AGI Sector Scan — Networking, Connectivity & Telecom Infrastructure

Executive Summary

AGI workloads are the most bandwidth-intensive applications ever built. Training clusters already saturate every link from GPU-to-GPU interconnect (microseconds) to undersea cables (intercontinental). Recursive self-improvement amplifies this: each generation of smarter models demands larger clusters, more data movement, and lower-latency interconnect. The networking stack is the binding constraint on how fast intelligence can scale.

Highest conviction: Optical transceivers (800G/1.6T ramp), InfiniBand/high-speed interconnect, optical fiber/cable, and fiber connectors sit directly in the data path of every AI cluster.
Strong but broader: Network switches/routers, undersea cables, and network test & measurement benefit from aggregate bandwidth growth without being single-vendor bottlenecks.
Moderate: Cell towers, satellite comms, and 5G equipment get indirect lift from edge inference and AI-augmented networks but are not core to training infrastructure.
Lower: Traditional cable ISPs and legacy telecom hardware face secular decline that AGI does not reverse.

Narrow Sectors

Companies

High Verdict

Medium Verdict

Low Verdict

High — Direct, high-multiplier AGI demand driver; supply constrained or pricing power intact

Medium — Clear AGI tailwind but shared with secular trends; moderate supply elasticity

Low — Indirect or minimal AGI impact; competitive or declining dynamics dominate

Optical Transceivers & Components

High

How It Works

Optical transceivers convert electrical signals to light (and back) at the ends of fiber-optic cables, enabling high-speed data transmission between switches, servers, and across data center fabrics. Each generation (400G → 800G → 1.6T) roughly doubles bandwidth per port. Components include lasers (VCSELs, EMLs, CW lasers), photodetectors, modulators, and silicon photonics engines.

Supply / Demand

AI Demand

Every GPU in a training cluster needs multiple high-speed optical links. A 100k-GPU cluster can require 200k+ transceivers. The 800G-to-1.6T transition is being pulled forward 12-18 months by hyperscaler AI buildouts. Demand is growing 30-40% annually in the AI segment alone.

Supply Constrained?

Yes. 800G transceivers have been in allocation since 2025. Laser chip supply (especially high-power EMLs and silicon photonics co-packaged optics) is the bottleneck. Lead times stretched to 30-50 weeks. Capacity expansions take 18-24 months to come online.

Key US-Listed Companies

COHRCoherent

LITELumentum

AAOIApplied Optoelectronics

CIENCiena (optics + networking)

Optical Fiber & Cable

High

How It Works

Optical fiber is the physical glass medium that carries light signals over distances from meters (intra-data center) to thousands of kilometers (long-haul and undersea). Cable manufacturers draw ultra-pure glass into hair-thin fibers, bundle them into cables with protective jackets, and sell to telcos, hyperscalers, and enterprises. Single-mode fiber dominates long-haul; multi-mode handles short-reach data center links.

Supply / Demand

AI Demand

Data center interconnects, new AI campus builds (each requiring millions of fiber-meters), and long-haul capacity upgrades to connect distributed training clusters are all driving incremental fiber demand. Hyperscalers are ordering 2-3x their historical run rates. Fiber-to-the-home (FTTH) also provides a secular floor.

Supply Constrained?

Moderately. Post-COVID overcapacity in telecom fiber has been absorbed. Specialty fiber (bend-insensitive, ultra-low-loss for long-haul) is tighter. Corning has >55% global market share and pricing discipline. New capacity takes 12-18 months to bring online.

Key US-Listed Companies

GLWCorning

OFSFurukawa (ADR: FRKWF)

COMMCommScope

InfiniBand & High-Speed Interconnect

High

How It Works

InfiniBand and proprietary high-speed interconnects (NVLink, NVSwitch, Ultra Ethernet) provide the ultra-low-latency, high-bandwidth fabric that connects GPUs within a training cluster. These fabrics handle the all-reduce and all-to-all collective operations that dominate distributed training time. Latency is measured in microseconds, bandwidth in terabits per second aggregate across a cluster.

Supply / Demand

AI Demand

This is arguably the single most AI-specific networking category. Every large training cluster uses either InfiniBand (NVIDIA) or a custom interconnect. As clusters scale from 100k to 1M+ GPUs, interconnect complexity and cost grow super-linearly. Recursive self-improvement means ever-larger clusters, making this a compounding demand driver.

Supply Constrained?

Yes. NVIDIA's InfiniBand switches and cables have been in allocation alongside GPUs. The Ultra Ethernet Consortium is building an open alternative but is 18-24 months behind. Broadcom's custom interconnect ASICs for hyperscalers are also capacity-constrained.

Key US-Listed Companies

NVDANVIDIA (NVLink/NVSwitch/InfiniBand)

AVGOBroadcom (custom interconnect ASICs)

MRVLMarvell (interconnect PHYs, PAM4)

Fiber Connectors & Assemblies

High

How It Works

Fiber connectors (MPO/MTP, LC, SC) and cable assemblies terminate optical fibers and plug into transceivers, patch panels, and switches. Structured cabling systems, including trunk cables, harnesses, and cassettes, form the passive optical plumbing of every data center. High-density MPO connectors supporting 12, 24, or 32 fibers per ferrule are critical for high-radix AI fabrics.

Supply / Demand

AI Demand

AI data centers use 5-10x more fiber connections per rack than traditional cloud. Each GPU server connects to multiple top-of-rack switches via multi-fiber trunks. The shift to 800G/1.6T optics requires new high-density connector systems. Active optical cables (AOCs) and direct-attach copper (DAC) cables for short-reach GPU-to-switch links are in massive demand.

Supply Constrained?

Moderately. High-density MPO connectors and AOC/DAC assemblies are seeing extended lead times. Precision polishing and alignment requirements limit how fast capacity can scale. Amphenol has been adding capacity aggressively but still reports backlog growth.

Key US-Listed Companies

APHAmphenol

TELTE Connectivity

COHUCohu (connector test)

Network Switches & Routers

High

How It Works

Switches and routers direct data packets across networks. In AI data centers, leaf-and-spine switch architectures with high-radix (many-port) switches form the backbone that connects thousands of GPU servers. Each switch uses merchant-silicon or custom ASICs to forward packets at line rate (51.2 Tbps per chip in the current generation). Routers handle inter-data-center and WAN traffic.

Supply / Demand

AI Demand

AI clusters require fat-tree or Clos network topologies with far more switches per server than traditional workloads. A 100k-GPU cluster can need 10,000+ switches. The transition to 51.2T and 102.4T switch ASICs is being accelerated by AI bandwidth needs. Ethernet-based AI networking (via Ultra Ethernet) is expanding the addressable market beyond InfiniBand-only designs.

Supply Constrained?

Moderately. Switch ASIC supply (Broadcom Memory, custom chips) is the bottleneck more than switch assembly. White-box switch makers can ramp faster than branded vendors. The AI switching market is growing 40%+ annually.

Key US-Listed Companies

ANETArista Networks

CSCOCisco Systems

JNPRJuniper Networks

AVGOBroadcom (Memory/Memory2 switch ASICs)

Network Test & Measurement

High

How It Works

Test and measurement companies build instruments that validate, characterize, and troubleshoot optical and electrical links. This includes high-speed oscilloscopes, bit-error-rate testers (BERTs), optical spectrum analyzers, network protocol analyzers, and automated test systems for transceiver production. Every new speed generation (400G → 800G → 1.6T) requires new test equipment.

Supply / Demand

AI Demand

The accelerating cadence of new transceiver and switch ASIC generations is a direct demand driver. Every optical transceiver fab line, every switch ASIC validation lab, and every data center commissioning project needs test equipment. 800G and 1.6T testing is more complex (PAM4 modulation, coherent optics), increasing ASP per instrument.

Supply Constrained?

Not supply-constrained in the traditional sense, but these are high-ASP, low-volume businesses with strong pricing power. Keysight and Viavi have dominant positions and limited competition in next-gen optical/electrical test. Revenue growth tracks the speed-generation upgrade cycle.

Key US-Listed Companies

KEYSKeysight Technologies

VIAVViavi Solutions

TERTeradyne (production test)

Undersea / Submarine Cables

High

How It Works

Submarine cables carry 99%+ of intercontinental data traffic via fiber-optic strands laid on the ocean floor. A modern cable system contains 12-24 fiber pairs, each carrying multiple terabits per second using wavelength-division multiplexing (WDM). Cable ships lay the cables; repeaters (optical amplifiers) boost the signal every 60-100 km. A single trans-Atlantic cable costs $250-500M and takes 2-3 years to build.

Supply / Demand

AI Demand

Distributed AI training across geographies (US-Europe, US-Asia) and inference serving to global users drives massive new undersea capacity demand. Google, Meta, and Microsoft are each ordering dedicated private cables. The total submarine cable market is expected to double from ~$3B/yr to ~$6B/yr by 2028. AI-specific routes (connecting data center hubs) are the fastest-growing segment.

Supply Constrained?

Yes, severely. There are only ~4 cable ships in the world capable of laying modern submarine cables. SubCom (private, US-based), NEC, and Alcatel Submarine Networks (Nokia) are the only three manufacturers. Backlog is 3-4 years. This is one of the tightest bottlenecks in the entire AI supply chain.

Key US-Listed Companies

NOKNokia (ADR; Alcatel Submarine Networks)

NPTNNeoPhotonics (acquired by Lumentum)

GLWCorning (submarine fiber)

CIENCiena (submarine line systems)

Software-Defined Networking (SDN) & Network OS

Medium

How It Works

SDN separates the control plane (routing decisions) from the data plane (packet forwarding), allowing centralized, programmable network management. Network operating systems (NOS) like SONiC, Arista EOS, and Cisco NX-OS run on white-box or branded switches. SDN controllers orchestrate traffic flows, enable network automation, and optimize paths dynamically — critical for managing the complex topologies of AI clusters.

Supply / Demand

AI Demand

AI data centers need programmable, automated network fabrics that can handle rapidly changing traffic patterns (training vs. inference, model parallelism reshuffling). SDN and automation tools reduce the operational burden of managing 10,000+ switch networks. Hyperscalers build their own SDN stacks, but enterprises buying AI infrastructure need vendor-provided solutions.

Supply Constrained?

No. Software is inherently scalable. Competition is intense between open-source (SONiC) and proprietary (Arista, Cisco) stacks. The real constraint is the network engineering talent to deploy and tune these systems, not the software supply itself.

Key US-Listed Companies

ANETArista (CloudVision, EOS)

CSCOCisco (ACI, NX-OS)

VMWVMware/Broadcom (NSX)

Load Balancers, ADCs & Network Security Appliances

Medium

How It Works

Application delivery controllers (ADCs), load balancers, and next-generation firewalls sit at the edge of data centers and between network tiers, distributing traffic across servers, enforcing security policies, and optimizing application performance. Modern versions are increasingly software-defined and run on commodity hardware or in cloud-native environments. DDoS mitigation, WAF, and SSL offload are key functions.

Supply / Demand

AI Demand

AI inference endpoints need load balancing and security just like any web service, and the explosive growth of AI API traffic (billions of inference calls per day) drives incremental demand. AI also creates new attack surfaces that network security appliances must address. However, hyperscalers largely build their own load balancers in software, limiting the addressable market growth.

Supply Constrained?

No. This is a competitive market with ample supply from multiple vendors plus open-source alternatives. Pricing power is limited. The shift to software and cloud-native form factors is deflationary for hardware vendors.

Key US-Listed Companies

FTNTFortinet

PANWPalo Alto Networks

FFIVF5 Networks

Cell Towers & Telecom REITs

Medium

How It Works

Tower companies own and operate the steel-and-concrete structures (macro towers, rooftops, small cells) on which wireless carriers mount antennas and radio equipment. They earn recurring lease revenue from multiple tenants per tower. Data center REITs own and lease wholesale and retail data center space. Both are infrastructure REITs with long-duration contracts and high recurring revenue visibility.

Supply / Demand

AI Demand

Edge AI inference, autonomous vehicles, and AI-powered mobile applications increase wireless data demand, which drives carriers to densify their networks and lease more tower capacity. Data center REITs benefit directly from hyperscaler AI buildouts. However, tower demand growth is tied to carrier capex cycles, which are currently moderating post-5G initial buildout.

Supply Constrained?

Towers: Not meaningfully constrained — there is ample tower inventory in most US markets. Permitting for new builds is slow (12-24 months) but demand for net-new towers is moderate. Data center REITs: Increasingly constrained — power availability and permitting are bottlenecks for new data center construction.

Key US-Listed Companies

AMTAmerican Tower

CCICrown Castle

SBACSBA Communications

5G Equipment & Radio Access Network (RAN)

Medium

How It Works

5G RAN equipment includes base station radios (massive MIMO antennas), baseband processing units, and fronthaul/midhaul/backhaul connections that link cell sites to the mobile core network. Open RAN disaggregates these components, allowing mix-and-match from different vendors. RAN is the most capital-intensive part of a wireless network, representing 60-70% of carrier infrastructure spending.

Supply / Demand

AI Demand

AI at the edge (on-device inference, real-time AR/VR, autonomous systems) requires high-bandwidth, low-latency wireless connectivity that only 5G (and eventually 6G) can deliver. AI also makes RAN operations more efficient (AI-RAN, self-optimizing networks). However, the US 5G initial deployment wave has peaked; incremental demand is densification and mid-band expansion, which is slower growth.

Supply Constrained?

No. The 5G equipment market is a duopoly (Ericsson, Nokia) with Samsung and Open RAN vendors adding competition. Carrier capex is flat to slightly down in 2025-2026. Equipment pricing is under pressure. This market is demand-constrained, not supply-constrained.

Key US-Listed Companies

NOKNokia (ADR)

ERICEricsson (ADR)

COMMCommScope (antenna/RAN)

Satellite Communications

Medium

How It Works

Satellite comms provide connectivity from orbit — geostationary (GEO) satellites for broad coverage with higher latency, medium-earth-orbit (MEO) for moderate latency, and low-earth-orbit (LEO) constellations (Starlink, Kuiper) for low-latency broadband. Ground-station equipment, satellite buses, payloads, and user terminals form the value chain. LEO constellations require hundreds to thousands of satellites and continuous replenishment launches.

Supply / Demand

AI Demand

AI inference needs to reach every human on Earth, including billions without fiber or cable access. Satellite broadband is the path to universal AI access. Additionally, AI-powered satellite imagery analysis, IoT sensor networks, and autonomous maritime/aviation systems drive demand for satellite links. Direct-to-device (D2D) satellite services augment cellular coverage for AI-enabled phones.

Supply Constrained?

Mixed. Launch capacity has expanded dramatically (SpaceX). Satellite manufacturing capacity is scaling (Airbus, Maxar). User terminal production (phased-array antennas) remains a bottleneck for LEO operators. Spectrum is inherently finite and increasingly contested. The biggest US-listed exposure is through component and terminal suppliers since SpaceX/Starlink is private.

Key US-Listed Companies

IRDMIridium Communications

VSATViaSat

GSATGlobalstar

Wireless Infrastructure & Small Cells

Medium

How It Works

Small cells are low-power radio access nodes deployed on streetlights, utility poles, and building walls to densify wireless coverage in high-traffic areas. Distributed antenna systems (DAS) serve indoor venues like stadiums and airports. These complement macro towers by filling coverage gaps and adding capacity in dense urban environments. Private 5G/LTE networks (CBRS spectrum) for enterprises are a growing segment.

Supply / Demand

AI Demand

AI inference at the edge — powering real-time AR, autonomous retail, smart factories, and robotic warehouses — needs reliable, high-bandwidth indoor/outdoor wireless. Private 5G networks for AI-heavy enterprise use cases (manufacturing, logistics) are a clear growth vector. However, deployment pace is limited by permitting, utility pole access, and carrier budget cycles.

Supply Constrained?

No. Equipment supply is ample. The constraint is deployment velocity — municipalities control permitting, and carriers pace their investments. Small cell deployments have consistently undershot forecasts. The US has ~1M small cells deployed vs. 5M+ projected need for full 5G densification.

Key US-Listed Companies

CCICrown Castle (small cell focus)

CALXCalix (edge platforms)

DPI, Network Intelligence & Traffic Management

Low

How It Works

Deep packet inspection (DPI) and network intelligence platforms analyze traffic flows in real time to classify applications, enforce QoS policies, detect anomalies, and optimize bandwidth allocation. They sit inline on carrier and enterprise networks, inspecting packet headers and payloads. Use cases include carrier traffic management, lawful intercept, and network analytics.

Supply / Demand

AI Demand

AI inference traffic is growing but DPI vendors don't benefit proportionally — most AI traffic runs on private hyperscaler networks that don't use third-party DPI. AI could actually disrupt this market by enabling smarter, software-based traffic management that replaces dedicated DPI appliances. Net AI impact is likely neutral to slightly negative for incumbents.

Supply Constrained?

No. Competitive market with pricing pressure from software-defined alternatives and open-source tools. Legacy DPI vendors face secular headwinds from encrypted traffic (which renders DPI less effective) and carrier cost-cutting.

Key US-Listed Companies

NTCTNetScout Systems

SNPSSynopsys (network IP)

Cable & Fiber ISPs

Low

How It Works

Cable and fiber internet service providers (ISPs) own the last-mile infrastructure (coaxial cable, fiber-to-the-home) that connects households and businesses to the internet. Revenue comes from broadband subscriptions, with declining video (cord-cutting) and voice revenue. Fiber overbuilds (FTTH) are replacing legacy coaxial DOCSIS networks in many markets, requiring heavy capex.

Supply / Demand

AI Demand

AI inference consumed via cloud APIs does increase downstream bandwidth demand, but consumer broadband usage growth has been modest (5-10% annually) and ISPs have ample capacity on existing infrastructure. AI does not change the fundamental economics of last-mile access — it's a commodity pipe, and consumers don't pay more for it because they're using AI. Fixed wireless access (FWA) from T-Mobile/Verizon adds competitive pressure.

Supply Constrained?

No. Overbuild competition (fiber, FWA, satellite) is increasing in most US markets. ARPU growth is flat. The sector faces regulatory risk (net neutrality, broadband subsidies) and high capex requirements for fiber upgrades. Not a supply-constrained opportunity.

Key US-Listed Companies

CMCSAComcast

CHTRCharter Communications

FYBRFrontier Communications

Optical Networking Systems (DWDM / Coherent)

High

How It Works

Dense wavelength-division multiplexing (DWDM) and coherent optical systems multiplex dozens of wavelengths onto a single fiber pair, each carrying 400G-800G per wavelength. These systems form the long-haul and metro transport layer connecting data centers to each other and to internet exchange points. Coherent DSPs (digital signal processors) compensate for fiber impairments, enabling higher capacity over longer distances without regeneration.

Supply / Demand

AI Demand

Inter-data-center traffic is exploding as AI clusters span multiple sites. Training data must be replicated across geographies, model weights synchronized, and inference served globally. Every new AI data center campus needs multiple 400G/800G DWDM links to the outside world. DCI (data center interconnect) is the fastest-growing segment of optical networking, driven almost entirely by hyperscaler AI buildout.

Supply Constrained?

Moderately. Coherent DSP chips (Ciena WaveLogic, Marvell, Broadcom) have limited foundry capacity. Ciena has reported order-book growth exceeding 30% YoY. However, the optical transport system market is more competitive than transceivers, with Ciena, Nokia, Huawei (non-US), and Infinera all competing.

Key US-Listed Companies

CIENCiena (WaveLogic coherent DSP)

INFNInfinera (acquired by Nokia, delisted)

ADTNAdtran

Network Edge & Content Delivery Networks (CDN)

Low

How It Works

CDNs cache and serve content (video, web pages, software downloads) from edge servers located close to end users, reducing latency and backbone bandwidth usage. Edge computing extends this by running compute workloads (containers, serverless functions) at these edge locations. Revenue is typically usage-based (per GB delivered or per compute-second).

Supply / Demand

AI Demand

AI inference at the edge is a potential growth driver — serving AI responses from nearby edge PoPs reduces latency. However, hyperscalers (AWS CloudFront, Google Cloud CDN, Azure Front Door) have largely commoditized CDN pricing, and most AI inference runs in centralized data centers, not at the CDN edge. The traditional CDN revenue model (video delivery) faces deflationary pressure from improved video codecs and peering.

Supply Constrained?

No. CDN capacity is abundant and pricing is deflationary. Edge compute is a small market that hyperscalers can serve at cost. Pure-play CDN companies face margin pressure from scale disadvantage vs. cloud giants.

Key US-Listed Companies

NETCloudflare

AKAMAkamai Technologies

FSLYFastly

Master Company List — 48 Tickers

#	Ticker	Company	Narrow Sector	Verdict
1	COHR	Coherent	Optical Transceivers & Components	High
2	LITE	Lumentum	Optical Transceivers & Components	High
3	AAOI	Applied Optoelectronics	Optical Transceivers & Components	High
4	CIEN	Ciena	Optical Transceivers / DWDM / Undersea	High
5	GLW	Corning	Optical Fiber & Cable / Undersea	High
6	COMM	CommScope	Optical Fiber & Cable / 5G RAN	Medium
7	NVDA	NVIDIA	InfiniBand & High-Speed Interconnect	High
8	AVGO	Broadcom	Interconnect ASICs / Switch ASICs	High
9	MRVL	Marvell Technology	Interconnect PHYs / Coherent DSP	High
10	APH	Amphenol	Fiber Connectors & Assemblies	High
11	TEL	TE Connectivity	Fiber Connectors & Assemblies	High
12	COHU	Cohu	Fiber Connectors (Test)	High
13	ANET	Arista Networks	Network Switches & Routers / SDN	High
14	CSCO	Cisco Systems	Network Switches & Routers / SDN	High
15	JNPR	Juniper Networks	Network Switches & Routers	High
16	KEYS	Keysight Technologies	Network Test & Measurement	High
17	VIAV	Viavi Solutions	Network Test & Measurement	High
18	TER	Teradyne	Network Test & Measurement	High
19	NOK	Nokia (ADR)	Undersea Cables / 5G RAN	Medium
20	ERIC	Ericsson (ADR)	5G Equipment & RAN	Medium
21	ADTN	Adtran	Optical Networking Systems	High
22	FTNT	Fortinet	Network Security Appliances	Medium
23	PANW	Palo Alto Networks	Network Security Appliances	Medium
24	FFIV	F5 Networks	Load Balancers / ADC	Medium
25	AMT	American Tower	Cell Towers & Telecom REITs	Medium
26	CCI	Crown Castle	Cell Towers / Small Cells	Medium
27	SBAC	SBA Communications	Cell Towers & Telecom REITs	Medium
28	IRDM	Iridium Communications	Satellite Communications	Medium
29	VSAT	ViaSat	Satellite Communications	Medium
30	GSAT	Globalstar	Satellite Communications	Medium
31	CALX	Calix	Wireless Infrastructure / Edge	Medium
32	NTCT	NetScout Systems	DPI / Network Intelligence	Low
33	CMCSA	Comcast	Cable & Fiber ISPs	Low
34	CHTR	Charter Communications	Cable & Fiber ISPs	Low
35	FYBR	Frontier Communications	Cable & Fiber ISPs	Low
36	NET	Cloudflare	Network Edge / CDN	Low
37	AKAM	Akamai Technologies	Network Edge / CDN	Low
38	FSLY	Fastly	Network Edge / CDN	Low
39	SNPS	Synopsys	Network IP / DPI	Low
40	VMW	VMware (Broadcom)	SDN / Network Virtualization	Medium
41	INFN	Infinera (Nokia)	Optical Networking Systems	High
42	FRKWF	Furukawa Electric (OTC ADR)	Optical Fiber & Cable	High
43	IIVI	II-VI (now Coherent)	Optical Components	High
44	CEVA	CEVA Inc	Wireless Infrastructure IP	Medium
45	CLFD	Clearfield	Fiber Connectors & Management	High
46	MTSI	MACOM Technology Solutions	Optical Components / RF	High
47	SLAB	Silicon Laboratories	Wireless / IoT Connectivity	Medium
48	LPTH	LightPath Technologies	Optical Components / Infrared	High

Methodology & Caveats

Verdict criteria: "High" = the company's core revenue is directly accelerated by AGI-scale compute buildout with defensible positioning (supply constraints, IP moats, market share). "Medium" = clear AGI tailwind but the company also has large non-AI segments or faces competitive dilution. "Low" = AI impact is indirect, marginal, or potentially negative.
Overlap: Some companies appear in multiple sectors (e.g., CIEN in transceivers, DWDM, and undersea; AVGO in interconnect and switches). Their master-table verdict reflects the aggregate across all relevant sectors.
ADRs: NOK, ERIC, FRKWF are foreign-domiciled but US-listed. INFN was acquired by Nokia in 2024 and may have limited post-merger liquidity.
Not included: SpaceX/Starlink (private), SubCom (private), Huawei (sanctioned/non-US). These are important players but uninvestable for US public equity portfolios.
Data vintage: Analysis based on publicly available information through May 2026. Company fundamentals, competitive positions, and supply dynamics can shift rapidly.