Fixed Wireless Archives

How Cambium Networks, GFiber, and Quext delivered a first-of-its-kind X7-35X Wi-Fi 7 MDU deployment

Some projects arrive and you immediately recognize them for what they are: not just a deployment, but a defining moment. When the opportunity to co-architect GFiber’s first managed MDU Wi-Fi deployment came across my desk, I knew within the first technical briefing that this would be one of those projects: complex, ambitious, and without a precise precedent at scale, pushing the boundaries of what is possible in enterprise wireless networking.

I am proud and fortunate to have served as the key technical interface between GFiber, Quext, and Cambium Networks throughout this engagement. The result is a live, production deployment at Pacifica Place, an 896-unit luxury apartment complex in Irvine, California, that represents a new benchmark for managed MDU Wi-Fi in the United States, and a testament to what strong cross-organizational collaboration and disciplined engineering can accomplish.

As covered by Fierce Network in their recent article “GFiber Gets into Bulk-Billing, Managed Wi-Fi Business for MDUs,” this deployment is already attracting industry-wide attention. This blog goes behind the headlines to share the technical architecture, the real engineering challenges we overcame, and the collaborative spirit that made it possible.

The Deployment: Pacifica Place at Irvine Spectrum

Pacifica Place is a newly constructed, 896-unit luxury residential community located within the Irvine Spectrum corridor in Irvine, California. GFiber, in partnership with The Irvine Company, selected this property as the launch site for its managed MDU Wi-Fi offering, a product that delivers dedicated 3 Gbps symmetrical fiber broadband to every individual unit rather than relying on a single shared pipe distributed across the property.

Each unit is served by a home-run fiber connection terminating in a Google Optical Network Terminal (ONT). A Cambium X7-35X Wi-Fi 7 access point, ceiling-mounted in every unit, provides the in-unit wireless layer. Quext, a smart-apartment technology provider, integrates the network with the property’s IoT ecosystem, enabling residents to manage thermostats, lighting, door locks, and access control systems directly over their unit’s Wi-Fi network.

Interior of a Pacifica Place apartment unit, with a Cambium X7-35X Wi-Fi 7 access point ceiling-mounted above the kitchen and living area

The scope extended beyond individual units. GFiber, Quext, and Cambium also blanketed all community spaces, including the pool area, parking garage, corridors, and amenity zones, with managed Wi-Fi, ensuring residents remain seamlessly connected throughout the property.

The Pacifica Place fitness center, one of the community amenity spaces covered by managed Wi-Fi, with the pool deck visible through floor-to-ceiling windows

The Architecture: Solving for Scale and Security from Day One

From the outset, this deployment presented a set of architectural requirements that, in combination, had no off-the-shelf answer. The engineering team, spanning Cambium, GFiber, and Quext, had to design for three simultaneous imperatives: per-unit network isolation, IoT device integration at the apartment edge, and enterprise-grade security compliance for the 6 GHz band.

Challenge 1: Per-Unit Micro-Segmentation and Re-Architecting the Tunnel Model

A core requirement from GFiber was true per-tenant wireless isolation: ensuring that every resident’s connected devices are logically separated from their neighbors’, even across a shared access infrastructure. To meet it, a conventional tunneling approach would not suffice. The Cambium engineering team had to rethink and re-architect the traditional tunnel model to fit the unique demands of this deployment.

The result is true micro-segmentation at the wireless edge: each resident’s devices operate within their own isolated network segment, with no lateral visibility to other tenants, regardless of the underlying access infrastructure.

The architecture we built is robust, scalable, and purpose-fitted to GFiber’s deployment requirements, and it does not depend on the physical access topology beneath it.

Challenge 2: WPA3-ePSK at Scale Across an 896-Unit Property

Wi-Fi 6 GHz operation mandates WPA3 as a security requirement, with no option to fall back to WPA2 on 6 GHz radios. For a large-scale MDU deployment where every unit has its own provisioned access point and potentially dozens of connected devices, this creates a non-trivial operational challenge: how do you provision, manage, and scale WPA3-ePSK (Enterprise Pre-Shared Key) credentials across hundreds of units without creating an administrative nightmare?

Our solution was to architect the ePSK management workflow entirely within Cambium’s cnMaestro cloud management platform. cnMaestro provides a centralized provisioning pipeline that allows per-unit ePSK credentials to be generated or configured, assigned, and managed at scale. Each unit’s AP is provisioned with a unique ePSK profile, ensuring that even on a shared SSID, each tenant’s connection is cryptographically isolated.

Challenge 3: Solving the SAE-Only Client Problem

One of the most technically nuanced challenges in this deployment surfaced in how wireless clients behave when associating to 5 GHz or 6 GHz BSSIDs on a WPA3-ePSK network. The reality of a large-scale MDU environment is that the wireless client population is highly heterogeneous: smartphones, laptops, smart TVs, IoT sensors, and legacy consumer electronics all coexisting on the same network. And not all of them speak the same security dialect.

The specific problem: a subset of wireless clients, particularly those with specific WPA3 implementations, send only SAE (Simultaneous Authentication of Equals) authentication frames when attempting to associate. On a WPA3-ePSK network, this causes the authentication to fail outright. The client attempts to connect, the AP cannot reconcile the SAE-only request with the ePSK policy, and the association is rejected. For a resident whose device falls into this category, the result is simply: no connection.

This was not an edge case we could dismiss. Across 896 units with potentially thousands of diverse devices, SAE-only clients were an unavoidable reality. The Cambium engineering team developed a solution that gracefully handles SAE-only client authentication on a WPA3-ePSK network. Clients that send only SAE authentication frames are now able to connect successfully, without compromising the security posture or micro-segmentation integrity of the overall deployment. I’m proud that we identified, engineered, and validated this solution in alignment with the GFiber and Quext deployment timeline, without a single day of delay.

The Partnership: What Great Collaboration Looks Like

Technical complexity is, in some ways, the easier part. The harder part, and the part that truly determines whether a project of this scale succeeds, is the quality of collaboration between the organizations involved. I am genuinely proud to say that the partnership between Cambium, GFiber, and Quext on this project set a standard I will carry into every future engagement.

Working alongside the GFiber team was, simply put, a pleasure. Their engineers are sharp, their expectations are high, and their commitment to building a product that genuinely serves residents is evident in every technical conversation. There is something energizing about working with a team that asks hard questions and won’t settle for “good enough.” Every architecture review, every edge case discussion, every whiteboard session pushed the solution to be better. That intellectual rigor from the GFiber side made our engineering work stronger.

Quext brought far more than smart-apartment integration expertise to this engagement; they brought genuine, hands-on capability in wireless deployment and ePSK provisioning within MDU environments. Their team demonstrated a deep understanding of how WPA3-ePSK behaves at the residential edge: the nuances of per-unit credential management, the real-world variability in how residents’ devices onboard to an ePSK-provisioned network, and the practical considerations that only surface when you have deployed at scale across multi-family properties. That operational knowledge was a force multiplier for the overall team. Their understanding of how residents actually interact with connected home technology, including the non-obvious failure modes that arise in production, was invaluable in ensuring that the wireless and IoT integration was not just technically correct, but genuinely seamless from the resident’s perspective. Quext’s deployment experience in MDU wireless is a real differentiator, and it showed throughout this project.

Serving as the central technical interface across all three organizations was one of the most professionally rewarding experiences of my career. It required translating deeply technical wireless networking concepts into product decisions, aligning timelines across teams with different internal pressures, and maintaining clarity of purpose even when the engineering path was uncertain. I am grateful to have had the opportunity to play that role.

Why This Matters for the MDU Industry

Pacifica Place is not just one building. It is a proof point: a repeatable, scalable architecture for managed MDU Wi-Fi.

The architecture we built at Pacifica Place, including the tunnel-based micro-segmentation model, the scalable WPA3-ePSK provisioning pipeline, and the SAE client compatibility solution, is designed to be repeatable and scalable.

For the MDU Wi-Fi industry more broadly, this deployment signals several important directions: the move from best-effort shared Wi-Fi to per-unit dedicated fiber with managed wireless; the operationalization of WPA3 and 6 GHz at residential scale; and the growing importance of IoT-aware network design in multi-family residential environments.

Cambium Networks’ X7-35X Wi-Fi 7 access points and cnMaestro management platform were purpose-built for exactly this kind of deployment: high-density, high-security, and centrally managed at scale. I believe this is a model that will define how managed MDU Wi-Fi is delivered across the country.

Closing Thoughts

I began this project knowing it would be complex. What I didn’t fully anticipate was how much I would learn from the teams I worked alongside, and how much satisfaction would come from solving problems that, at the outset, had no obvious answers. The WPA3-ePSK SAE client challenge alone pushed our team to innovate in ways that I believe will benefit future deployments far beyond Irvine.

To the GFiber and Quext teams: it was an honor and a genuine pleasure. I look forward to what we build next.

For further reading on the commercial context of this deployment, I recommend the Fierce Network article: “GFiber Gets into Bulk-Billing, Managed Wi-Fi Business for MDUs” (Fierce Network, June 22, 2026).

Key Takeaways

First managed MDU build for GFiber — A live, production Wi-Fi 7 deployment across 896 units at Pacifica Place in Irvine, California.
Per-unit isolation — A re-architected tunnel model delivers true micro-segmentation at the wireless edge, with no lateral visibility between tenants.
WPA3-ePSK at scale — cnMaestro centralizes per-unit credential provisioning so every unit receives a unique, cryptographically isolated profile.
SAE-only client fix — A purpose-built solution lets SAE-only WPA3 clients connect without weakening security or segmentation.
Repeatable model — The architecture is designed to scale beyond one property to managed MDU Wi-Fi nationwide.

Building managed Wi-Fi for multi-dwelling units?

See how the Cambium X7-35X Wi-Fi 7 access point and cnMaestro deliver secure, per-unit connectivity at scale.

Explore the X7-35X

About the Author

Azif Abdul Salam is a Director of Solutions Engineering at Cambium Networks, specializing in cnMaestro, Intent-Based Networking (IBN), and MDU/Hospitality Wi-Fi solutions. He holds multiple Wi-Fi patents and works across technical architecture, customer deployments, product design, and cross-functional collaboration with engineering and field teams.

L’infrastruttura che alimenta l’AI — data center, reti, energia — vale già 400 miliardi di dollari l’anno e crescerà fino a un trilione entro il 2028; ciò anche perché l’AI agentica richiede reti riprogettate da zero: latenza sub-100ms, identità crittografiche per ogni agente, policy unificate tra cloud e ambienti on-premise. Nel frattempo, la disponibilità di energia elettrica ha superato il capitale come principale collo di bottiglia allo sviluppo. Europa e USA si confrontano con reti elettriche al limite: chi pianifica prima vince la corsa all’AI.

C’è qualcosa di paradossale nel modo in cui il mondo parla di intelligenza artificiale. I titoli dei giornali inseguono i modelli linguistici, i chatbot, le applicazioni consumer. Ciò che rimane fuori campo – invisibile e necessario come l’aria – è l’infrastruttura che rende tutto possibile: i data center, le reti, la corrente elettrica, i protocolli di connessione tra agenti autonomi. È lì che si gioca la partita vera. Ed è lì che stanno confluendo, silenziosamente, centinaia di miliardi di dollari l’anno.

I numeri che ridisegnano il mondo

Nel 2026, la spesa globale per i data center AI ha raggiunto una soglia che pochi avrebbero immaginato solo tre anni fa: tra i 400 e i 450 miliardi di dollari di capital expenditure, con una proiezione che tocca il trilione entro il 2028. Non si tratta di una stima di mercato ottimistica: è la convergenza di dati provenienti da Goldman Sachs, Deloitte, McKinsey e da chi, ogni giorno, prende decisioni concrete su dove costruire, quanta potenza installare e come finanziare l’espansione. [1]

Dietro questa cifra c’è un’equazione semplice e brutale: la domanda di calcolo cresce quattro o cinque volte più velocemente di quanto l’efficienza dei chip riesca a compensare. Ogni nuova generazione di server è più efficiente per unità di calcolo, ma consuma più potenza per singolo rack. I server Vera Rubin di NVIDIA, ad esempio, offrono il 16% in più di velocità di calcolo per kilowatt rispetto alla generazione Blackwell precedente – ma la loro potenza massima per server è aumentata del 68%. L’efficienza non basta: l’appetito dell’AI è più vorace. [2][1]

A questo si aggiunge la transizione dal training all’inferenza, che molti avevano interpretato come un segnale di rallentamento. Deloitte smentisce questa lettura con dati precisi: l’inferenza nel 2026 rappresenta circa i due terzi di tutto il calcolo AI (contro un terzo nel 2023), ma le tecniche di post-training scaling richiedono 30 volte più calcolo del training originale, mentre il test-time scaling – il cosiddetto “pensiero lungo” dei modelli – ne richiede oltre 100 volte. Il risultato è che i data center non diventano più piccoli: crescono, si densificano, si moltiplicano. [1]

L’infrastruttura invisibile che connette tutto

In questo contesto, parlare di “connettività avanzata” significa qualcosa di molto più ampio della semplice banda larga. Significa ripensare l’intera architettura di rete che permette a decine di migliaia di GPU di comunicare tra loro in millisecondi, che consente a workload distribuiti su datacenter multipli di funzionare come un unico sistema coerente, e – frontiera emergente – che abilita agenti AI autonomi a collaborare tra di loro in tempo reale senza compromettere sicurezza e governance.

McKinsey identifica la connettività avanzata come una delle tendenze tecnologiche di frontiera per il 2025-2026, sottolineando come la sua importanza sia amplificata proprio dall’esplosione dell’AI. “One of the most pronounced connectivity trends during the last year was a significant growth in data center connectivity demand driven by AI. Hyperscalers are looking to build or buy significantly more fiber than before”. Non è una tendenza secondaria: è una condizione abilitante per tutto il resto. [3]

A livello tecnico, i requisiti di rete per i data center AI di nuova generazione sono radicalmente diversi da quelli tradizionali. I cluster GPU per il training di modelli fondazionali richiedono comunicazioni inter-nodo a latenza bassissima e bandwidth elevatissima. Meta, ad esempio, ha sviluppato reti specializzate per i propri cluster GPU che utilizzano RDMA over Converged Ethernet version 2 (RoCEv2) come trasporto per le comunicazioni inter-nodo in AI training distribuito. Non è un dettaglio tecnico, bensì la colonna vertebrale che determina se un training run da settimane può completarsi o collassa al primo ritardo di rete. [3]

La fotonica al silicio sta emergendo come tecnologia abilitante per portare questa connettività a scala. Marvell ha dimostrato in laboratorio un motore 3D di fotonica al silicio capace di 6,4 terabit al secondo, con 32 canali da 200 gigabit ciascuno – un’integrazione che riduce il costo per bit e offre scalabilità modulare da 1,6 a 6,4 terabit e oltre. NVIDIA ha annunciato Spectrum-X Photonics, switch di rete con ottica co-impacchettata per scalare le AI factory fino a milioni di GPU, con prestazioni 10 volte superiori e 63 volte migliore integrità del segnale rispetto ai metodi tradizionali. [3]

L’AI agentica e la crisi delle reti tradizionali

Ma c’è una sfida più sottile e più urgente che emerge con forza nel 2026: le reti aziendali non sono state progettate per l’AI agentica. Questa è la tesi centrale di Equinix, che gestisce oltre 260 data center in 36 paesi e ha costruito la propria riflessione su questo tema partendo dall’osservazione diretta di migliaia di deployment aziendali. [4]

L’AI agentica – sistemi in cui molteplici agenti autonomi collaborano, prendono decisioni e agiscono in modo indipendente – non è semplicemente “più AI”. È un paradigma radicalmente diverso che richiede un’infrastruttura altrettanto diversa. McKinsey misura l’investimento in equity nell’AI agentica a 985 miliardi di dollari nel 2024, con una crescita più rapida di qualsiasi altra tendenza tecnologica. Deloitte rileva che quasi tre aziende su quattro (74%) pianificano di impiegare l’AI agentica entro due anni. [5][3]

Il problema è che queste organizzazioni stanno cercando di eseguire sistemi multi-agente distribuiti, su infrastrutture progettate per un’architettura completamente diversa: reti perimetrali, controlli di sicurezza frammentati, policy di identità non coordinate tra ambienti cloud multipli. Quando agenti diversi girano su cloud diversi o ambienti on-premises, i loro controlli di sicurezza, le policy di rete e i confini di fiducia diventano progressivamente frammentati e difficili da gestire. [4]

La risposta che sta emergendo – e che Equinix ha dettagliato in un documento tecnico pubblicato nel febbraio 2026 – è il concetto di “secure agent enclave”: un’architettura che taglia trasversalmente i silos infrastrutturali e crea un approccio unificato alla gestione e alla sicurezza dei workload agentici, indipendentemente da dove siano ospitati. [4]

Anatomia di un’enclave sicura per agenti AI

La struttura di un’enclave sicura per agenti AI rappresenta l’evoluzione più sofisticata del networking aziendale. Non si tratta più solo di connettere sistemi: si tratta di creare un “trust fabric” che attraversa ogni strato dell’infrastruttura – rete, dati, identità, policy.

I requisiti tecnici sono precisi e non negoziabili: gestione del ciclo di vita di ogni agente attraverso un control plane centralizzato; comunicazione zero-trust con autenticazione e cifratura end-to-end per ogni interazione agente-agente e agente-tool; latenza sub-100 millisecondi per mantenere il ragionamento in tempo reale; supporto per comunicazioni asincrone e streaming per workload a lungo termine; topologie mesh per resilienza e scalabilità orizzontale; e, in modo cruciale, identità decentralizzata e verificabile crittograficamente per ogni agente e risorsa. [4]

Standard come il Model Context Protocol (MCP) di Anthropic e l’Agent2Agent (A2A) Protocol di Google forniscono le basi di autenticazione e controllo degli accessi. Google ha introdotto A2A come standard aperto per la collaborazione sicura tra agenti AI di vendor diversi, supportato da oltre 50 partner, abilitando use case come candidate sourcing e supply chain coordination. Ma i protocolli da soli non bastano: richiedono un’infrastruttura di rete profondamente integrata che estenda fiducia, identità e enforcement delle policy end-to-end senza rallentare l’innovazione. [3][4]

Il punto più critico – e spesso sottovalutato – è che nell’AI agentica la separazione tradizionale tra policy di rete e policy dei dati semplicemente non esiste più. Ogni hop di rete coinvolge prompt sensibili, dati embedded o output di modelli. Il routing di rete deve essere controllato deterministicamente: anche se un agente ha una policy dati valida, può comunque fare data leakage se il traffico passa su un percorso di rete non controllato. Network controls, data governance e agent-level policies devono operare come un unico fabric unificato. [4]

Il collo di bottiglia che non è il denaro

Se la sfida della connettività per l’AI agentica è sofisticata, quella dell’energia per i data center è brutale e materiale. La principale limitazione allo sviluppo dei data center nel 2026 non è il capitale, ma la disponibilità di energia elettrica. [6]

I dati convergono su questo punto da tutte le direzioni. Ropes & Gray, in un report pubblicato il 21 maggio 2026 dopo aver partecipato alla conferenza IMN Data Centers Private Equity a New York, certifica che la disponibilità di energia – non il capitale – è il driver primario delle decisioni di investimento. Goldman Sachs prevede una crescita della domanda elettrica dei data center del 220% entro il 2030 rispetto ai livelli del 2023, con il 60% di questa crescita concentrata negli Stati Uniti. Il Lawrence Berkeley National Laboratory stima che la domanda dei data center americani passerà dai 176 TWh del 2023 a 325-580 TWh entro il 2028. [7][2][6]

Le interconnessioni alla rete elettrica richiedono fino a quattro anni per essere completate. In questo vuoto, le soluzioni BYOP (Bring Your Own Power) – generazione onsite, principalmente a gas naturale – stanno diventando non più un’eccezione, ma una strategia strutturale: oltre un terzo dei data center prevede di utilizzare il 100% di energia onsite entro il 2030, un aumento del 22% rispetto alle previsioni di sei mesi prima. [8][6]

La conseguenza geografica è immediata: il capitale si concentra nelle regioni con vantaggi energetici. Nel mercato americano, il Texas è proiettato a diventare il principale mercato di data center entro il 2028, con una crescita del 142% di market share, mentre mercati storici come California e Oregon perderanno rispettivamente il 50% e il 67% della loro quota relativa. In Europa, il quadro è simile: le FLAP-D (Frankfurt, London, Amsterdam, Paris, Dublin) stanno perdendo terreno a favore dei Paesi nordici e di mercati emergenti come Italia e Belgio, dove i tempi di connessione alla rete sono più brevi. [9][8]

Europa: la rete elettrica come infrastruttura strategica

Il caso europeo merita attenzione particolare, perché sintetizza in modo esemplare come la connessione – quella elettrica, prima ancora di quella digitale – sia diventata un fattore competitivo strategico.

La domanda elettrica dei data center in Europa è proiettata a crescere del 150% tra il 2024 e il 2035, passando da 96 TWh a 236 TWh. L’Unione Europea ha fissato l’obiettivo di triplicare la capacità dei data center nei prossimi cinque-sette anni attraverso l’AI Continent Action Plan e il programma InvestAI, che mobilita 200 miliardi di euro. Ma le previsioni di ICIS, IEA, IMF e McKinsey mostrano come l’UE sia destinata a raddoppiare la capacità, non a triplicarla, se non si interviene strutturalmente sulle reti. [9]

Il problema è la congestione delle reti elettriche nei mercati tradizionali. L’Irlanda, per esempio, ha imposto una moratoria di fatto sui nuovi data center a Dublino fino al 2028. Nei Paesi Bassi e in Germania, i gestori di rete hanno adottato misure simili. La reazione del mercato è stata immediata: il capitale si è spostato verso mercati con minor congestione. [9]

La lezione è chiara: le grid non sono più solo infrastruttura energetica. Sono infrastruttura competitiva. I paesi che investono oggi in pianificazione proattiva delle reti si posizionano per attrarre i flussi di investimento in AI degli anni a venire.

L’AI agentica come rivoluzione del networking aziendale

L’impatto dell’AI agentica sull’infrastruttura di rete va compreso nella sua dimensione sistemica. Non si tratta solo di un nuovo tipo di workload da gestire: è una trasformazione del ruolo stesso della rete aziendale.

Nelle architetture tradizionali, la rete è infrastruttura passiva: trasporta dati tra sistemi che hanno logica applicativa propria. Nell’architettura agentica, la rete diventa piattaforma attiva per la fiducia, le performance e la scalabilità. Non connette solo sistemi: abilita il trust, l’agilità e l’innovazione in ambienti AI-driven. Questo richiede reti che siano identity-aware (consapevoli dell’identità di ogni agente e risorsa), policy-driven (dunque che l’enforcement delle regole avvenga a livello di fabric, non di applicazione) e self-optimizing (che si adattino dinamicamente alle topologie agentiche in evoluzione). [4]

I protocolli emergenti – MCP, A2A, Agent Communication Protocol (ACP), Agent Network Protocol (ANP) – stanno definendo le basi di questo nuovo strato di rete. Ma la loro efficacia dipende dall’infrastruttura sottostante: interconnessione fisica e virtuale, colocation strategica, private connectivity per il controllo del routing. È per questo che Equinix, nel posizionare il proprio modello di business per l’era agentica, pone l’enfasi sulle interconnection hub come punto di convergenza di ecosistemi digitali: oltre 10.000 aziende nello stesso fabric, con on-ramp diretti a tutti i major cloud provider in 39 mercati globali. [4]

La posta in gioco

Rileggere i numeri nel loro insieme produce una vertigine. Goldman Sachs stima oltre 300 miliardi di dollari di revisioni al rialzo alle previsioni di capex dei grandi hyperscaler solo negli ultimi due mesi, con la spesa totale in capex e R&D da parte dei principali hyperscaler globali destinata a raddoppiare entro il 2029 rispetto al 2025. Il mercato mondiale dei data center europeo era valorizzato 47 miliardi di euro nel 2024 ed è proiettato a raggiungere 97 miliardi entro il 2030. Deloitte prevede che il mercato degli agenti AI autonomi potrebbe raggiungere 8,5 miliardi di dollari entro il 2026 e 35 miliardi entro il 2030, con un potenziale di crescita aggiuntiva del 15-30% se le aziende orchestreranno meglio i propri agenti. [2][1][9]

La posta in gioco non è solo economica. L’IMF stima che il boom dell’AI possa aumentare il tasso di crescita annuale del PIL globale di 0,5 punti percentuali tra il 2025 e il 2030 – ma solo se l’infrastruttura dei data center viene predisposta in tempo. La Germania vede i data center contribuire già per 10,4 miliardi al proprio PIL nel 2024, con previsioni di raddoppio entro il 2029. Il Regno Unito stima 55 miliardi di valore aggiunto annuo per un decennio grazie all’AI. [9]

Siamo in quello che Goldman Sachs chiama la “Appraisal Phase” del ciclo di innovazione AI: la fase in cui l’investimento in infrastruttura è più aggressivo, i multipli di valutazione più elevati, e la posta in gioco più alta. La transizione verso la “Execution Phase” – quando la domanda di calcolo sarà definita, i ritorni si normalizzeranno, e il vantaggio competitivo si sposterà dall’infrastruttura alla capacità di estrarne valore – è inevitabile. Ma non è ancora arrivata. [2]

Nel frattempo, l’infrastruttura invisibile continua a crescere: rack su rack, fibra su fibra, agente su agente. E chi la costruisce oggi, determinerà dove e come l’AI trasformerà il mondo nei prossimi dieci anni.

Fonti

[1] Deloitte TMT Predictions 2026
[2] Goldman Sachs GS SUSTAIN AI Data Center Power Demand (febbraio 2026)
[3] McKinsey Technology Trends Outlook 2025
[4] Equinix – Your Network Wasn’t Built for Agentic AI (febbraio 2026)
[5] Deloitte State of AI in the Enterprise – The Untapped Edge (2026)
[6] Ropes & Gray – Data Center Investment in 2026 (maggio 2026)
[7] Belfer Center Harvard – AI, Data Centers, and the U.S. Electric Grid (febbraio 2026)
[8] Bloom Energy 2026 Data Center Power Report
[9] Ember Energy – Grids for Data Centres in Europe (giugno 2025)

Today, many of the moments that shape the guest experience depend on the network.

Guest Expectations Are Reshaping Hospitality

Travel demand continues to show resilience. UN Tourism reported that 307 million people traveled internationally in the first quarter of 2026, a 2% increase over the prior year. At the same time, digital expectations inside the property continue to rise. Oracle Hospitality and Skift found that more than 60% of hospitality executives expect fully contactless basic hotel transactions, including check-in and check-out, food and beverage, and room keys, to be widely adopted.

That combination puts more pressure on the network. Connectivity now powers far more than guest Wi-Fi. It supports mobile check-in, streaming, staff communications, point-of-sale systems, security, IoT devices, conference services, outdoor amenities and back-office operations. The network has become the digital foundation for the entire property.

That shift is changing what owners, operators and IT teams need from their technology partners. The challenge is no longer simply connecting guests in the lobby. It is delivering a consistent experience across rooms, restaurants, meeting spaces, pool decks, outdoor areas and staff environments while keeping operations simple for lean IT teams and managed service providers.

From Connectivity to a Repeatable Operating Model

This is where Cambium Networks helps hotels turn connectivity into a more consistent, supportable operating model.

Cambium helps hospitality organizations move from fragmented connectivity to a more unified, repeatable network model. With indoor and outdoor Wi-Fi, wall-plate access points, switching, security, fixed wireless broadband and cloud-based management, Cambium gives hotels a practical way to connect the full property, not just isolated parts of it.

That repeatable model is where the value grows. Owners can protect reviews, pricing confidence and portfolio consistency. General managers can reduce guest complaints and front-desk escalations. IT teams and managed service providers can lower tickets, accelerate rollout and simplify support across one property or many.

Cambium’s advantage is delivering reliable connectivity with the operational simplicity and disruptive economics hotels need: standardization, supportability, lower long-term friction and a network foundation that can grow with the property. Looking ahead, Cambium’s next-generation Intent Based Networks will help deliver repeatable, optimized configurations focused on outcomes for our customers rather than Wi-Fi access alone.

Building the Foundation for the Connected Guest

Across the Americas, hospitality environments vary widely, from business hotels and regional chains to resorts, mixed-use properties and outdoor destinations. But the need is consistent: secure, reliable, manageable connectivity that supports both the guest experience and the business behind it.

As hotels continue to modernize, digital services will keep expanding across the property. The winners will be the properties that treat connectivity as a strategic operating platform, not just an amenity. Cambium helps hospitality organizations make that transition with a solution that is scalable, supportable and ready for the demands of the connected guest.

Reach out to Cambium today to explore how we can support you to deliver exceptional guest experiences across your property.

Network planning is most effective when teams can model links accurately, work efficiently, and rely on up-to-date product and regulatory information. With the online release of LINKPlanner V6.4.2, Cambium Networks continues to make wireless network design simpler, faster, and more dependable.

This update brings new planning capabilities, workflow improvements, and expanded product support to help users move from design to deployment with greater confidence.

Plan with More Confidence

LINKPlanner V6.4.2 adds support for the ITU-R P.530-19 Prediction Model, giving users another trusted option for evaluating microwave link performance. For planners designing PTP networks, this helps support more informed decisions during the design stage and improves confidence before equipment is deployed in the field.

Save Time in Everyday Project Work

This release also includes usability enhancements that reduce repetitive steps and make projects easier to manage.

A new project-level Revert Profile option gives users a faster way to roll back project profile changes when needed. Custom styles and colors for PTP, PMP, and Mesh elements make it easier to visually organize complex projects, quickly distinguish network elements, and improve project readability.

The addition of product-list search also helps users find the right products more quickly, especially when working across large portfolios or detailed project designs.

Stay Current with Product and Regulatory Updates

LINKPlanner V6.4.2 expands support for several Cambium Networks product families, including updates for PTP 820, PTP 850, ePMP, and PMP 450. The release also includes regulatory updates for supported regions, helping users design with current product availability and regional requirements in mind.

For ePMP 4000 Series users, updated values aligned with Release 5.11.0 help improve planning accuracy for 6 GHz AFC-related calculations and other performance parameters.

A Better Planning Experience

Beyond new product and regulatory updates, LINKPlanner V6.4.2 includes refinements that improve the overall user experience. Table views now remember the Show All Rows setting across windows, helping users maintain their preferred view while working through project details.

Together, these updates help reduce manual effort, improve project clarity, and support more accurate network designs.

Learn More

For the full list of updates and additional release details:

Visit the Cambium Community Announcement

81% of ePMP radios tracked in this snapshot are running firmware that’s been superseded — some by over two years. Here’s what those networks are leaving behind.

~19%

On recommended firmware, 5.11.0 or 5.11.1

~36%

Are 16 to 34+ months behind, running pre-5.10 firmware

5.12

Beta this summer, and the gap will keep growing

In a sampling of 14,238 ePMP radios, only about one in five is running the current recommended firmware. 64% are on 5.10.0 or later and already have access to the latest features like MU-MIMO, Asymmetric Uplink Bandwidth, and 6 GHz AFC tools — one upgrade away from 5.11.x. The remaining 36% are on pre-5.10 firmware, 16 to 34+ months behind. Every one of these improvements shipped as a free software update. Operators who stay current continuously extract more value from the hardware they’ve already paid for. Those who don’t are leaving that investment sub-optimally deployed — risking support calls, churn, and missed opportunities to exceed customer expectations.

How Far Behind Is Each Firmware Group?

Approximate device counts by version · Release dates per Cambium support index · Staleness as of June 2026

5.11.1 May 5, 2026

~130 devices · 1 month behind

5.11.0 Mar. 24, 2026

~2,580 devices · 2 months behind

5.10.4 Nov. 4, 2025

~2,100 devices · 7 months behind

5.10.3 Aug. 12, 2025

~1,450 devices · 10 months behind

5.10.1 Jul. 3, 2025

~630 devices · 11 months behind

5.10.0 Apr. 1, 2025

~2,150 devices · 14 months behind

Pre-5.10 5.9.x and earlier

~5,100 devices · 36% · 16–34+ months behind

Current (5.11.1)

1–3 months (5.11.0)

4–8 months (5.10.4)

9–12 months (5.10.1–5.10.3)

13–15 months (5.10.0)

16–34+ months (pre-5.10)

Four New Capabilities That Require 5.10.x or Later

The 5.11.0 release (March 2026) and the 5.10.0 release (April 2025) together delivered substantial new capabilities — not just bug fixes. Networks still running older software are missing out on these features.

New in 5.10.0

Asymmetric Uplink Bandwidth

Run a wide downlink channel where capacity matters and a narrower uplink where interference is the bigger concern. Same spectrum, less noise exposure, better SNR and fewer retries. Since upload demand is typically a fraction of download demand, this is a free performance improvement for most deployments. Unique in the industry and only on ePMP 4000 series from Cambium.

New in 5.11.0

MU-MIMO Wireless Link Test

Identifies which SMs are eligible for MU2 or MU3 grouping and compares SU-MIMO vs. MU-MIMO throughput directly. Without this, you cannot confirm whether MU-MIMO is actually delivering gains on a given sector, or diagnose why it is not. Available on ePMP 4600 and 4500 series.

New in 5.11.0

Antenna Gain Profiles Across Spectrum

Integrated antennas now use 8-point gain curves from real measurements rather than a flat gain value, more accurate EIRP calculations, better regulatory compliance, and automatic transmit power optimization where the antenna is less efficient. No configuration required.

New in 5.11.0

AFC Analyzer for 6 GHz

Visualizes EIRP limits per channel for every AP and SM across the 6 GHz band. Operators can see exactly how individual SMs constrain available channels before making changes, eliminating trial-and-error during deployments or channel reconfigurations.

The Economic Case for Staying Current

Firmware upgrades do take effort and planning, but delaying them means missing capabilities you’ve already paid for by investing in 4000 Series hardware.

What Outdated Firmware Actually Costs

Underperforming hardware. The ePMP 4500 and 4600 series are physically capable of MU-MIMO, asymmetric uplink, and 6 GHz AFC optimization. Operators on pre-5.10 firmware are running hardware they haven’t fully activated.

Support tickets for closed issues. Cambium resolved dozens of defects across 5.10.x and 5.11.x — crashes, SNMP failures, GPS instability, PPPoE reconnection delays. Every ticket filed against a bug that’s already fixed in an available release is a support cost that shouldn’t exist.

Subscriber churn from degraded reliability. Intermittent disconnects, latency spikes under load, and throughput inconsistency are the top drivers of churn in fixed wireless. Several defects fixed in recent releases directly affect link stability under real-world traffic patterns. Running a known-fixed version is the cheapest churn prevention available.

Security posture. The 5.10.2 release addressed EU RED cybersecurity requirements. Operators on older firmware are carrying default security configurations that have since been hardened — an exposure that becomes relevant the moment an audit or incident occurs.

You don’t have to upgrade blind. Check the Cambium Community forums and ePMP Facebook groups before rolling out networkwide. WISPs share real-world results, flag edge cases, and validate performance on specific hardware. That peer knowledge, combined with Cambium support, reduces upgrade risk significantly.

What’s Coming Next

The 5.12.x beta ships this summer with 4000 series SM-as-an-AP, AFC improvements (failover, caching, alternate channel), and more performance work. Networks already on 5.11.x can evaluate it quickly; those three or four releases behind should treat this as the right moment to catch up. Firmware management is an important aspect of running a profitable network. The operators with the lowest upgrade risk aren’t the ones who wait longest — they’re the ones who stay closest to the latest and greatest release.

Ready to Close the Gap?

Use cnMaestro to filter your network inventory by software version. Start with any AP on 5.9.x or below. Those sectors have the longest list of unresolved issues and missed capabilities.

VIEW RELEASE NOTES AND UPGRADE GUIDE

Mobile data demand continues to grow at an unprecedented pace. At the same time, expanding traditional cellular infrastructure is becoming more expensive and increasingly difficult to scale. As carriers look for smarter ways to improve coverage and reduce congestion, Wi-Fi offload is emerging as a practical and profitable solution.

For Managed Service Providers (MSPs) and Wireless Internet Service Providers (WISPs), this shift creates a significant opportunity: monetize existing Wi-Fi and broadband infrastructure while delivering better mobile experiences.

What Is Wi-Fi Offload?

Wi-Fi offload enables mobile devices to seamlessly transition from cellular networks to trusted Wi-Fi infrastructure using technologies like Passpoint and OpenRoaming.

To the user, the experience feels automatic. Devices connect securely without manual logins or interruptions. Behind the scenes, carriers are shifting traffic from overloaded mobile networks onto carrier-grade Wi-Fi infrastructure.

The result:

Better coverage
Reduced cellular congestion
Improved user experiences
Lower infrastructure costs for carriers

For MSPs and WISPs, it means existing networks can now participate in the broader mobile ecosystem.

The Business Model Many Providers Miss

One of the biggest misconceptions about Wi-Fi offload is assuming revenue comes directly from end users. In reality, carriers fund offload initiatives because they benefit from reduced mobile network strain.

Most MSPs and WISPs participate through ecosystem partners such as:

Offload aggregators
OpenRoaming brokers
Managed service platforms
Technology integration providers

Companies like WayFi and XNET already maintain carrier relationships and simplify onboarding into the offload ecosystem. This allows providers to focus on network performance rather than negotiating directly with mobile operators.

Where Wi-Fi Offload Delivers the Most Value

Not every deployment generates meaningful revenue. The strongest opportunities typically exist in environments with:

High user density
Heavy mobile data usage
Weak or congested cellular coverage
Long user dwell times

High-value deployment locations include:

Transportation hubs
Campuses and commercial buildings
Shopping malls and retail centers
Hospitality venues and restaurants
Hotels and multi-dwelling units (MDUs)

These environments create ideal conditions for sustained mobile traffic offload.

Operational Simplicity Matters

Carrier-grade Wi-Fi offload requires more than adding access points. Providers need secure onboarding, centralized policy control, seamless roaming, and consistent application performance.

This is where Cambium Networks’ ONE Network architecture helps simplify deployment and operations. Cambium integrates Wi-Fi, switching, SD-WAN, security, fixed wireless, and fiber management into a unified platform managed through cnMaestro™.

Key advantages include:

Integrated Passpoint and OpenRoaming configuration
Centralized multi-site management
Application-aware traffic visibility and control
Scalable Wi-Fi performance for high-density environments
Reduced operational complexity and lower OpEx

For MSPs and WISPs, operational efficiency directly impacts profitability and scalability.

Why MSPs and WISPs Are Positioned to Win

Most MSPs and WISPs already own the foundational assets needed for Wi-Fi offload:

Broadband backhaul
Indoor and outdoor Wi-Fi infrastructure
Distributed managed network environments
Existing customer relationships

This makes Wi-Fi offload less about building new infrastructure and more about unlocking new value from networks already in place.

As mobile carriers continue to look for scalable alternatives to costly macro network expansion, providers that align with the offload ecosystem today will be better positioned for future growth.

Wi-Fi is no longer just a convenience layer. It is becoming a strategic extension of the mobile network.

To learn more about how MSPs and WISPs can participate in the growing Wi-Fi offload ecosystem, read the full solution paper here:

Download the Wi-Fi Offload Solution Paper

Enterprise networks are under more pressure than ever. Cloud applications, AI-driven workflows, video collaboration, high-density environments, IoT devices, and real-time services all demand faster, more reliable wireless connectivity. Wi-Fi 7 is built for this moment.

Cambium Networks is helping organizations move beyond traditional wireless limitations with Wi-Fi 7 solutions that deliver higher throughput, lower latency, and more predictable performance across the network. As part of Cambium’s ONE Network architecture, Wi-Fi 7 becomes more than just faster wireless. It becomes the foundation for resilient, scalable enterprise connectivity.

Why Wi-Fi 7 Matters

The evolution of Wi-Fi has consistently improved speed and efficiency, from Wi-Fi 4 introducing MIMO to Wi-Fi 6E expanding into the 6 GHz spectrum. Wi-Fi 7 takes the next step with technologies designed to make wireless a true wired replacement.

Key advancements include:

Multi-Link Operation (MLO) for better reliability and responsiveness
320 MHz channels doubles max channel width for significantly higher capacity
Deterministic low latency for mission-critical applications
Improved resiliency in high-density environments

The result is a network that can support demanding enterprise applications without compromising user experience.

Designed for Real-World Enterprise Demands

Cambium’s Wi-Fi 7 solutions are purpose-built for organizations where performance and uptime directly impact operations.

Ideal Use Cases Include:

Education networks needing seamless coverage for students and staff
Smart cities and public Wi-Fi deployments supporting IoT and citizen connectivity
Service providers leveraging Wi-Fi offload to improve subscriber experiences
Hospitality and retail environments supporting guests and transaction systems
High-density venues such as campuses, stadiums, and auditoriums
Healthcare and industrial operations requiring low latency and reliability

A Complete Wi-Fi 7 Solution

Modern enterprises need more than access points. They need a unified platform that simplifies operations while scaling for future demands.

Cambium delivers this through four integrated pillars:

Wi-Fi 7 Access Points

Cambium’s portfolio includes solutions tailored to different deployment requirements:

X7-35X – Flexible tri-band 2×2 AP for mainstream enterprise deployments
X7-55X – High-performance software-definable tri-band 4×4 AP for demanding environments
X7-53X – Cost-effective dual-band 4×4 AP optimized for deployment efficiency

cnMatrix™ Multi-Gigabit Switching

With 2.5 Gbps access ports, uplinks up to 25 Gbps and PoE delivery up to 90W per port, cnMatrix switches scale to deliver the performance and power required for Wi-Fi 7 infrastructure while supporting IoT devices, VoIP systems, and high-performance endpoints.

NSE Next-Generation Firewall

Cambium’s Network Service Edge (NSE) integrates advanced security, SD-WAN, and application-aware network services into a single platform, helping organizations simplify end-to-end network deployments.

cnMaestro™ Network Management

cnMaestro delivers centralized cloud or on-premises management across the entire network with capabilities including:

AIOps-driven Assurance for faster troubleshooting
EasyPass for secure onboarding of users and IoT devices
MarketApps to extend network functionality and reduce complexity
Application Control for visibility and control of thousands of applications
APIs for third-party integrations and workflow automation

Together, these capabilities create a network that is easier to manage, more resilient, and ready for future growth.

Stronger Portfolio. Smarter Network.

Cambium’s ONE Network approach unifies Wi-Fi, switching, security, SD-WAN, and network management into a single architecture. The benefits are clear:

Future-ready performance for next-generation applications
Resilient infrastructure with high availability and multi-gigabit scalability
Simplified operations through centralized management and automation
Lower total cost of ownership with integrated tools and streamlined deployment

Wi-Fi 7 is not simply about delivering more speed. It’s about enabling organizations to confidently support the next generation of digital experiences with connectivity that just works.

Read the Full “Unleashing the Power of Wi-Fi 7” Solution Paper

For LINKPlanner users, a bill of materials is more than a parts list. It is often the bridge between network design, project planning, customer quoting, and deployment. The faster and more confidently teams can create and manage BOMs, the easier it becomes to move from design to execution.

That is why the latest LINKPlanner BOM Estimator v1.2.0 release is an important step forward. This update is designed to make BOM creation easier to manage, harder to lose, and better aligned with real deployment workflows.

You can view the full release details in the LINKPlanner BOM Estimator v1.2.0 Community post.

Keep Your BOM Projects Organized

One of the biggest improvements in this release is the ability to manage BOM projects directly from your account. Users can now save, open, rename, clone, and delete projects, making it much easier to work across multiple designs or customer scenarios.

This matters because BOM work is rarely one-and-done. Network plans evolve. Customers ask for options. Teams compare product combinations, revise assumptions, and return to previous versions. With v1.2.0, users have a more practical way to keep that work organized instead of relying only on manual exports or local files.

Spend Less Time Recreating Work

Few things are more frustrating than building out a BOM and realizing later that changes were not saved. This release helps reduce that risk with clearer save behavior and warnings when unsaved changes are present.

That may sound simple, but it is a meaningful workflow improvement. LINKPlanner BOM Estimator users often work through multiple decisions in a single session. Better save visibility helps protect progress and gives users more confidence when switching projects or starting something new.

Move More Easily Between Planning Sessions

With cloud-based project handling, users can come back to their BOM work more easily. Projects are sorted by recent activity, and work started before logging in can sync after login.

For teams juggling several opportunities, sites, or design alternatives, this helps turn the BOM Estimator into a more persistent project workspace rather than a single-session tool.

Better Support for Real-World Deployments

Version 1.2.0 also adds product updates that help BOMs better reflect the latest product changes. This includes support for new cnMatrix™ EX3000 25 Gbps fiber transceiver options, Egypt-specific Wi-Fi 6 SKUs, and updated PoE part numbers.

These updates are important because BOM accuracy directly affects quoting, ordering, and installation planning. The closer the estimator is to current product and regional requirements, the more useful it becomes as a planning tool.

Easier Access for More Users

The release also expands login support with Cambium SSO and Channel Partner Portal SSO. This helps make the BOM Estimator easier to access for both internal teams and partners who rely on the tool as part of their planning and proposal workflows.

A Smoother Experience from Design to Deployment

The LINKPlanner BOM Estimator v1.2.0 release is focused on practical improvements that help users work more efficiently: saving projects, protecting changes, managing multiple BOMs, and keeping product selections current.

For LINKPlanner BOM Estimator users, these updates help reduce friction in the planning process and make it easier to move confidently from a network design to a complete, actionable bill of materials.

Check out the LINKPlanner BOM Estimator today.