Networking Watch: Enterprise Networking Growth, World's First AI-RAN Call, Toshiba's QSN Platform & Nokia AI
Enterprise Networking Grows 15% While Fundamentals Diverge, Nokia and NVIDIA Move AI-RAN Into Live Networks, Toshiba Launches a Quantum-Safe Platform, and the 1.6T Optical Transition Begins
The Networking Market Is Growing — But the Reasons Are Diverging Fast
Enterprise networking revenue grew 15% year over year in the first half of 2026, with all five major segments posting double-digit gains, according to Dell'Oro Group. On the surface, that looks like a healthy, uniform recovery. It is not. The headline figures conceal increasingly different market dynamics — some segments are growing on genuine shipment volume, others on price increases caused by component shortages. In campus switching, port shipments fell 5% in Q2 2026 while average selling prices surged 19%. That is revenue growth driven by price, not by demand. Understanding which is which has become essential for enterprise buyers planning 2027 refreshes.
The same divergence is visible one layer up, in telecom infrastructure. Nokia's Network Infrastructure net sales grew 15% at constant currency in Q3 2026, led by IP Networks up 43% and Optical Networks up 18%, with AI and Cloud net sales growing 123%. On October 21, Nokia and NVIDIA completed the world's first AI-RAN call on a live commercial network with T-Mobile in Bellevue, Washington, running over Nokia Massive MIMO radios and NVIDIA Aerial RAN Computer infrastructure. AI is no longer only a data-center workload. It is moving into the radio.
Two further developments widen the frame. Toshiba launched its Quantum-Safe Networking (QSN) Platform on October 21, combining post-quantum cryptography, quantum key distribution and quantum random number generation into a single architecture that integrates with existing fiber networks rather than requiring a rebuild. And in the AI fabric itself, the industry is crossing into the 102.4 Tbps switching generation, with 1.6T optics positioned to replace 800G as the 2027 mainstream. The through-line: networking is being reshaped simultaneously by AI demand, component scarcity and a cryptographic migration — and the vendors that win will be the ones that solve all three at once.
Networking at a Glance
| Development | Vendor / Source | Category | Status |
|---|
Signal summary: Three of these seven developments are shipping or complete. Three are previews, trials, or early adoption. One is pure market data. That distribution matters: networking is being reshaped by genuine deployed technology — live AI-RAN calls, shipping 102.4T switches — while several adjacent initiatives remain announcements awaiting validation.
The Stories Shaping Networking
Enterprise Networking Grew 15% in 1H26 — But Some Segments Grew on Price, Not Demand
What changed: Dell'Oro Group reported on October 26 that worldwide enterprise networking revenue grew 15% year over year in 1H26, with all five major segments — WLAN, Network Security, Campus Switching, Branch Routing and Enterprise Data Center Switching — posting double-digit gains (Dell'Oro). Growth accelerated further in Q2.
Why it matters: The headline number is real but misleading if read as uniform demand strength. The segments are growing for different reasons, and those reasons will not persist on the same timetable. Segment-by-segment: WLAN led growth on higher access point shipments as Wi-Fi 7 adoption accelerated, with enterprises ordering ahead of anticipated price increases. Wi-Fi 7 now accounts for most indoor access point revenue. Network Security contributed the most incremental revenue. Campus Switching grew at a double-digit rate — but port shipments were nearly flat, and in Q2 2026 shipments fell 5% while ASPs surged 19%. That is price-driven growth, not volume-driven growth.
The component shortage: DDR4 memory shortages are the primary source of pricing pressure, because AI infrastructure buildouts have tightened component supply. Enterprise IT vendors have raised prices, redesigned products and sought alternative sources. Because enterprise equipment is often quoted well before delivery, these increases take time to appear in reported revenue — meaning more price-driven growth is likely still flowing through.
Who is affected: Enterprise buyers planning campus refresh cycles face a materially different calculation than they did in 2023. Higher switch prices are squeezing IT budgets and making some refresh projects harder to justify, even with a large installed base of end-of-support equipment in the field. Smaller customers have pulled back on enterprise data center switching purchases while large organizations continue to invest.
What comes next: Dell'Oro expects DDR4 shortages to push WLAN equipment prices higher in 2H26. The central question for 2027 is how much of the current expansion reflects durable investment demand and how much reflects pricing and purchase timing.
The CODEW Angle: This is the most important data point in the edition because it reframes everything else. If a meaningful portion of networking revenue growth is price-driven rather than demand-driven, then vendor earnings in 2027 may not track shipment volumes — and buyers who delay refreshes are, paradoxically, helping to keep prices elevated by holding back the volume that would justify capacity expansion.
The World's First AI-RAN Call Ran on a Live Commercial Network. That Changes the Timeline.
What changed: On October 21, Nokia, T-Mobile and NVIDIA completed the world's first AI-RAN call at a commercial cell site in Bellevue, Washington, over T-Mobile's live production network using Nokia Massive MIMO radios, Nokia 5G anyRAN software and infrastructure built on the NVIDIA Aerial RAN Computer platform. The site was connected to a commercial 5G core and commercial user equipment.
Why it matters: The distinction between a lab demonstration and a live-network call is the entire story. AI-RAN has been discussed as a 6G concept for years. Running it on a commercial cell site with real spectrum and production infrastructure moves it from research to deployment planning. Nokia describes it as the first software-defined RAN call using accelerated computing — a claim that, if accurate, means the RAN is now at parity with cloud and software environments, benefiting from continuous CI/CD innovation.
The operator evidence: The milestone is not isolated. T-Mobile completed the live call. BT is preparing a live-network trial in Birmingham for 2027. Elisa ran voice, mobile data and video analytics simultaneously on the same AI-RAN platform in Helsinki, with distributed deployment and commercial service evaluation expected in 2027. Deutsche Telekom validated the commercial AI-RAN platform in Bonn and Oulu labs, and its AI-based channel estimation work showed 10–15% typical uplink throughput improvements in initial lab measurements, with greater gains under more complex radio conditions. Nine EMEA operators are building or expanding NVIDIA-powered infrastructure.
What AI-RAN actually does: Nokia and NVIDIA are applying AI within the 5G signal-processing chain — channel estimation, beamforming, multi-user MIMO pairing and scheduling. That is a materially different proposition from running an AI application on a server next to a cell tower. It means AI is inside the radio function itself, improving spectral efficiency. Nokia's stated ambition is 2x spectral efficiency. The deployment reality: Commercial AI-RAN services at scale are expected in 2027, not now. The live call validates the architecture; it does not validate the business case or the operational economics at thousands of sites.
The CODEW Angle: AI-RAN is the clearest example in networking of AI moving from the data center into carrier infrastructure — and it reframes the competitive map. Nokia gains a differentiation story against Ericsson and Samsung in RAN. NVIDIA extends its accelerated computing footprint from hyperscale data centers into radio sites, which is a substantially larger installed base. And operators gain a path to improve 5G economics without waiting for 6G spectrum.
Toshiba's QSN Platform Bets That Quantum-Safe Migration Happens Incrementally, Not at Once
What changed: Toshiba launched its Quantum-Safe Networking (QSN) Platform on October 21 — a carrier-grade architecture combining Post-Quantum Cryptography (PQC), Quantum Key Distribution (QKD) and Quantum Random Number Generation (QRNG) into a single centrally managed system. The critical design decision is that organizations do not need to rebuild their networks to adopt it; the platform integrates with existing fiber infrastructure and supports deployments from a single dedicated connection up to large metro, hub-and-spoke, ring, mesh and long-distance networks.
Why it matters: Quantum-safe migration has been framed as an all-or-nothing infrastructure project, which has made it easy for enterprises to defer. Toshiba is explicitly attacking that framing. Its transceivers combine transmitters and receivers in a single unit, and its hubs use optical switching to connect multiple sites with less equipment than conventional point-to-point QKD deployments — reducing footprint and cost per site. That lowers the threshold for starting with the most critical data paths and expanding later.
The threat model: Toshiba frames the risk around "harvest now, decrypt later" attacks, in which adversaries capture encrypted data today and retain it for future decryption. That threat is already present, which is why governments and regulators are establishing migration timelines for critical systems. Toshiba's argument is that QKD and QRNG are not susceptible to AI-driven attacks used to find vulnerabilities in algorithmic cryptography — a distinction that matters as AI-accelerated vulnerability discovery becomes more common.
The interoperability dimension: Toshiba has demonstrated the platform integrated with HPE Networking SRX next-generation firewalls via SpeQtral's key orchestration software, showcased at GovWare 2026 in Singapore. That matters because quantum-safe networking only works if it spans multi-vendor infrastructure.
The deployment reality: Toshiba is launching the platform, not announcing broad customer deployments. Targeted trials and commercial deployments of quantum-safe technology have already demonstrated the technology works for critical connections; the open question is how quickly it extends across more sites and network types.
The CODEW Angle: Security is becoming a network architecture decision rather than a software overlay, and Toshiba is positioning quantum safe as a scalable network service rather than a bespoke project. The strategic risk is timing: if enterprises conclude the quantum threat is far enough away, adoption stalls regardless of how elegant the migration path is. The commercial opportunity depends on regulators setting deadlines that force the issue.
Nokia's AI and Cloud Sales More Than Doubled. The Bottleneck Is Now Supply, Not Demand.
What changed: Nokia reported Q3 2026 results on October 22 with net sales up 9% year over year at constant currency and Network Infrastructure up 15%, led by IP Networks growing 43% and Optical Networks growing 18%. Group net sales to AI and Cloud customers grew 123%, with order intake up 118%. Nokia secured €1.5 billion in new AI and Cloud orders in the quarter.
Why it matters: The optical transport layer is where AI infrastructure spending is now showing up. That is a meaningful shift in the AI capex story. For most of the buildout, the spending has been concentrated in accelerators, servers and data-center switches. Nokia's results indicate that the connectivity layer between data centers — long-haul optical, IP routing, metro transport — is now growing at a rate comparable to the compute layer.
The Infinera factor: Nokia is on track to complete the Infinera integration by the end of 2026, one year ahead of plan, with €200 million of net synergies. Optical Networks is the business that houses Infinera, so the 18% growth there reflects both organic demand and the combined portfolio. Nokia expects its combined Optical and IP Networks businesses to grow 25–30% in 2027.
The margin caveat: Network infrastructure gross margin declined 150 basis points to 41.2% due to product mix and ramp-up costs for new AI and cloud products. That is an important detail — growth in this segment is currently coming at a cost. Ramp-up expenses for new products compress margins before volume efficiencies arrive.
The supply constraint: Nokia noted component supply constraints alongside the demand growth. That connects directly to the Dell'Oro pricing story: AI infrastructure buildouts are competing for the same memory and component supply as enterprise networking gear, driving up prices across both markets.
The hyperscaler relationship: Nokia's partnership with Google in support of Google's €13 billion investment in Finland is cited as an example of how optical and IP networks provide the connectivity AI infrastructure requires. That kind of anchor customer relationship is what makes the 25–30% growth target credible.
The CODEW Angle: AI capex is broadening beyond the data center. Nokia's results are the clearest evidence yet that optical transport is becoming a growth market in its own right, not just a supporting layer. The risk is margin: if ramp-up costs persist as volumes scale, revenue growth will not translate proportionally into profit.
The 102.4 Tbps Era Arrives — and Optics Is Becoming the Binding Constraint
What changed: The AI networking fabric crossed into the 102.4 Tbps switching generation in production, with Broadcom's Tomahawk 6 silicon underpinning systems from Arista Networks, Celestica and Micas Networks. Celestica's DS6005 is a liquid-cooled 102.4 Tbps Ethernet switch based on the Tomahawk 6–Davisson co-packaged optics platform, offering 64 × 1.6 Tbps ports in a 1.5OU form factor (Dell'Oro). On the silicon side, Cisco Systems reported initial customer shipments of G300-powered 102.4T systems and introduced the 51.2T Silicon One G302 with 200G SerDes and 1.6T connectivity for top-of-rack applications.
Why it matters: As AI clusters scale, the network that connects accelerators determines how efficiently they can work together. The industry is transitioning from 800G to 1.6T optics, with 1.6T positioned to replace 800G as the mainstream in 2027. That is not a minor generational bump; it changes the economics and physical design of AI data centers.
The co-packaged optics milestone: CPO has moved from roadmap to early commercial deployment. NVIDIA moved its Spectrum-X Ethernet Photonics switches into production with early customer adoption. This matters because CPO places the optical engine directly alongside the switch silicon, dramatically reducing power consumption per bit compared with pluggable optics.
The pluggable counterattack: Pluggables are not standing still. NVIDIA and Cisco introduced plans for an Ultra-Dense Small Form Factor Pluggable (USFP) — a proposed 400 Gb/s-per-lane design with a path to 12.8 Tbps of aggregate module capacity, targeting roughly four times the bandwidth density and 1.3 times the power efficiency of 200 Gb/s-per-lane designs. The competitive dynamic is not CPO versus pluggables; it is a portfolio of optical approaches competing on density, power, reach, reliability and serviceability.
The rack-density outlook: Arista identified a path from 1.6 Pb/s with current OSFP-based designs to 6.5 Pb/s per rack using open CPO/NPO and XPO approaches. That is an architectural ambition, not evidence of volume deployment — but it shows where the physics is heading. The deployment reality: 102.4T systems are shipping. 1.6T optics are entering early adoption with mainstream expected in 2027. CPO is in early production. But early commercialization should not be confused with broad adoption: customer qualification, serviceability, production yields and meaningful shipment volumes remain the metrics to watch.
The CODEW Angle: This is the follow-through on the open Ethernet and optics story The CODEW covered on October 13. The strategic insight is that optics is no longer a component decision — it is becoming a rack-level architectural decision that determines how much compute can be packed into a data center. Vendors that control the optical integration path will influence AI data-center design in a way that switch silicon alone no longer does.
Also in Networking
Extreme Networks previews Exchange Zones: Extreme Networks has previewed an on-premises AI infrastructure for networking that pools idle compute capacity in access points and switches to run AI workloads locally at branch and edge sites. Planned for general availability in the first half of calendar 2027, the capability is designed for data sovereignty, latency reduction and operation during cloud outages. This is a preview, not a shipping product — the architecture is notable but unvalidated.
Telstra and Ericsson expand autonomous network collaboration: Announced October 20, the pair are expanding their collaboration to accelerate the shift toward autonomous networks and will co-lead a new industry working group focused on the knowledge, trust and governance foundations needed for network autonomy. The emphasis on a common knowledge taxonomy and governance is the substantive part — autonomous networks require shared semantic understanding before automation can be trusted.
DE-CIX launches AI-agent network management: DE-CIX released an open-source Model Context Protocol server that lets customers use AI agents to discover, order and manage connectivity services using natural-language instructions. It is an early but concrete example of agentic AI applied to network operations rather than to end-user applications.
Micas Networks ships 102.4T open Ethernet platforms: Micas introduced three Tomahawk 6-based systems: a 4RU air-cooled switch with 128 ports at 800G, a 2RU liquid-cooled switch with 64 ports at 1.6T, and a 1RU co-packaged optics switch. The three-SKU approach illustrates how vendors are hedging across cooling and optics architectures rather than committing to one.
Competitive Implications: Three Architectures, Three Bets
Read together, this week's developments reveal three distinct architectural bets being made simultaneously across the networking stack. They are not mutually exclusive, but they compete for the same capital and the same engineering attention.
| Architecture | The Bet | Who Is Making It |
|---|---|---|
| AI inside the RAN | Spectral efficiency gains justify accelerated computing at cell sites | Nokia, NVIDIA, T-Mobile, BT, Deutsche Telekom, Elisa |
| Quantum-safe by architecture | Security becomes a network layer, migrated incrementally | Toshiba, HPE, SpeQtral |
| Optical density over everything | 1.6T and CPO determine how much compute fits in a rack | Broadcom, NVIDIA, Cisco, Arista, Celestica, Micas |
| AI at the edge, on-prem | Data sovereignty requires local inference capacity | Extreme Networks |
| Autonomous operations | Shared knowledge and governance precede automation | Telstra, Ericsson, DE-CIX |
Where the bets intersect. The optical density bet and the AI-RAN bet are connected more than they appear. AI-RAN requires moving large volumes of data between radio sites and centralized compute; the same optical transport capacity Nokia is selling into AI and Cloud customers is what makes distributed RAN intelligence practical. Toshiba's quantum-safe platform, meanwhile, is a prerequisite for both — every one of these architectures expands the encrypted traffic surface that must eventually be protected against quantum decryption.
Where the bets conflict. Extreme Networks' on-premises edge AI and Nokia's AI-RAN both assume compute moves toward the network edge. But they are competing for the same operator and enterprise capital, and both depend on component supply that is already constrained. If DDR4 and optics component shortages persist into 2027, some of these architectures will be delayed not by technical failure but by inability to source parts.
What to Watch Next
- Dell'Oro's 2H26 enterprise networking data. The key test is whether campus switching port shipments recover. If ASPs keep rising while shipments stay flat, the market is pricing-driven and 2027 refresh plans will be harder to justify.
- BT's Birmingham AI-RAN trial in 2027. The first European live-network trial will show whether the Bellevue result generalizes across spectrum bands and operator environments.
- Commercial AI-RAN service launches in 2027. Elisa's distributed deployment and evaluation of commercial AI-RAN services is the milestone that moves this from validation to revenue.
- Toshiba QSN customer deployments. The platform is launched, not deployed at scale. Watch for named operator or data-center customers and for regulator-driven migration deadlines that force adoption.
- 1.6T optics qualification and volume. The transition from 800G to 1.6T mainstream in 2027 depends on customer qualification and production yields, not on switch silicon availability.
- CPO serviceability and field replacement. Co-packaged optics is shipping in early production, but field-replaceable external lasers and serviceability models will determine whether operators accept it at scale.
- Nokia's optical and IP growth against margin recovery. Nokia targets 25–30% growth in combined Optical and IP Networks in 2027. Watch whether gross margin recovers from the 41.2% level as AI and cloud product ramp-up costs normalize.
The larger industry pattern connecting these stories is that networking is being pulled in three directions at once — toward AI compute density, toward distributed intelligence at the edge and radio, and toward cryptographic migration — and the same physical supply chain constrains all three.
That constraint is the most underappreciated story in networking right now. Dell'Oro attributes campus switch and WLAN price inflation directly to DDR4 shortages caused by AI infrastructure buildouts. Nokia cites component supply constraints alongside 123% AI and Cloud growth. Extreme's edge AI architecture depends on pooling compute that itself requires components in short supply. The 1.6T optics transition requires laser and photonic component capacity that is being built out but is not yet abundant. In other words, the AI boom is simultaneously creating demand for more networking and limiting the industry's ability to supply it.
The strategic consequence is that networking vendors are being forced to choose where to allocate constrained supply. A component that goes into a 1.6T optical module for a hyperscale AI cluster cannot go into a campus switch for an enterprise refresh. The vendors with the strongest hyperscaler relationships — Nokia with Google and its €1.5 billion in AI and Cloud orders, Broadcom and NVIDIA with their AI fabric platforms — will get preferential allocation. Enterprise buyers competing for the same components will pay more and wait longer. That dynamic is already visible in the 19% ASP increase on campus switching.
What would change the assessment. If DDR4 and optics component supply loosens in 2027, price-driven revenue growth reverts to volume-driven growth and the enterprise refresh cycle resumes at normal economics. If AI-RAN commercial services launch on schedule in 2027 with measurable spectral efficiency gains, carrier infrastructure becomes a genuine second growth market for accelerated networking. And if regulators set firm quantum-safe migration deadlines, Toshiba's incremental adoption model becomes a template rather than a niche approach.
Conversely, if supply remains constrained and AI capex keeps absorbing component capacity, enterprise networking becomes a market where revenue grows while customer value stagnates — more money for less equipment. That is the scenario enterprise buyers should plan against, because it is the one currently visible in the data.
The CODEW Research Flywheel
Networking Watch tracks the infrastructure that connects compute, users and machines — and how AI is reshaping it. It sits alongside the other Watch series and feeds into the deeper CODEW research architecture:
| Product | Core Question |
|---|---|
| Networking Watch | Enterprise networking, AI data-center connectivity, optical infrastructure and telecom |
| Semiconductor Watch | Chip design, memory, packaging and the semiconductor supply chain |
| AI Infrastructure Watch | Data centers, compute, power and the physical infrastructure of AI |
| Cybersecurity Watch | Network security, identity and critical infrastructure protection |
| Hardware Watch | Switches, optics, edge devices and computing hardware |
| Enterprise Software Watch | Network automation, AIOps and enterprise platform strategy |
Networking Watch → Enterprise networking → AI data-center fabric → Optical transport → Telecom AI-RAN → Quantum-safe security → Network automation → Semiconductor Watch / AI Infrastructure / Cybersecurity Watch
The CODEW Stat
19% — the increase in campus switching average selling prices in Q2 2026, against a 5% decline in port shipments. It is the single most revealing number in this edition because it quantifies the divergence at the heart of the networking market. Revenue grew at a double-digit rate, but the equipment being sold did not grow with it. DDR4 memory shortages driven by AI infrastructure buildouts are the cause, and those shortages are expected to push WLAN prices higher in the second half of 2026 as well. For enterprise buyers, the implication is uncomfortable: the networking refresh cycle is becoming more expensive at exactly the moment end-of-support equipment volumes make it most urgent. For vendors, it means growth that looks healthy on an income statement may not reflect the underlying demand for their products.
Sources: Dell'Oro Group (Enterprise Networking Market Growth and 2027 Outlook; OCP 2026 optics analysis); Nokia newsroom, Nokia Q3 2026 Interim Report, Nokia blog; NVIDIA Newsroom; Toshiba Corporation news release; Extreme Networks; DE-CIX; Telstra and Ericsson. Independent reporting: Computer Weekly, IT Brief UK, Network World, StorageReview, Business Standard, Benzinga, TrendForce, Decision Insights. Coverage window: October 15–26, 2026. Developments announced before October 27 are presented as continuing developments rather than new October 27 news. The available reporting does not establish five newly announced developments specifically on October 27; this edition uses verified recent leads.
All factual claims regarding market data, product availability, financial results and corporate announcements are drawn from company disclosures and contemporaneous reporting. Editorial analysis is clearly distinguished from reported facts throughout. Vendor-reported performance claims — including Nokia's AI-RAN spectral efficiency ambition, Deutsche Telekom's 10–15% uplink throughput improvement, and Nokia's 25–30% growth target — are labeled as company-reported and have not been independently verified. Products described as preview, trial or planned are not generally available. Early commercialization of co-packaged optics should not be confused with broad adoption. Arista's 6.5 Pb/s per rack figure is an architectural density outlook, not evidence of volume deployment. This series does not provide investment advice.
Reviewed by Erwin Castro
on
Tuesday, October 27, 2026
Rating:
No comments: