DSP) Own SiP fab (Maynard, MA)

Cisco / Acacia

Acacia fully integrated, now central to Cisco's "Internet for the Future" strategy. Pioneer and market leader in coherent pluggables — over 70% of coherent ports are now pluggable, and Cisco/Acacia drove this shift. New 3nm Kibo PAM4 DSP (1.6T capable) sampling late 2025 for intra-DC optical. Strategy: collapse IP and optical layers via pluggable coherent in their routing/switching installed base. Ranked #5 optical transport, #3 DCI, gained 1pp share in 2025.

This isn't head-to-head competition in optical line systems. It's a long-term structural threat — pluggable optics embedded in routers eliminating the need for standalone optical gear. Affects Nokia and Ciena equally.

Huawei

#1 optical transport vendor globally by revenue share. #1 across all six major telecom infrastructure categories (Dell'Oro). Controls ~40-41% of global telecom equipment market excluding North America.

  • Banned/restricted: US, UK, Australia, Japan, Sweden
  • Dominant: China (near-total domestic), much of Asia, Africa, Middle East, LatAm
  • Contested: Continental Europe (bans announced but enforcement is slow — Nokia and Ericsson CEOs publicly frustrated)

Massive R&D spend, US sanctions paradoxically strengthened domestic supply chain. Recent win: Algeria Telecom national 400G WDM backbone (Feb 2025).

For Nokia investors: Huawei is the elephant, but mostly in markets Nokia doesn't target head-to-head. The bigger risk is Huawei gaining ground in Europe. In optical DCI specifically, Huawei doesn't compete in the hyperscaler space driving growth — that's Nokia/Ciena/Cisco.

Other Optical Competitors

Adtran (formerly ADVA): EUR 438M revenue in 2024 (-29%). ~$1.2B market cap. Metro/access optical, enterprise. Not a real competitor at scale.

Ribbon Communications (RBBN): $834M revenue in 2024. ~$690M market cap. IP optical, comms software. Niche player.

HPE/Juniper: $14B acquisition completed July 2025. Juniper's strength is routing/switching, not optical line systems. New PTX 12,000 routers designed for next-gen optical networks. Could become more relevant if they push into disaggregated optical.

RAN Market (2025 Estimates)

Vendor Global Share Trend Key Markets
Huawei ~32-34% Gaining China, emerging, parts of Europe
Ericsson ~27-29% Stable N. America, Europe, Japan, Korea
Nokia ~15-17% Slight gain Europe, Japan, parts of N. America
Samsung ~5-7% Stable US (Verizon), Japan, Korea
Others ~4% Stable Open RAN, NEC, Fujitsu

Total market: ~$35B (flat). Top 5 = 96% of market.

Samsung in RAN

#5 globally, ~$1.4B network revenue (first 9 months 2024, -27% YoY). Anchored by Verizon ($6.65B 5G deal 2020-2025). Strong in vRAN and Open RAN, leverages semiconductor expertise for custom chipsets. Verizon's new CTO is a strong Open RAN/Samsung advocate. Growing in Japan.

The contained threat: Samsung has struggled to break beyond a few anchor customers. Revenue decline is concerning. Most likely beneficiary if Open RAN takes off, but that's still a slow burn.

Ericsson — The Primary RAN Rival

Ericsson is executing better by most financial measures:

Metric Nokia Ericsson
EV/Revenue 2.0x 1.5x
EV/EBITDA 16.0x 9.5x
Forward P/E 24.1x 17x
Gross margin 43.5% 48.1%
Operating margin 7.7% 13.7%

Nokia's 68% EBITDA premium over Ericsson is entirely the optical/AI narrative. If that narrative falters, the premium compresses violently. At Ericsson's 9.5x EBITDA, Nokia trades at ~$5.00 — 40% downside.

Industry Structure (Porter's Five Forces)

  • Supplier power: Moderate. Custom silicon reduces dependence on merchant chips.
  • Buyer power: High. A handful of operators control most spend and negotiate ruthlessly.
  • New entrant threat: Low. $4-5B annual R&D, decades of standards influence, deep switching costs.
  • Substitutes: Low-moderate. Pluggable optics, LEO satellite, Wi-Fi 7 all chip at edges.
  • Rivalry: Very high. Oligopoly with massive fixed costs fighting over a flat market.

Bottom line: Tough industry. The saving grace is the oligopoly structure and Huawei's exclusion from Western markets, which creates a protected arena for Nokia and Ericsson.

Emerging Threats

Three structural threats to monitor beyond the obvious competitive dynamics:

  1. Ciena's 1.6T lead solidifying before Nokia ships competitive products. If PSE-7/ICE-8 slips past mid-2027, Ciena locks in hyperscaler DCI relationships that become very sticky. First-mover advantage in optical design wins is real — operators don't like to re-qualify mid-deployment.

  2. Pluggable optics commoditization via merchant DSPs. Marvell, Broadcom, and others are building merchant coherent DSPs that any module maker can buy. If coherent becomes a commodity (like SFP transceivers), Nokia's vertical integration advantage in embedded systems erodes. The InP differentiation helps on performance, but the pluggable market — which is eating embedded — could commoditize faster than expected.

  3. Ericsson's custom silicon proving more efficient than Nokia's GPU approach. If operators conclude that ASIC-based RAN is definitively more power-efficient and cost-effective than GPU-based RAN, Nokia's irreversible NVIDIA commitment becomes a strategic liability. Ericsson's approach is lower-risk technically even if Nokia's has more theoretical upside.

Management

CEO Justin Hotard — Deep Profile

First American-born CEO in Nokia's history. 25+ years in tech, heavy on data center, AI, HPC.

Career timeline:

Period Company Role
2025-present Nokia President & CEO
Feb 2024-Feb 2025 Intel EVP & GM, Data Center & AI Group (DCAI)
2015-2024 HPE EVP & GM, HPC, AI & Labs (final role)
Earlier NCR, Symbol Technologies, Motorola Senior leadership; started as engineer deploying mobile networks

Education: BS Electrical Engineering, UIUC. MBA, MIT Sloan.

Headline achievement at HPE: Delivered the world's first exascale supercomputer — Frontier for DOE Oak Ridge. 1.1 exaflops, faster than the next seven largest combined, simultaneously #1 on Green500 for energy efficiency. Under his watch, HPE's AI/HPC unit grew revenue to $1.18B/quarter (+37% YoY). Not a career bureaucrat — the man built and shipped the most powerful computer ever constructed.

Intel stint: Brief (~1 year). DCAI was already hemorrhaging share to AMD and NVIDIA. Honest assessment: not enough runway to judge. When Nokia called, he left — which says more about his view of Nokia's upside than Intel's mess.

No controversies, no legal issues. Clean record.

First-year actions at Nokia (Apr 2025 - Mar 2026):

  1. Simplified four segments into two (NI + MI)
  2. Set 2028 targets: EUR 2.7-3.2B comparable OP
  3. Landed NVIDIA's $1B investment — the signature deal of year one
  4. Completed Nokia Shanghai-Bell buyout (~EUR 500M)
  5. Restructured Group Leadership Team
  6. Set growth KPIs: 6-8% net sales CAGR, 10-12% Optical + IP

Skin in the game: Invested EUR 2.8M of his own money in Nokia shares via co-investment LTI (609,274 shares at EUR 4.63, locked 3 years with 2:1 matching performance shares). That's real conviction, not just sign-on grants. Total target comp remains below U.S. median benchmark.

LTI performance metrics: 50% relative TSR, 40% cumulative EPS, 10% GHG emission reduction. Well-designed — TSR aligns with shareholders, EPS prevents revenue growth at profitability's expense, ESG component meaningful but not dominant.

Former CEO Pekka Lundmark (2020-2025)

Solid B+ CEO. Inherited Nokia at a low point (stumbled on 5G, stock at $4-5). Restored 5G tech leadership, executed massive portfolio restructuring (ASN divestiture, Infinera acquisition, Shanghai-Bell buyout), returned EUR 1.5B+ via well-timed buybacks (avg EUR 3.83-4.69 vs. current EUR 7.50+), reinstated dividends, laid NVIDIA partnership groundwork. Margins improved even as revenue fell in the industry downturn.

Year Net Sales (EUR B) Comp. OP (EUR B) Comp. OP Margin
2020 ~21.9 ~1.6 ~7.3%
2022 ~24.9 ~3.1 ~12.5%
2024 ~19.3 ~2.6 ~13.5%
2025 ~19.9 ~2.0 ~10.1%

Genuinely planned succession — he initiated it, stayed as Special Advisor through end 2025.

CFO Marco Wiren

Joined alongside Lundmark in September 2020. Proper industrial CFO (ex-Wartsila, SSAB). Nokia has generally met or exceeded guidance under his watch. FY2024 delivered EUR 2.6B vs. 2.3-2.9B guide (above midpoint). FY2025 delivered EUR 2.0B vs. ~1.85B midpoint. FCF conversion: 72%. Not heroic, not sandbagged — appropriate for a company in transition.

Board of Directors

  • Incoming Chair: Timo Ihamuotila — former Nokia CFO (2009-2016), leaving ABB for this role. Knows the DNA.
  • Interesting addition: Meredith Whittaker — Signal Foundation president, AI ethics advocate. Unconventional but smart for AI credibility with European regulators.
  • 9/10 proposed members independent. Five committees including a dedicated Technology Committee (unusual but appropriate given EUR 4.3B+ annual R&D).
  • Solidium (Finnish state) at 5.96%. Passive, supportive holder — keeps Nokia Finnish-controlled.

Insider Ownership & Transactions

CEO skin in the game: Hotard holds 609,274+ shares plus sign-on restricted shares. EUR 2.8M personal co-investment at EUR 4.63/share, locked 3 years with 2:1 performance matching. That's real money, not just option exercises.

Recent insider transactions (last 12 months):

Person Action Shares Price Date
Justin Hotard (CEO) Co-investment buy 609,274 EUR 4.63 Jun 2025
David Heard (President NI) Share incentive 65,122 — Oct 2025
Tommi Uitto (departing) Sold 57,589 EUR 6.32 Oct 2025*

*Uitto sold right after NVIDIA announcement spiked the stock. He departed Group Leadership Team at end 2025. His associate also sold 35,686 shares at EUR 5.49.

Small insider buys in Feb 2026 (Sahgal: 2,391 shares; Heard: 251; Fisk: 174) — insignificant in dollar terms. On balance, insider activity leans positive — CEO has genuine skin in the game, selling concentrated around one departing executive.

NVIDIA's 2.9% stake: $1B at $6.01/share, ~166.4M new shares. No board seats — strategic partnership, not governance. Second-largest shareholder after Solidium.

Solidium (Finnish state) at 5.96%: Steadily accumulating since initial 3.3% purchase in 2018. Passive, supportive. National strategic interest — keeps Nokia Finnish-controlled.

Capital Allocation Grade: B+

Action Assessment
Buybacks (EUR 1.5B+, avg EUR 3.83-4.69) Excellent timing — stock now EUR 7.50+
Infinera ($2.3B) Bold, strategically sound. Jury out on execution
ASN divestiture (EUR 350M) Smart portfolio pruning, removed EUR 1B low-margin rev
Shanghai-Bell buyout (~EUR 500M) Reasonable for full China control, EUR 200M synergies
Dividend reinstatement Measured, sustainable — EUR 0.14/share, well-covered
R&D (EUR 4.0-4.5B annually, ~22% of rev) Essential for competitive positioning, consistent through cycles

Buyback detail:

Program Period Amount Shares Avg. Price
Phase 1 Feb-Nov 2022 EUR 300M ~65M ~EUR 4.60
Phase 2 Jan-Nov 2023 EUR 300M 78.3M ~EUR 3.83
Infinera offset Nov 2024-Apr 2025 EUR 900M cap 150M ~EUR 4.69

These were value-creating buybacks. Average prices of EUR 3.83-4.69 vs. current EUR 7.50+ means substantial value accrual.

Compensation & Governance Detail

CEO compensation vs. peers:

CEO Company Total Comp (2024, approx.)
Pekka Lundmark (Nokia) Nokia ~EUR 6-8M
Borje Ekholm Ericsson SEK 71.6M (~EUR 6.3M)
Gary Smith Ciena $14.1M (~EUR 13M)

Hotard's first-year all-in was elevated (~EUR 12-14M) due to sign-on packages, but ongoing will be ~EUR 7-9M. Board explicitly stated total target comp "remains below the US median benchmark."

STI metrics: Revenue, operating profit, free cash flow — weighted, adjusted annually. FY2025 payout: 129% of target for Hotard's CEO portion.

LTI metrics: 50% relative TSR, 40% cumulative EPS, 10% GHG reduction (scopes 1, 2, 3). Well-designed — TSR aligns with shareholders, EPS prevents growth-at-any-cost, ESG meaningful but not dominant.

Governance provisions:

  • Double-trigger change of control (both event AND termination required)
  • CEO severance: up to 18 months total comp
  • Two clawback policies: (1) erroneously awarded comp from restatements (3-year lookback), (2) reputational damage, willful breach, gross misconduct
  • SBC runs ~1-2% of revenue — modest vs. US tech companies

Legal/regulatory: No current SEC enforcement. Legacy Alcatel-Lucent FCPA settlement ($137M in 2010) for bribing officials in Costa Rica, Honduras, Malaysia, Taiwan — pre-Nokia ownership, fully resolved with enhanced compliance. Clean record since.

No activist investors. Recent AGMs saw all board proposals approved without significant opposition. Broad Finnish retail base (~108,000 shareholders at 2023 AGM).

Management DD verdict: Green. Nokia's management and governance are genuinely strong. The CEO transition was well-executed, the new CEO has relevant experience and real skin in the game, capital allocation is disciplined, and the board is independent and well-composed. Management isn't the risk — execution on the AI pivot is.

Financials

Valuation Snapshot

Metric Value
Price ~$8.24
Market cap $48.3B
Enterprise value $45.9B
Forward P/E 24.1x
EV/EBITDA 16.0x
P/FCF 28.6x
FCF yield 3.5%
Dividend yield 1.25%
52-week range $4.00 - $8.66

Income Statement (EUR Millions)

Metric FY2023 FY2024 FY2025 FY2026E
Revenue 21,138 19,220 19,889 ~20,800
Revenue growth — -9.1% +3.5% +4-5%
Gross margin 40.4% 46.1% 43.5% ~44-45%
Comparable OP ~2,400 2,584 2,024 2,000-2,500
Comparable OP margin ~11% 13.4% 10.2% ~10.5-12%
Net income 665 1,277 651 —
Comparable EPS — EUR 0.39 EUR 0.29 ~EUR 0.33-0.36

FY2024 GM elevated by Tech licensing catch-ups. FY2025 comp OP declined from integration costs + RAN headwinds.

Cash Flow & Balance Sheet (EUR Millions)

Metric FY2023 FY2024 FY2025
Free cash flow 665 2,021 ~1,500
FCF margin 3.1% 10.5% ~7.4%
FCF conversion — — 72%
Net cash 2,600 3,500 2,000
Debt/equity — — 0.21
ROIC — — 5.77%

Fortress balance sheet — EUR 2B net cash after Infinera and Shanghai-Bell. ROIC at 5.77% sits below estimated 6.5% WACC — technically destroying value on invested capital. Patent licensing (high ROIC) masks mediocre hardware returns. Needs to improve toward 7%+ as synergies materialize.

2028 Targets (CMD 2025)

Metric Target
Comparable OP EUR 2.7-3.2B
NI sales CAGR 6-8%
NI operating margin 13-17%
FCF conversion 65-75%
Optical + IP CAGR 10-12%

Incremental Margin Analysis

Quarterly seasonality (Q4 = 30%+ of revenue) and restructuring charges distort incrementals. The directional story:

  • Network Infrastructure (+22.5%) is the healthy incremental driver — operating margin hit 19.6% in Q4 2024
  • Mobile Networks (-4.3%) is the drag — posted operating loss in first 9 months of 2025
  • Mix shift toward optical improves incrementals 2026-2028 as NI becomes majority of revenue
  • Infinera synergies (EUR 200M by 2027: ~1/3 COGS, ~2/3 opex) directly boost margins
  • Key watch: NI operating margin expansion from ~10% toward 13-17% 2028 target

Q4 2025 comparable gross margin hit 48.1% — matching Ericsson's figure and showing the direction of travel as licensing and optical mix increases. Full-year at 43.5% is depressed by H1 integration costs and Mobile Networks weakness.

Peer Valuation Comparison

Metric Nokia Ericsson Ciena
EV/Revenue 2.0x 1.5x 10.1x
EV/EBITDA 16.0x 9.5x 90.6x
Forward P/E 24.1x 17x 55x
Gross margin 43.5% 48.1% 42.0%
Operating margin 7.7% 13.7% 8.2%
R&D % of revenue ~23% ~21% ~18%

Nokia's premium to Ericsson is the optical/AI narrative. Ciena's valuation is in another stratosphere — any execution miss and those multiples compress violently.

The Valuation Gap with Ciena

Nokia and Ciena compete directly in optical networking, yet their valuations diverge dramatically:

Metric Nokia (NOK) Ciena (CIEN) Ciena Premium
EV/Revenue 1.5x 6-7x ~4-5x
Forward P/E ~17x ~52-57x ~3x
EV/EBITDA (adj) ~11x ~28-44x ~3-4x
Net cash/(debt) EUR 3.4B net cash ~$200M net debt —
Revenue growth +3% (reported) +19% —
Gross margin ~46-48% (comparable) ~42-43% (adj) Nokia higher

The discount exists for legitimate reasons: Nokia's blended revenue growth is anemic because Mobile Networks drags; Ciena is pure-play optical directly riding the AI wave; Ciena joined the S&P 500 in February 2026 triggering passive inflows; Nokia is a Finnish ADR with conglomerate structure investors penalize.

But the discount appears excessive. Nokia's gross margin actually exceeds Ciena's thanks to patent licensing. And the sum-of-parts analysis (see Thesis section) implies $9-12 per share — 25-60% upside.

The re-rating catalyst is structural: new two-segment reporting makes NI growth visible, Infinera acquisition gives optical scale, NVIDIA validates AI-RAN, and the appointment of an AI/data center CEO signals the pivot.

What the Market Is Pricing

At $8.24, the market assumes: successful Infinera integration, NI growing 6-8%, comparable OP reaching ~EUR 3B by 2028, AI-RAN narrative intact.

Stretched: 68% premium to Ericsson on narrative alone. Trading above all consensus targets except Morgan Stanley $8.00. Simple DCF (EUR 1.5B FCF, 3-4% growth, 8% discount rate) yields fair value EUR 6.50-7.50. Stock is at the upper end.

Upside: If optical beats expectations, NI segment re-rates toward Ciena multiples. GPU-RAN = unpriced optionality. Patent licensing worth EUR 12-15B standalone.

Downside: At Ericsson's 9.5x EBITDA, Nokia trades at ~$5.00 — 40% downside.

Analyst sentiment: Morgan Stanley initiated Overweight at $8.00 (Jan 2026), "Top Pick," citing AI & Cloud growing from 6% of revenue at ~1pp per quarter. JP Morgan: $8.20 target. Jefferies: Buy, "top pick" in European telecom equipment. Stock +65-97% over past 12 months.

Consensus: Buy (12), Hold (6), Sell (4) of 21 analysts. Average target ~$7.01. Stock at $8.24 trades 18% above average.

Ownership

Holder % Outstanding
Solidium (Finnish state) 5.96%
NVIDIA (strategic) ~2.9%
FMR/Fidelity ~2.9%
Artisan Partners ~1.8%

Institutional: ~22.4% (513 institutions). Short interest: 0.70% (minimal).

Catalysts & Risks

What Could Go Right (6-12 Months)

  • Hyperscaler deal flow: Google win + engagements with Microsoft, Amazon, Meta, CoreWeave, Nscale. Combined hyperscaler capex >$600B in 2026.
  • New segment reporting (Q1 2026): NI growth clearly visible for first time. Could trigger analyst re-modeling and multiple expansion.
  • Infinera synergies: EUR 200M targeted by 2027, milestones throughout 2026.
  • PSE-7/ICE-8 milestones: The single most important near-term catalyst. Competitive 1.6T demonstration by late 2026 closes the Ciena gap.
  • ICE-X pluggable volume ramp: From thousands to >100K units in 2026.
  • T-Mobile AI-RAN trials: Proof-of-concept for NVIDIA partnership.
  • EU Huawei removal mandates: Forced RAN replacement directly benefits Nokia.
  • Patent licensing stability: All major licenses newly signed, no cliff for years.
  • Share buyback: EUR 900M program active, 150M shares to offset Infinera dilution.
  • Margin expansion: Targets EUR 2.0-2.5B comparable OP in 2026, EUR 2.7-3.2B by 2028.

What Could Go Wrong

Risk Likelihood Closeable?
Ciena extends 1.6T optical lead Medium-High Partially — PSE-7/ICE-8 ships late 2026/2027
RAN flat through 2028 High No — structural until 6G
Infinera integration stumbles Medium Yes — progressively through 2027
GPU-RAN fails commercially Medium Partially — depends on operator economics
CEO transition execution risk Medium Yes — 12-18 months to evaluate
Hyperscaler capex pullback Low-Medium No — macro risk
Pluggable optics commoditization Medium Partially — InP differentiation helps

Additional risk factors:

  • Huawei gaining European ground: Has gained ~3pp of global telecom equipment share since sanctions began. Dominates optical transport globally with ~29-31% share. European ban enforcement is patchy — if Huawei isn't fully removed from European networks, Nokia's protected market shrinks.
  • India market collapse: Revenue down 41% in 2024. Recovery uncertain.
  • Currency risk: Nokia reports in EUR with significant USD revenue. EUR/USD moves directly impact reported numbers.
  • Potential US tariff escalation: Additional uncertainty for cross-border equipment sales.
  • Ericsson Alcatel-Lucent precedent: Ericsson's troubled acquisition of Nokia's networks business in 2016 is a cautionary tale for telecom M&A integration. Infinera is simpler (optical only), but execution risk is real.
  • Merchant DSP commoditization: Marvell, Broadcom building merchant coherent DSPs. If coherent becomes commodity, vertical integration advantage erodes.

Bear case: RAN declines, Ciena lead solidifies, Infinera costs overrun, GPU-RAN stalls. De-rate to Ericsson 9.5x EBITDA -> ~$5.00 (40% downside).

Thesis invalidation (any one breaks it):

  1. Major hyperscaler optical customer loss (especially if Google deal doesn't materialize)
  2. PSE-7/ICE-8 delay past mid-2027
  3. NVIDIA downscaling partnership

Industry Cycle Position

The critical insight: Bottom of the RAN cycle, early innings of an AI-driven optical supercycle.

RAN / Mobile:

  • 5G capex peaked 2021-2022
  • Steep declines 2023-2024 (~$8-9B of RAN revenue evaporated industry-wide)
  • Stabilized 2025 at ~$35B
  • Expected to remain flat through 2026-2027
  • Next meaningful uptick: 6G investment toward end of decade
  • Dell'Oro: 1% CAGR through 2030
  • 5G Advanced / 5.5G: incremental software upgrades, not major capex
  • 6G: Standards ~2028-2029, commercial 2030+. That's a 4-5 year doldrums.

Traditional telco capex:

  • Global telecom capex declining at -2% CAGR through 2027
  • Capex-to-revenue ratio falling: 26.9% (2022) -> 22.9% (2024)
  • European telecoms actively deleveraging
  • Wireless capex specifically declining -3% CAGR through 2026

The counter-narrative — AI/DC optical:

  • Hyperscaler capex: $600B+ in 2026 (+36% from 2025), ~75% AI-related
  • Data center capex up 59% YoY in Q3 2025 — 8th consecutive quarter of double-digit growth
  • Approaching $1 trillion global DC capex in 2026
  • Optical interconnect in AI DCs: $9.9B (2025) -> projected $31B by 2033
  • AI cluster optical transceivers: $5B (2024) -> $10B+ (2026) — doubling in two years
  • 800G optics growing 60%+ in 2025
  • DCI-specific WDM grew ~40% in 2025 alone
  • Direct cloud provider optical purchases grew ~50% in 2025

Operator spending trends 2024-2026:

  • Wireless capex: declining at -3% CAGR
  • Fiber/broadband: positive but decelerating (US tipping cable to fiber in 2026)
  • Enterprise/private networks: growing but small base
  • Backhaul/DCI: growing, driven by AI traffic

This divergence is exactly why Nokia bought Infinera. The growth is in optical, not radio. Nokia is late to the DCI party — Ciena has the tech lead and customer relationships. Nokia needs PSE-7/ICE-8 to ship competitively to avoid losing more ground in the highest-growth segment.

Decision Log

Pre-Buy Checklist Summary (March 17, 2026 @ $8.24)

Rating Count Key Items
PASS 16 Business understanding, balance sheet (EUR 2B+ net cash), FCF (EUR 1.5B, 72% conversion), catalyst identification, uptrend, low short interest, capital allocation, governance
CAUTION 11 Tech moat (narrow not wide), pricing power (mixed), management transparency (Hotard too new), valuation (24x fwd P/E for 3% grower), FOMO risk
FAIL 3 ROIC < WACC (5.77% vs. 6.5%), above consensus target (18% above $7.01 avg), no margin of safety

Verdict: WAIT

Good company, wrong price. Nokia is a legitimate business with real catalysts, but the entry point is wrong.

At $8.24: 24x forward P/E for a 3% revenue grower with below-WACC ROIC, trading 15-25% above analyst consensus, after a 62% run. RSI overbought at ~75. Insider buying negligible.

What changes this to BUY:

  • Pullback to $6.50-7.00 (closer to consensus)
  • Two quarters of Hotard execution proving optical/AI-RAN stories are real revenue, not narrative
  • ROIC improvement above WACC (needs ~7%+)
  • Telco capex confirmation beyond one-off AI deals

Position Sizing Framework

  • Entry: Wait for pullback to $6.50-7.00
  • Size: 2-3% at initial entry
  • Add: PSE-7/ICE-8 success or accelerating optical share gains
  • Trim: >$9-10 without earnings upgrade
  • Re-evaluate thesis: Below $5.50

Key Open Questions

  1. Does the Google $1.5-2B deal get officially confirmed? Multi-year framework or one-time?
  2. Can PSE-7/ICE-8 ship by late 2026 and close the Ciena 1.6T gap?
  3. Will Infinera EUR 200M synergies materialize on schedule by 2027?
  4. Can Nokia's GPU-RAN approach overcome the power/cost disadvantage vs. Ericsson's custom silicon?
  5. Does the new two-segment reporting actually trigger analyst re-modeling and multiple expansion?
  6. Can ROIC improve from 5.77% to above WACC as integration costs roll off?

The Bottom Line

Nokia presents a rare setup: EUR 7.6B in high-growth infrastructure revenue (growing 22%), EUR 3.4B net cash, a EUR 1.4B pure-profit patent licensing business, trading at 1.5x EV/Revenue because investors can't look past the declining mobile business.

The thesis doesn't require Nokia to beat Ciena in optical tech. It requires three things: (1) continued hyperscaler order momentum at or above 10-12% CAGR target, (2) successful Infinera integration delivering EUR 200M synergies, and (3) the new reporting structure making the growth story visible to generalist investors.

But at $8.24 after a 62% rally, the asymmetry that made this compelling has eroded. Wait for the pullback. It will come — either from a broader correction, a quarter where optical growth decelerates, or the stock running out of momentum. When Nokia trades at $6.50-7.00 with the same fundamentals, the risk/reward tilts decisively in your favor. That's when you buy.

Sources

  • Nokia Q4/FY2025 Results; Nokia CMD 2025
  • Nokia Completes Infinera Acquisition (Feb 2025)
  • NVIDIA $1B Nokia Investment (Oct 2025)
  • Dell'Oro: Optical Transport Market $16B (2025), RAN Market Stabilized
  • Ciena FY2025 Results, WaveLogic 6 Extreme specs
  • Ericsson FY2025 Results
  • Infinera ICE7/InP PIC documentation
  • Nokia PSE-6s, CSTAR Silicon Photonics specs
  • Nokia MantaRay SON/SMO/RIC documentation
  • NVIDIA Aerial/AI-RAN platform documentation
  • SemiAnalysis: Nokia-Google optical deal report (Mar 2026)
  • Jefferies, Morgan Stanley, JP Morgan analyst coverage
  • O-RAN Alliance specifications
  • Hyperscaler capex projections (>$600B in 2026)
  • StockAnalysis, MarketBeat, TipRanks consensus data
  • Nokia remuneration/governance filings
  • Dell'Oro/Omdia market share estimates

Topics

Source updates (auto-maintained)

Intake (Jun 4, 26) - gtp-cn-revision-brief-2026-06-04

The article is a GTP Concept Note revision brief for Pink's GGGI work, referencing Nokia only in passing as an operating firm active in Vietnam via MOUs (Viettel–Nokia, VNPT–Nokia partnerships from the Oct 2025 Finland–Vietnam Strategic Partnership).

Relevant to your thesis: Confirms Nokia's enterprise/carrier presence in Southeast Asia, a minor positive for the Mobile Networks segment but tangential to the optical/AI-infrastructure bull case.

Source: intakefile://gtp-cn-revision-brief-2026-06-04.md

Intake (Jun 4, 26) - gtp-value-add-by-layer-2026-06-04

Nokia appears in this article only as an example of a Finnish layer-1 company with established local operations in Southeast Asia — it is listed alongside Wärtsilä, KONE, and others to illustrate firms that already have country offices and distribution, and therefore do not need basic market-access intermediation.

Relevant to your thesis: Tangential — flagged for review.

Source: intakefile://gtp-value-add-by-layer-2026-06-04.md

Intake (Jun 3, 26) - gtp-id-subsector-research

Nokia appears in this article only as a listed company in Indonesia's digital/IoT/comms backbone layer with a Jakarta office, alongside other Finnish firms; no material information about Nokia's core optical, mobile, or AI infrastructure business is provided.

Relevant to your thesis: Tangential — flagged for review.

Source: intakefile://gtp-id-subsector-research.md

Intake (Jun 3, 26) - gtp-cn-draft5-2026-06-03

Nokia (NOK) is cited as a Finnish-technology firm relevant to digital/energy intelligence in GTP's industrial green-transition program across Vietnam, Indonesia, and the Philippines, specifically in the monitoring/MRV and energy-efficiency layers.

Relevant to your thesis: Consistent with the bull case that Nokia's industrial digital/monitoring portfolio has emerging-market demand beyond hyperscaler optics, though this is BD-stage, not revenue.

Source: intakefile://gtp-cn-draft5-2026-06-03.md

Intake (Jun 2, 26) - gtp-cn-draft4-2026-06-02

The article is a GGGI concept note draft for a Finland-funded green transition program in Southeast Asia; Nokia (NOK) appears only as an incidental ticker mention, not as a subject of analysis.

Relevant to your thesis: Tangential — flagged for review.

Source: intakefile://gtp-cn-draft4-2026-06-02.md

Intake (Jun 2, 26) - gtp-cn-draft3-2026-06-02

The article is a GGGI/MFA Finland development program concept note for green transition work in Vietnam, Indonesia, and the Philippines; Nokia and NOK appear only as incidental text matches unrelated to the document's content.

Relevant to your thesis: Tangential — flagged for review.

Source: intakefile://gtp-cn-draft3-2026-06-02.md

Drop/Citrini Articles (Apr 22, 26) - Agentic Utilities - Citrini Research

Nokia's own Global Traffic Report is cited by Citrini, projecting agentic M2M traffic to grow from 66 EB/month in 2025 to 537 EB/month by 2034, validating the optical networking demand thesis.

Relevant to your thesis: Direct third-party validation of the AI-driven bandwidth explosion that underpins Nokia's Optical Networks growth story and the 40% order surge cited in the wiki.

Source: dropfile://Citrini Articles/Agentic Utilities - Citrini Research.PDF

Pages that link here (1)
Optical Networking & Photonics
Open in Obsidian