Galleries

AI Is Reshaping Global Connectivity: Why Subsea and Terrestrial Networks Matter

NJFX Blog /

Insights  /  SecureWAVES

AI Is Reshaping Global Connectivity
Why Subsea and Terrestrial Networks Matter

How artificial intelligence is reshaping global connectivity and bringing new importance to the infrastructure connecting continents, networks, and data centers.

  • NJFX
  • October 2026
  • 6 min read

Artificial intelligence is transforming the global digital infrastructure landscape. From the rapid expansion of GPU-powered data centers to the growing adoption of AI applications, the industry is investing heavily in the computing power behind this next generation of technology. Data center spending currently accounts for most of that capital, but the long-haul subsea and terrestrial networks linking those facilities are becoming more critical.4

But behind every AI deployment is an equally important requirement: the network.

As AI workloads grow in scale and complexity, moving data between data centers, cloud environments, and global markets is becoming more critical. That is bringing renewed attention to the submarine cable systems that connect continents and the terrestrial networks that carry that capacity to its destination.

Subsea and terrestrial infrastructure have long served complementary roles in global communications. AI is accelerating the need for them to operate as one integrated connectivity ecosystem.

The future of AI infrastructure will depend not only on where computing power resides, but on how effectively it connects to the rest of the world.

Subsea Networks: The Global Foundation of AI Connectivity

Submarine cable systems carry more than 99% of intercontinental data traffic and form the backbone of global communications.12 They connect international markets, cloud regions, financial centers, and increasingly, the infrastructure supporting artificial intelligence.

AI training requires access to enormous datasets, often collected, stored, and processed in different geographic locations. Training tends to happen where power is abundant, while inference is placed close to users.3 Inference is creating new traffic patterns as applications deliver more sophisticated services to users worldwide, and as inference compute spreads geographically, the bandwidth needed between data center sites grows with it.5

TeleGeography forecasts used international bandwidth to grow about 24% a year from 2025 to 2035, roughly doubling every three years. Content providers, including hyperscalers, cloud providers, AI platforms, and neoclouds, already account for 75% of used international capacity.4

The industry is not simply preparing for more traffic. It is preparing for a changing mix of workloads that place different demands on capacity, latency, geographic reach, and resilience.

That shift is shaping investment in higher-capacity cable systems, new international routes, and the infrastructure that connects those systems to domestic networks.

Where the Subsea Network Meets the AI Data Center

Submarine cables provide the international connectivity that links global markets. The value of that capacity depends on what happens when it reaches land.

At the cable landing station, international subsea systems transition into terrestrial infrastructure, where connectivity extends toward data centers, cloud platforms, metropolitan networks, and enterprise locations.

Subsea Cables

Intercontinental capacity

Cable Landing Station

Ocean to land, global to domestic

Terrestrial Fiber

Diverse routes inland

AI Data Centers

Training, inference, cloud

For AI, the entire connectivity path matters.

This transition matters more as AI changes the geography of computing. AI infrastructure is not confined to a single data center or region. Training clusters, inference environments, cloud services, and enterprise applications may run across multiple locations, each requiring access to a broader network ecosystem.

A high-capacity international cable must be supported by terrestrial networks that can carry traffic efficiently to the computing environments where it is needed. Likewise, data centers supporting AI must consider their access to international connectivity, network diversity, and room to scale capacity as demand grows.

The result is a closer relationship between subsea network operators, terrestrial fiber providers, and data center developers.

The cable landing station is no longer simply where international connectivity reaches shore. It is a strategic point where global networks and AI infrastructure intersect.

AI Is Changing the Requirements for Global Network Architecture

AI is not only increasing the amount of data crossing networks. It is influencing how those networks are designed.

International connectivity strategies have traditionally focused on available capacity, geographic reach, and cost. Those remain essential, but AI is bringing four requirements to the front.

01

Greater network capacity

As datasets and distributed computing environments grow, operators must keep expanding their ability to move large volumes of data between international and domestic markets.

02

Physical route diversity

Greater dependence on interconnected infrastructure raises the importance of geographically diverse cable systems, terrestrial routes, and network entry points. The ITU and ICPC's 2026 resilience report names route diversity a global priority.2

03

Low-latency connectivity

Some distributed AI applications need responsive connections between compute and end users, which makes network architecture and infrastructure placement more important.

04

Scalable interconnection

AI developers need the flexibility to connect with multiple network providers, cloud ecosystems, and computing locations as requirements change.5

Together, these priorities move the conversation away from individual cable systems and network segments toward a view of end-to-end connectivity. A submarine cable may deliver substantial international capacity, but its performance and resilience depend on the terrestrial infrastructure and interconnection options behind it.

For AI, the entire connectivity path matters.

SecureWAVES: Bringing Subsea, Terrestrial, and AI Infrastructure Together

At NJFX, the convergence of subsea and terrestrial infrastructure has been central to the development of our carrier-neutral connectivity campus.

Located in Wall Township, New Jersey, NJFX provides direct access to international submarine cable systems alongside an ecosystem of terrestrial network operators, interconnection capabilities, and colocation infrastructure. Global connectivity and domestic networks come together in one strategically important location.

Through SecureWAVES, NJFX is highlighting this convergence as the digital infrastructure industry evolves to support AI. With access to multiple network operators and diverse connectivity options, customers can build network strategies that extend beyond a single carrier, route, or traditional interconnection market.

NJFX's continued evolution toward supporting high-density AI infrastructure reinforces the link between compute and connectivity. The objective is not simply to bring AI computing closer to international networks. It is to support an ecosystem where subsea capacity, terrestrial fiber, interconnection, and computing requirements are considered together.

The Next Chapter of Subsea Connectivity Will Be Driven by AI

The subsea industry has long played a foundational role in connecting economies, businesses, and digital services. Artificial intelligence is introducing another period of transformation.

As AI investment accelerates, demand for international capacity will continue to shape the development of submarine cable systems and the terrestrial networks supporting them. But the opportunity extends beyond larger cables or more fiber. It lies in stronger connections between the international networks carrying global traffic and the domestic infrastructure delivering that capacity to compute.

Cable landing stations, carrier-neutral interconnection points, and geographically diverse terrestrial routes will remain core components of that architecture. For NJFX, SecureWAVES reflects this next chapter: bringing together the infrastructure that connects global networks with the growing requirements of AI.

The next generation of AI will not be powered by computing capacity alone. It will depend on the global networks connecting it.

Talk to NJFX about SecureWAVES

Sources

  1. Mauldin, A. "Do Submarine Cables Account for Over 99% of Intercontinental Data Traffic?" TeleGeography, May 2023. resources.telegeography.com
  2. International Telecommunication Union. "International Advisory Body approves landmark report to strengthen submarine cable resilience." ITU, July 2026. itu.int
  3. Carlson, K. "The Impact of AI on Submarine Cable Networks." TeleGeography, January 2025. resources.telegeography.com
  4. Mauldin, A. "Connecting the AI Revolution: Long-Haul Networks and Bandwidth Growth." TeleGeography, June 2026. resources.telegeography.com
  5. Sant'Anna, F. "AI Inference Is the Next Network Stress Test." Ciena, January 2026. ciena.com

AI Is Reshaping Global Connectivity: Why Subsea and Terrestrial Networks Matter Read More ยป

Where the GPUs Go, The Networks Have to Follow

1 Comment / NJFX Blog /

NJFX Insights  |  AI Infrastructure  |  7 min read

Wherever the GPUs Go, the Network Has to Follow

AI is already network intensive inside the data center. Between roughly 2027 and 2030, that intensity reaches carrier, metro and subsea networks. The commercial opportunity starts sooner.

Artificial intelligence is changing what the industry expects from a data center.

Most of the conversation around AI infrastructure centers on GPUs, power and cooling. Those resources are essential, but they describe only part of the system. AI is also extraordinarily network intensive. The more useful questions are where that intensity shows up, and when.

Wherever GPUs are deployed, the network stops being a connection to the infrastructure. It becomes part of the infrastructure.

Inside the facility first, outside next

Today, most AI traffic stays inside the building. A text prompt and its response may amount to a few megabytes or less. The heavy communication happens between GPUs and storage systems within the AI facility.

Meta's AI Research SuperCluster illustrates the scale. Its GPU systems communicate over a 1,600 Gb/s InfiniBand fabric with no oversubscription, supported by a storage service built to serve terabytes per second of bandwidth.1

East-west

Traffic between GPUs, storage and compute clusters inside the facility. Already intense today.

North-south

Traffic entering and leaving the facility: cloud connections, enterprise access, data movement and inference results reaching users. Scaling next.

AI training has already made the data center network essential. Widespread inference will make the external network strategically essential, with carrier, metro and subsea demand becoming meaningful between roughly 2027 and 2030. The distinction matters, because each network layer reaches that point on a different timeline.

Network layerTimingWhat drives demand
Inside an AI clusterAlready happeningThousands of GPUs exchanging model parameters at 400, 800 and 1,600 Gb/s
Data center interconnection2026โ€“2028Distributed training, data movement, replication, storage and disaster recovery
Metro and regional networks2027โ€“2029Inference moving closer to users; AI clouds connecting to enterprises
Long-haul and subsea2028โ€“2032Global inference, sovereign AI replication, multimodal content and distributed AI factories
Consumer access networksGradualAI video, real-time translation, autonomous agents, robotics and spatial computing
Projected timing of AI-driven network demand by layer.

Three shifts that move AI traffic outside the building

The inflection point for external networks arrives as three developments scale together.

Distributed inference

Applications route workloads dynamically across multiple AI facilities based on latency, capacity and power availability. Traffic follows the workload from site to site.

Multimodal AI

Continuous video, voice, sensor data and AI-generated video require far more bandwidth than text.

Agentic AI

Millions of software agents run continuously, retrieving data and communicating with multiple cloud platforms instead of waiting for individual human prompts.

Metro networks become the AI on-ramp

Inference places AI closer to the applications and people it serves. Traffic needs efficient paths between AI infrastructure, enterprises, cloud platforms, network exchanges and population centers.

Metro fiber becomes the on-ramp and off-ramp for AI. For carriers, that creates demand for more than bandwidth. It calls for route diversity, lower latency and scalable connections into AI-ready facilities.

Where compute is located matters. So does the network architecture surrounding it.

Carrier networks have to scale with compute

A high-density AI rack concentrates tremendous computing power in a small physical footprint. The effects of that concentration reach well beyond the rack.

As deployments grow from individual installations to multi-megawatt environments, carriers must be ready to carry the traffic those environments produce. That requires higher-capacity optical systems, additional fiber, diverse routes and stronger interconnection ecosystems.

The relationship between the data center and the carrier changes as a result. Connectivity stops being a service delivered into the building.

The carrier ecosystem becomes part of the facility's ability to support AI at scale.

AI is global, and subsea infrastructure carries it

AI runs inside a physical building, but the data it depends on does not stop at borders. Enterprises operate internationally. Cloud platforms span continents. A model can be trained in one market, refined in another and accessed by users around the world. Sovereign AI requirements add another layer, as models and data are replicated within specific jurisdictions.

Subsea cable systems carry that traffic between regions, and their share will grow as AI increases the volume of data moving across borders. Facilities positioned close to subsea infrastructure reduce unnecessary network hops and create more direct paths between domestic and international networks.

As AI deployments scale, the connection between AI infrastructure and subsea infrastructure becomes a design decision rather than an afterthought.

The commercial window opens sooner

The external network becomes strategically essential over the next two to four years. The commercial opportunity begins earlier, between 2026 and 2028, as AI tenants start to require:

  • Multiple diverse 100G, 400G and 800G routes
  • Direct cloud and GPU cloud connectivity
  • Low-latency access to New York and Ashburn
  • Dataset and checkpoint transport
  • Private connectivity between AI campuses
  • Subsea access for sovereign and international AI workloads

None of these requirements are solved by adding GPUs. Each one is a network decision.

NVIDIA describes the AI factory as infrastructure that manages the full AI lifecycle, from data ingestion through training, fine-tuning and high-volume inference.2 Every stage of that lifecycle moves data, and increasingly it moves data between sites.

AI is turning compute facilities into networked factories. NJFX provides the carrier-neutral middle mile connecting those factories to clouds, enterprises, subsea systems and other AI campuses.

At our carrier-neutral cable landing station in Wall Township, New Jersey, high-density infrastructure, terrestrial fiber, carrier networks and subsea systems already operate as parts of the same architecture. Through SecureWAVES, customers and network operators can plan beyond adding capacity and build the bandwidth, diversity and resiliency AI requires as it scales.

AI training made the data center network essential. Widespread inference will make the external network strategically essential.

Talk with the NJFX team

Sources

  1. Meta AI, "Introducing the AI Research SuperCluster," January 2022. ai.meta.com/blog/ai-rsc
  2. NVIDIA, "What Is an AI Factory?" NVIDIA Glossary. nvidia.com/glossary/ai-factory

Where the GPUs Go, The Networks Have to Follow Read More ยป

Capacity Europe 2026

Events /
Capacity Europe 2026 NJFX Attending

Capacity Europe 2026

The connectivity industry's leading European gathering, bringing together carriers, cloud, and infrastructure leaders shaping global digital networks.

Dates

October 13–15, 2026

Location

InterContinental O2, London

Audience

Carrier, cloud & infrastructure executives

About the event

Capacity Europe brings together carriers, cloud and hyperscale providers, data centre operators, and enterprise buyers for three days of dealmaking and discussion on the state of global connectivity infrastructure.

Carriers Cloud & Hyperscale Data Centres Enterprise & Content

NJFX team attending

Meet the NJFX delegation in London. Find us on the show floor or reach out below to schedule time.

Felix Seda

Felix Seda

Chief Revenue Officer · NJFX

LinkedIn Bio
Emily Newman

Emily Newman

Marketing Director · NJFX

LinkedIn Bio

Meet us in London

The NJFX team is available for meetings throughout Capacity Europe 2026. Fill out the form and we'll be in touch to confirm a time that works.

Events

CCT Global 2026

NJFX is excited to attend CCT Global 2026 in Dublin, bringing together leaders shaping the future of cloud, AI, connectivity, and digital infrastructure. As a premier carrier-neutral cable landing station and interconnection hub, NJFX continues to support the evolving demands of global networks, enterprise, financial markets, and next-generation AI workloads. Meet the NJFX team at CCT Global as we explore innovative partnership strategies, scalable infrastructure, and the future of high-capacity connectivity across Europe and North America.

Meet Us ยป
Explore More

Capacity Europe 2026 Read More ยป

Why AI needs more fiber not just more GPUs

NJFX Blog /

NJFX Insights

Why AI needs more fiber, not just more GPUs

The AI infrastructure conversation has centered on compute. But every GPU cluster depends on a network diverse and large enough to move what it produces — and that requirement is now impossible to ignore.

WHERE NETWORK REQUIREMENTS ARE HEADED

Gigabits where most connections started
100s of Gb where AI workloads sit today
Terabits where compute-scale demand is heading

The infrastructure around the GPU is changing

The rapid deployment of AI infrastructure is pushing capacity requirements higher throughout the network. As more GPUs come online, organizations need to move increasingly large datasets into and out of those compute environments and that puts pressure on the fiber, optical equipment, and network infrastructure connecting them.

It's no longer enough to ask whether a facility can support the GPUs. Organizations also need to ask whether the network can support what those GPUs will produce. The pipes connecting the compute have to scale with the compute itself.

Scaling AI isn't just a compute challenge. It's a connectivity challenge.

More GPUs mean more data in motion

A GPU doesn't operate in isolation. During training, massive datasets need to reach compute clusters and move between interconnected systems. During inference, the challenge becomes even more distributed — every AI request has to travel from a user or application to compute infrastructure, get processed, and return across the network.

Multiply that across enterprises, applications, devices, and millions of users, and the scale of network traffic changes dramatically. More compute creates the ability to process more data. That also creates the need to move more data — which is why adding GPU capacity without considering network capacity simply relocates the bottleneck.

The infrastructure behind the infrastructure

Meeting terabit-scale connectivity requires more than adding bandwidth to a service. For network operators and infrastructure providers, it means upgrading the systems and pathways that sit around the compute itself.

  • Optical transport equipment
  • Routers and switching infrastructure
  • Fiber capacity
  • Cross-connect architecture
  • Power supporting network equipment
  • Physical pathways and network diversity
  • Data center and cloud connections
  • Subsea system access

Fiber is part of the AI supply chain

When the AI supply chain comes up, the conversation usually centers on GPUs, power, and cooling. Fiber belongs on that list. GPUs provide the processing power. Power keeps them operating. Cooling allows increasingly dense infrastructure to function. Fiber connects all of it to the rest of the world.

Unlike software-defined resources, physical network infrastructure can't always be added instantly. Fiber routes, optical systems, cable landing infrastructure, and physical pathways require planning and investment well ahead of need — which means organizations planning AI infrastructure have to think beyond the compute capacity they require today, toward the network capacity they'll require tomorrow.

Compute concentrates. Demand doesn't.

Training can happen inside large, centralized compute environments. Inference has to reach users and applications wherever they are — which reshapes what the network around AI needs to do.

Compute
campus

Centralized training environments, concentrated in a small number of locations

↔
Users &
apps

Inference demand, distributed across enterprises, devices, and population centers

Preparing the network for AI-scale capacity

This is the shift NJFX is calling SecureWAVES. We are seeing networks increasing their capacity inside a secure, carrier-neutral facility to keep pace with AI-scale demand. NJFX sits at the intersection of subsea cable systems, terrestrial fiber networks, data center infrastructure, and emerging AI compute requirements, giving customers a strategically connected location to upgrade and deploy the space, power, equipment, fiber, and diverse network paths that terabit-scale connectivity depends on.

That diversity matters as much as the capacity. Building larger-scale fiber connectivity without diverse, resilient paths just creates a bigger single point of failure. As AI traffic grows, the goal isn't only bigger pipes — it's bigger pipes with more than one way to get where they're going.

AI doesn't just need more GPUs. It needs more fiber to connect them.

The next phase of AI infrastructure won't be defined by compute alone. As requirements move from megawatts to gigawatts, and from gigabits to terabits, the network carrying that traffic has to keep pace.

Learn more about the SecureWAVES movement at NJFX

Why AI needs more fiber not just more GPUs Read More ยป

Active vs. Passive Network Diversity

NJFX Blog /
Network Resilience

Active vs. Passive Network Diversity

Redundancy on paper isn't the same as resilience in practice. These are the two layers of diversity that keep a network running, and where most outages actually happen.

SITE A SITE B FAILOVER TRIGGERED
Path A Path B

Two physically separate paths (passive diversity). Automatic failover between them (active diversity). Click the button to simulate a cut.

When a backhoe severs a conduit, or an anchor drags across a subsea route, the question that decides how long an outage lasts isn't whether a network had a backup path. It's whether that path was actually separate from the one that failed, and whether anything was watching closely enough to use it in time.

"Diversity" gets used as a single word for two different jobs. One is a property of geography and physical construction. The other is a property of equipment and protocol behavior. A network can have one without the other, and almost every diversity-related outage happens in the gap between them.

Passive Diversity

Two routes that share no physical infrastructure: different conduits, different entry points, enough geographic separation that one event can't reach both.

Active Diversity

Equipment and protocols that detect a failure and move traffic to another path automatically, without a person making the call.

Passive diversity: the physical layer

Passive diversity means two paths share nothing physical between them: different conduits, different building entrances, different cable landing points, different rights of way, and enough distance that a single backhoe, cable cut, flood, or fire can't reach both at once.

It's called passive because the separation exists whether or not anything is actively managing it. No equipment has to make a decision for two physical routes to be independent of each other. That independence is either engineered in from the start, or it isn't there at all.

What it protects against

  • Fiber cuts and construction damage
  • Single-conduit or single-duct failures
  • Localized events: fire, flood, a single landing station

What it doesn't cover

  • What happens after the cut
  • Whether traffic actually finds its way to the second path
  • How long that handoff takes

Active diversity: the operational layer

Active diversity sits one layer up, in the equipment and protocols that detect a failure and move traffic without a human involved: automatic protection switching, BGP re-routing across diverse carriers, redundant electronics that fail over in milliseconds.

It's called active because it requires something to be doing something in real time: monitoring signal, comparing routes, executing a switch.

What it protects against

  • Hardware and equipment failure
  • A single carrier's outage or congestion
  • Slow, manual recovery

What it doesn't cover

  • A physical path that was never actually separate
  • A shared point of failure underneath two "diverse" circuits
  • Anything upstream of the equipment doing the switching

If both active paths still run through the same physical conduit, active diversity has nothing left to fail over to. It can reroute traffic instantly and still lose the connection entirely, because there was only ever one physical path underneath both logical ones.

Side by side

  Passive Diversity Active Diversity
Layer Physical / civil infrastructure Network / operational
What separates the paths Distance, conduit, landing point, right of way Redundant hardware, routing logic, carrier choice
Protects against Cuts, construction damage, single-site events Equipment failure, carrier outages, congestion
Fails silently when Routes converge at one shared point despite looking diverse The physical paths underneath were never actually separate
Typical examples Diverse conduit entries, separated cable landing points, dual meet-me rooms Automatic protection switching, BGP multi-homing, diverse carrier peering

The failure mode that matters most

It's rarely a total absence of diversity that causes the worst outages. It's diversity that looks complete on paper. Two carriers can represent route diversity on a circuit map, and both circuits can still run through the same duct bank for the last mile before they reach the building. Two cable systems can look diverse in a due-diligence deck, and both land at a station a mile down the coast.

CARRIER A CARRIER B BUILDING SHARED DUCT BANK

Two carriers, one physical entrance. Diverse on the circuit map, not diverse on the ground.

The most dangerous single point of failure is the one that looks like redundancy.

Why both layers have to hold

Passive diversity without active failover means a working second path and a slow, manual recovery, potentially hours of downtime while someone finds and executes a reroute by hand. Active diversity without passive diversity means fast, automated failover to a path that shares a point of failure with the primary, so the switch has nowhere to go when it matters.

Real resilience is both, stacked: physical separation engineered in from the start, and equipment above it that can find and use that separation in milliseconds. Verifying one without the other is the most common gap in a network resilience plan, and usually the hardest one to see from a circuit diagram alone.

Where NJFX Fits

Diversity has to start at the physical layer

As a carrier-neutral cable landing station and interconnection facility in Wall Township, NJ, NJFX sits at the physical layer where passive diversity is either built in or it isn't. Independent subsea cable landings, geographically diverse terrestrial backhaul routes, and a dense, carrier-neutral ecosystem inside the facility give network operators genuine physical separation to build on, and the choice of independent providers to build real active diversity on top of it.

Before calling a network diverse, it's worth tracing both layers separately: where the physical paths actually run, and what happens the moment one of them goes dark. If the answer to either question is unclear, the diversity is theoretical until it's tested by an outage.

Active vs. Passive Network Diversity Read More ยป

Behind every AI inference is a network

NJFX Blog /
NJFX Insights  ยท  SecureWAVES

Behind every AI inference
is a network

Training built the case for GPUs. Inference is building the case for the network moving data between them, the cloud, and everyone waiting on an answer.

NJFX Editorial Team โ€ข August 2026 โ€ข 6 min read

For years, the AI infrastructure conversation centered on one question: where will we find enough compute? GPUs, power, cooling, data center capacity.

Those questions haven't gone away. But as AI moves from training models to putting them to work for millions of users through inference, a second question is moving to the forefront: where will the data move, and how fast can it get there?

Inference doesn't end when a model is built. It happens every time someone uses it, an answer, an image, a recommendation, an autonomous agent completing a task. And every one of those requests has to travel: from the model, across a network, to the person or system waiting on the other end. As inference scales, AI becomes not only a compute challenge, but a data movement challenge.

The data behind the shift

Hover a stat for its source
0x
year-over-year growth in AI token consumption
Cisco, AI Impact on WANs 2026
0x
longer average flow duration vs. a typical web transaction (1,292ms vs. 643ms)
Cisco, AI Impact on WANs 2026
0%
of AI inference flows run upstream-heavy, vs. ~0.5% of web traffic
Cisco, AI Impact on WANs 2026
$0B
in hyperscaler AI infrastructure orders booked in fiscal 2026
Cisco, FY2026 earnings
The shift

Training and inference ask different things of infrastructure

Training concentrates compute. Inference has to reach people, applications and other AI systems, often in real time. That difference is reshaping how AI infrastructure gets planned.

Training

  • Concentrates massive compute into GPU clusters
  • Runs over days, weeks, or months
  • Can often tolerate geographic distance
  • Priority: raw processing power working together

Inference

  • Interacts with users, apps, and other AI in real time
  • Runs continuously, every time a model is used
  • Latency, route diversity, and location matter directly
  • Priority: capacity, diversity, and network optionality
Where compute is heading

AI infrastructure is no longer one massive campus. It's becoming distributed.

Instead of concentrating everything in centralized compute campuses, the industry's next phase connects infrastructure across many locations, closer to where inference demand actually originates.

NVIDIA ยท GTC 2026

AI grids

NVIDIA and telecom operators including AT&T, Spectrum, and Akamai are building "AI grids": geographically distributed, interconnected AI infrastructure that runs inference closer to users across existing network footprints.

Read more โ†’
Google Cloud

Campus-as-a-computer

Google is decoupling its data center network into distinct domains under a "campus as a computer" philosophy, designed for the scale, low latency, and predictability inference workloads demand.

Read more โ†’
Front-end networking

Net-new network buildout

Dell'Oro Group projects front-end data center networking opportunities will grow at nearly 40% CAGR, driven in large part by the shift from training toward inference and agentic workloads.

Read more โ†’
Power determines whether GPU infrastructure can operate. Cooling determines how densely it can operate. Networking determines where that intelligence can go.
THE ROLE OF NETWORKING IN THE NEXT GENERATION OF AI INFRASTRUCTURE

Power

Determines whether GPU infrastructure can operate at all.

Cooling

Determines how densely that infrastructure can run.

Networking

Determines where the intelligence built on top of it can actually go.

Where AI meets global connectivity

Compute and connectivity, meeting at one point on the network

A model may be trained in one location. The data behind it may sit somewhere else. The enterprise deploying it may operate across regions. And the user asking the question could be thousands of miles away. Every inference request is a journey between those points.

NJFX's carrier-neutral Cable Landing Station campus in Wall Township, New Jersey brings subsea systems, terrestrial fiber, network operators, and data center infrastructure together at a strategically positioned point on the U.S. East Coast, giving that journey more options for how it happens.

Through SecureWAVES, customers and network operators are adding and upgrading equipment to move terabits of data across terrestrial and subsea networks, building the capacity behind the next generation of AI-driven data movement.

Carrier-neutral by design Subsea systems, terrestrial fiber, and network operators converge on one campus, so AI traffic has more than one route to travel.
East Coast positioning Wall Township, NJ sits at a strategic point for U.S. East Coast connectivity between transatlantic systems and inland networks.
Built for scale SecureWAVES upgrades are adding the higher-capacity equipment needed to move terabits of data as inference-driven traffic grows.

The AI infrastructure race started with compute. It won't end there. More GPUs require more power. Higher densities require more advanced cooling. And more inference will require bigger, faster, more resilient network pipes.

NJFX SecureWAVES โ€” The Building Blocks of AI

Sources & further reading

  • Cisco AI Impact on Wide Area Networks 2026 โ€” token growth, flow duration, and upstream traffic data cisco.com โ†’
  • Cisco Fiscal 2026 Q4 results โ€” $9.3B in hyperscaler AI infrastructure orders marketbusinessnews.com โ†’
  • Dell'Oro Group Front-end networking market forecast โ€” inference and agentic AI driving ~40% CAGR prnewswire.com โ†’
  • NVIDIA AI grids โ€” distributed, interconnected AI infrastructure announced at GTC 2026 blogs.nvidia.com โ†’
  • Google Cloud "Campus as a computer" โ€” data center and network architecture built for AI cloud.google.com โ†’

Behind every AI inference is a network Read More ยป

NeoClouds Determining Infrastructure, Connectivity Needs and Optimum Partnership Strategies

NJFX Blog /
CCT Dublin · Panel Session

Neocloud Providers Insight: Determining Infrastructure and Connectivity Needs and Optimum Partnership Strategies

NJFX Chief Revenue Officer Felix Seda joined this CCT Dublin panel on neocloud infrastructure and partnership strategy, discussing how the industry is moving from raw compute buildout to the networks that carry trained models into inference.

EventCCT Dublin Session8:45 AM CategoryNeoclouds & AI Infrastructure SpeakerFelix Seda, CRO, NJFX
5
Panelists on stage
Neoclouds, hyperscale & telco
1
NJFX voice in the room
Felix Seda, CRO
8:45
Morning session slot
CCT Dublin agenda
↑
Capacity demand climbing
Training → inference traffic
The Panel

Inside the session

The panel brought together five infrastructure leaders to unpack how neoclouds are accelerating change across the industry, and what it means for the partnerships providers and network operators build next.

Session Lineup
Neocloud Providers Insight · CCT Dublin
Moderator: Jezzibell Gilmore, General Manager, Service Providers, Kentik
  • Bryan HillGlobal Director, Strategy & Business Development, AI/HPC & Digital Media · Digital Realty
  • Felix SedaChief Revenue Officer · NJFX
  • Sergio SoliveraGlobal Nokia Alliance Leader and Telco Partnerships · Nscale
  • Clara UlkenProgram Director, Edge & AI Infrastructure · Gcore
  • Michael ChampionChief Information Security Officer · Verda
The Concept

The AI Grid

Felix referenced a moment from Nvidia's GTC conference, where CEO Jensen Huang described the concept of the "AI Grid." In Felix's framing, the AI Grid is current network infrastructure: the physical and digital pathways that move data and compute across the world.

FS

Neoclouds are now in their next phase: connecting their sites and working with network operators to drive trained traffic to inference.

Felix Seda, Chief Revenue Officer, NJFX
The Next Phase

Connecting sites, driving traffic from training to inference

Felix's core point on the panel: neoclouds are now in their next phase of maturity. Having built out compute capacity, they are turning to the work of connecting their sites and partnering directly with network operators. That partnership work is what will ultimately drive trained traffic toward inference, moving AI workloads from where models are built to where they are put to use.

It's a natural extension of where NJFX sits in the ecosystem: at the intersection of subsea connectivity and the infrastructure decisions neoclouds are making as they scale.

The Capacity Shift

Bigger pipes for a bigger movement

A recurring thread across the panel was scale. As neoclouds move traffic from training environments to distributed inference, the volume of data moving between sites is growing well beyond what legacy network paths were built to carry. Panelists agreed that this shift is pushing operators toward larger amounts of capacity and higher-throughput connections, not as a future consideration, but as a present requirement.

That demand is reshaping how neoclouds and network operators plan interconnection. Smaller, incremental capacity additions are giving way to conversations about bulk capacity, higher-count fiber paths, and connections built to scale with AI workloads rather than be upgraded piecemeal after the fact. The panel framed this as the next constraint the industry has to solve for: it is no longer only about where compute sits, but whether the pipes connecting it can keep pace with the traffic AI is generating.

For carrier-neutral campuses like NJFX, that shift reinforces the value of sitting at a point where multiple networks and cable systems already converge, giving operators room to scale capacity without being boxed in by a single provider's infrastructure.

On the Panel
Digital Realty · Nscale · Gcore · Verda
Co-panelists alongside NJFX
Kentik
Session moderator organization
Training → Inference
Core traffic shift discussed
Takeaway

A strong showing for NJFX

Sharing the stage with leaders from Digital Realty, Nscale, Gcore, and Verda underscored how central these partnership questions have become across the industry, and how relevant NJFX's carrier-neutral, subsea-adjacent position is to the conversation.

It was a valuable morning session and a strong representation for NJFX at CCT Dublin.

Ready to connect at the subsea edge?

Talk to our team about how NJFX can anchor your global network strategy.

Get in touch

Get in touch

NeoClouds Determining Infrastructure, Connectivity Needs and Optimum Partnership Strategies Read More ยป

Welcome Grace Merkler: Powering NJFX CLS with AI

NJFX Blog /
GM
AI & Team

Powering the CLS with AI: Welcoming Grace Merkler

NJFX is excited to welcome Grace as part of the 2026 Summer Internship Program, bringing generative and agentic AI to the heart of the Cable Landing Station and investing in the next generation of digital infrastructure leaders.

Published
July 2026
Category
AI & Infrastructure
Author
NJFX Editorial
UR
University of Richmond
Junior, Business Administration
Marketing
Concentration in Marketing
Minor in Health Studies
ฮšฮšฮ“
Kappa Kappa Gamma
Recruitment Leadership
Sales Club
University of Richmond
Member
Trinity Hall
4 Years of Engineering Study
Founder, HUMM Institute
13.1
Half Marathon Finisher
Richmond and beyond
Summer '26
2026 Internship Program
AI Agents & Marketing Tech
Wall Township, NJ
NJFX Headquarters
Monmouth County resident
The Welcome

Investing in the Next Generation of Digital Infrastructure Leaders

NJFX is excited to welcome Grace Merkler to the team as part of the 2026 Summer Internship Program. As NJFX continues to invest in developing the next generation of leaders within digital infrastructure, marketing, and technology, we are proud to provide students with hands-on opportunities to learn, grow, and contribute within a rapidly evolving industry.

Grace joins us at a pivotal moment, as NJFX celebrates its 10th anniversary and continues to expand its role as a carrier-neutral cable landing station at the intersection of subsea connectivity, cloud infrastructure, and AI-era compute. Her curiosity, technical grounding, and drive to build are a welcome addition to the team.

AI at the CLS

Enabling the Team with Generative and Agentic AI

Since joining NJFX, Grace has been enabling the team by putting generative and agentic AI to work at the highest level, building and refining agents that take on the tedious, repetitive tasks so the team can focus on higher-value work. From researching the right AI tools to designing and testing agents built for NJFX's specific workflows, Grace has made efficiency a hands-on discipline rather than a talking point.

That work extends across the tools the team relies on every day: aligning automation within HubSpot, integrating Microsoft Copilot into daily workflows, and running internal demos that help the whole team put new tools to use with confidence. NJFX holds itself to a high standard as North America's first carrier-neutral Cable Landing Station, and Grace is here to make sure that standard extends to how the team works, building the agents, streamlining the workflows, and testing the tools that keep NJFX as efficient on the inside as it is reliable for the carriers, hyperscalers, and enterprises it connects on the outside.

About Grace

An Engineer's Mind for Business

Grace Merkler, NJFX 2026 Summer Marketing Intern

Grace is a junior at the University of Richmond, pursuing a Business Administration degree with a concentration in Marketing and a minor in Health Studies. A Monmouth County, NJ resident and a Dean's List student, she balances strong academic performance with hands-on professional experience.

Grace's passion for innovation and technology took root at Trinity Hall, where she spent four years studying engineering, developing a strong foundation in analytical thinking and a genuine curiosity for how technology solves real-world problems. That early exposure shaped how she approaches challenges, and inspired her to become a founder of the HUMM Institute at Trinity Hall, an entrepreneurial program centered on educating young women through experiential learning, using collaboration, design thinking, and a human-centered approach to inspire innovators.

Academic Focus

Business Administration with a concentration in Marketing and a minor in Health Studies, bridging her technical background with a future in sales.

Campus Involvement

Member of Kappa Kappa Gamma, where she holds a leadership position in recruitment, and an active member of the Sales Club at Richmond.

Experience

Building AI Agents from the Ground Up

At NJFX, Grace's role spans the full lifecycle of AI agent development: creating, testing, and designing agents that take on tedious, repetitive workloads and free up the team's capacity for higher-value work. Her responsibilities also include implementing immersive technology solutions, working within HubSpot to align automation with existing marketing and CRM operations, and integrating Microsoft Copilot across team workflows.

She also delivers internal demos to drive team-wide adoption of new tools, a responsibility that requires translating complex technology into something the whole team can put to use. Throughout, Grace has stayed committed to keeping up with changing technologies, making sure every project she takes on reflects current, relevant thinking.

This Summer

What Grace Is Looking to Learn at NJFX

This summer at NJFX, Grace is building toward a career in medical device and general technology sales, combining her interest in health and technology with hands-on experience in sales-adjacent, tech-forward environments. Her work here is giving her direct exposure to applied AI strategy, cross-platform integration, and technical enablement.

  • How subsea connectivity, cloud infrastructure, and global networks come together to power the digital economy
  • How AI agents can be designed and deployed to reduce repetitive workload and free up teams for higher-value work
  • How to translate complex tools like HubSpot and Microsoft Copilot into practical, team-wide adoption
  • How immersive technology fits into a corporate marketing and sales strategy
Looking Ahead

The Start of a Strong Industry Career

Along the way, Grace has developed a working understanding of how subsea cables physically link continents and carry the vast majority of international data traffic, how cloud infrastructure provides scalable computing power, and how global networks keep data moving reliably between providers and end users. That broader view has given her a deeper appreciation for the role companies like NJFX play in keeping the world connected.

We are excited to have Grace join the team this summer and look forward to supporting her professional growth throughout her internship journey at NJFX. As the digital infrastructure industry continues to accelerate, driven by the rise of AI, expanding subsea cable systems, and growing enterprise demand for carrier-neutral interconnection, there has never been a more exciting time to enter the field. We are glad to have Grace at the table for it.

NJ
NJFX Editorial Team

Published by NJFX, North America's first carrier-neutral Cable Landing Station campus, Wall Township, New Jersey.

Related Reading

Careers at NJFX Leadership Team Rethinking the Role of the CLS The Fiber Exchange NJFX Newsroom

Welcome Grace Merkler: Powering NJFX CLS with AI Read More ยป

Navigating Your Subsea Network Through Tense Scenarios

NJFX Blog /
Navigating Your Subsea Network Through Tense Scenarios
Subsea Infrastructure

Navigating Your Subsea Network Through Tense Scenarios

Ongoing tensions around the Strait of Hormuz have put a spotlight on a risk the industry has long understood: too much of the world's data still moves through too few narrow passages.

NJFX Insights Subsea Cables 8 min read
2
Major chokepoints, Hormuz and the Red Sea, have faced simultaneous disruption in 2026, a first for the industry
40+
Days is the typical minimum for a cable repair ship to reach, fix, and rebury a single fault under normal conditions
63
Cable repair ships operate worldwide, with only a handful assigned to the entire Middle East region

Subsea cables carry the overwhelming majority of the world's internet, financial, and cloud traffic. Most people never think about them, because the system is built to be invisible. Tension in the Strait of Hormuz is testing that assumption in real time.

Iranian officials and state-linked media have repeatedly signaled intent to assert control over cables transiting the Strait, including proposals to charge transit fees to the tech companies that depend on them. At the same time, ongoing hostilities have made the Strait and the nearby Red Sea largely inaccessible to the specialized repair vessels that fix cable faults when they happen. Faults that would normally take weeks to resolve are now facing indefinite delays.

This matters beyond the Gulf. Cables through Hormuz and the Red Sea connect Europe, the Middle East, and Asia, and they underpin the data center and AI buildout underway across the region. When a chokepoint closes, or even just becomes unpredictable, traffic has to find another way through. The question every network operator should be asking is whether their traffic has one.

Redundancy stopped being a theoretical best practice the moment two chokepoints were disrupted at once. It is now an operational requirement.

This is a pattern, not an isolated incident

Hormuz is the latest flashpoint, but it isn't the first, and it won't be the last. Subsea cables have absorbed a steady run of disruptions over the past two years, from deliberate targeting to anchor drags to natural events. A timeline of the major incidents since early 2024:

Feb 2024

Red Sea cables cut near Yemen

An anchor drag severed cables carrying Seacom/TGN-EA, EIG, and AAE-1 traffic, an early sign of how exposed the Red Sea corridor had become amid regional conflict.

Mar 2024

West Africa rockslide outage

A suspected underwater rockslide off Cรดte d'Ivoire damaged four cables converging at a single point (WACS, ACE, MainOne, SAT-3), knocking out or degrading internet service across 13 West and Central African countries.

May 2024

East Africa cable damage

Damage to the SEACOM and EASSy cables off the South African coast disrupted connectivity across Kenya, Tanzania, Uganda, Rwanda, Malawi, and Mozambique.

Nov 2024

Baltic Sea: BCS East-West & C-Lion1

Two cables linking Finland, Germany, and Lithuania were disrupted within a day of each other. A Chinese-flagged vessel under scrutiny at the time prompted NATO and EU accusations of hybrid warfare.

Dec 2024 โ€“ Jan 2025

Further Baltic incidents

The Estlink 2 power cable and a Latvia-Sweden data cable were both damaged within weeks of each other, continuing a pattern that led NATO to launch its Baltic Sentry patrol program in 2025.

2024โ€“2025

Cable damage near Taiwan

Five separate incidents damaged cables in waters around Taiwan, several involving vessels linked to China, adding a second geopolitical hotspot to the list of regions where cable security and territorial tension now overlap.

Sept 2025

Red Sea cable cluster severed again

A second major cut in the Red Sea slowed cloud services and cross-border financial transactions across Asia, the Middle East, and Europe. Limited regional repair access extended the disruption well beyond a typical fault window.

Dec 2025 โ€“ Jan 2026

Fitburg incident & Arelion cable damage

Finnish special forces boarded a vessel suspected of dragging its anchor across a Helsinkiโ€“Tallinn cable on the last day of 2025. Days later, Arelion's BCS East cable connecting Latvia and Lithuania was also damaged.

Feb 2026

EU responds with a โ‚ฌ347M package

The European Commission introduced a Cable Security Toolbox and a โ‚ฌ347 million subsea infrastructure initiative, including a Rapid Repair Pilot for the Baltic, the largest EU-level investment in cable protection to date.

Globally, the median time to repair a single subsea cable fault runs around 40 days under normal conditions, and considerably longer when repair ships can't safely reach the site. Regions with limited redundancy and thin repair capacity, including parts of West Africa, the Pacific, and now the Gulf, are the ones most exposed when a fault occurs.

How networks navigate through it

None of this is a reason to panic, but it is a reason to plan. Networks that weather these disruptions well tend to share a few habits in common:

1

Build in route diversity upfront

Contract capacity across more than one subsea corridor so a single strait, sea, or region isn't the only way traffic can move.

2

Spread risk across cable systems

Use more than one cable system and operator on a given route. Cables that share a corridor can still share risk from the same event.

3

Land in more than one place

Connect through multiple landing stations and geographies so an incident at one location doesn't isolate a region's traffic entirely.

4

Keep a terrestrial backup path

Pair subsea routes with terrestrial fiber options where possible, giving traffic an alternate path when ocean routes are compromised.

5

Monitor and reroute in real time

Maintain visibility into cable and route health so traffic engineering decisions happen before an outage cascades, not after.

6

Work from a carrier-neutral facility

Operate out of a facility with access to multiple carriers and cable systems, so shifting between them during a disruption is a configuration change, not a renegotiation.

A single-path network has a single point of failure

Europe
Origin
Red Sea
Contested
Strait of Hormuz
Contested
Gulf & South Asia
Destination

Two of the busiest connectivity corridors on earth run through the same two narrow passages. Diversifying the physical route, and the landing infrastructure on either end, is what keeps traffic moving when one of them is disrupted.

What "diverse connectivity" actually means

Diversity isn't a single feature you can buy off a shelf. It's a set of independent choices, stacked together, so that no single event, geopolitical, physical, or operational, can take a network fully offline.

Route

Multiple physical paths

Connectivity across more than one subsea corridor, and increasingly, terrestrial backup routes, so a single strait or sea isn't the only way through.

Cable System

Multiple cable systems

Access to more than one cable operator and system on a given route, since even parallel cables in the same waters can share risk from the same event.

Landing Station

Multiple landing points

Cables that come ashore in more than one location, so a disruption at a single landing station doesn't isolate a region's traffic.

Carrier

Carrier neutrality

Facilities that let a business connect to many carriers and networks in one place, instead of being locked into whichever single provider happens to reach them.

Where NJFX fits

NJFX operates a carrier-neutral cable landing station and Tier 3 data center on the New Jersey coast, one of the connectivity gateways between the U.S. and the rest of the world. That geography isn't a marketing point, it's physics: subsea cables land where the coastline allows them to, and that determines where diversity is actually possible to build.

Because NJFX is carrier-neutral, customers aren't limited to one cable system or one path across the ocean. They can connect to multiple subsea systems and terrestrial networks from a single facility, building the kind of route and carrier diversity that current events are making a practical necessity rather than a nice-to-have.

This isn't a new conversation for NJFX. Two years ago, NJFX hosted a Critical Infrastructure Forum session focused specifically on educating enterprise and financial markets on navigating their networks through geopolitical tension, and on what diverse connectivity solutions actually look like in practice. NJFX has continued to hold these forums as new trends and new conflicts have emerged, most recently with a SecureWAVES CIF built around the intersection of subsea infrastructure and AI. The throughline across all of them is the same: geopolitical risk to global connectivity isn't hypothetical, and the enterprises and financial institutions that depend on that connectivity need a working understanding of where their exposure sits.

None of this makes any single region immune to disruption. What it does is give network operators options when disruption happens elsewhere, which is precisely the point.

The lesson isn't new. It's just harder to ignore.

The industry has known for years that too much traffic depends on too few chokepoints. What's changed is that two of the busiest ones are now under strain at the same time, turning a known risk into an active one. Building diverse, carrier-neutral connectivity isn't a response to any single conflict. It's the ongoing work of making sure the network doesn't depend on any one of them.

Want to talk through what route and carrier diversity could look like for your network?

Get in touch

Navigating Your Subsea Network Through Tense Scenarios Read More ยป

Training to Inference: The Infrastructure Shift Reshaping AI

NJFX Blog /
Training to Inference: The Infrastructure Shift Reshaping AI | NJFX
NJFX Perspective  ยท  AI Infrastructure

Training to Inference: The Infrastructure Shift Reshaping AI

Across the industry, the AI conversation is moving from how models get built to how they get delivered at scale. That shift changes what infrastructure has to do.

NJFX EditorialยทJuly 2026ยท5 min read

For the past two years, the AI conversation centered on building bigger models. Headlines followed GPU clusters, trillion-parameter models, and hyperscale training campuses.

That conversation is changing. Across research firms, colocation providers, and infrastructure investors, the language is converging on one idea: training built the AI industry, but inference is what will run it going forward.

Inference is projected to account for roughly two-thirds of AI compute in 2026, up from about a third in 2023.

Deloitte, Technology, Media & Telecommunications Predictions 2026

That shift means every deployed model becomes a continuous, always-on workload rather than a one-time training run โ€” every prompt, recommendation, transaction, and automated decision it powers, delivered in real time.

The question is no longer just how AI gets built. It's how AI gets delivered, at scale, every day.

The Shift

Training Built AI. Inference Delivers It.

Training happens once, or periodically, when a model is built or updated. Inference happens continuously โ€” billions of times a day, every time that model is actually used. Understanding the shift starts with what each phase actually does.

Training

A model learns patterns from massive datasets by adjusting billions of internal parameters over many repeated passes. It's a batch process: computationally enormous, but finite. Once a model is trained, that phase is done until the next update or retrain.

Inference

The trained model is put to work: given a new input (a prompt, an image, a sensor reading), it produces an output. Unlike training, inference never really stops. It runs every time someone or something uses the model, at whatever scale that demand reaches.

For years, that distinction made inference look like the smaller job. Training was the expensive part; inference was assumed to be comparatively cheap and lightweight, run one prompt at a time. That assumption no longer holds.

โœ“

Every AI chat prompt

โœ“

Every AI-powered search result

โœ“

Every product recommendation

โœ“

Every fraud detection check

โœ“

Every autonomous vehicle decision

Two things changed. First, adoption: AI usage reached a majority of consumers within about three years, faster than personal computers or the internet, and enterprise adoption has moved even quicker. Second, inference itself got more sophisticated. Newer models use techniques like step-by-step reasoning at query time, working through a problem before answering rather than producing a single immediate response. That reasoning step can use 30 to 100 times the compute of a simple, single-pass inference query. Multiply that by billions of daily requests, and inference stops looking like the lightweight half of AI.

The scale of that shift keeps surprising forecasters. Total AI compute demand is now projected to keep growing four to five times a year through 2030, outpacing the efficiency gains from newer chips.

Where the Network Comes In

Connectivity Becomes Part of the Compute Stack

Training and inference don't just need different amounts of compute. They need infrastructure in different places. Training workloads scale up: large GPU clusters concentrated in a handful of core locations built for raw throughput. Inference workloads scale out: smaller deployments spread across many locations, positioned close to data sources and users because latency is the primary constraint.

โšก

Low Latency

Millisecond round-trips between user and model

๐Ÿ”—

Resilient Paths

Network routes that don't fail when one path does

๐ŸŒ

Global Reach

Connectivity that spans users wherever they are

๐Ÿ”Œ

Reliable Power

Continuous power behind a continuous workload

The GPU is only one component. Without the network to carry results back to the user, AI can't deliver an answer fast enough for it to matter.

Not Just One Company's View

The Industry Is Saying the Same Thing

This isn't one company's talking point. Analysts, colocation providers, and infrastructure investors are all describing the same shift, from different vantage points.

Deloitte

Enterprise data centers built for the pre-AI era are increasingly misaligned with inference's latency, cost, and resilience demands, pushing organizations toward hybrid infrastructure built workload by workload.

Equinix

As inference becomes the dominant enterprise AI workload, scaling out distributed infrastructure across locations is becoming as critical as scaling up raw compute capacity.

Mayfield (VC)

As AI moves from pilots to production, the questions shift from whether AI works to whether it's financially viable at scale โ€” and inference economics are where that answer gets decided.

A Return to Geography

Why Location Matters Again

AI doesn't have to live in one hyperscale campus. Training tends to concentrate in a handful of massive sites, wherever land and power are cheapest. Inference works differently: it becomes more valuable the closer it sits to the people and systems using it, which is pushing workloads to spread out.

Power availability, cooling efficiency, and energy cost increasingly determine how fast, and whether, inference capacity can scale in a given location โ€” alongside the connectivity to actually reach the users the workload is meant to serve.

FactorTrainingInference
FrequencyPeriodic, batch runsContinuous, real-time
Latency toleranceMinutes to hoursMilliseconds
Scaling approachScale up: concentrated GPU clustersScale out: distributed regional capacity
Ideal locationWherever power and land are cheapestClose to users and data sources
Infrastructure priorityRaw compute densityNetwork connectivity and proximity
In Context

Where NJFX Fits

As AI inference expands globally, digital infrastructure has to evolve alongside it. Carrier-neutral ecosystems that connect international subsea systems, terrestrial fiber providers, cloud platforms, and enterprise networks are becoming increasingly important in reducing latency, improving resilience, and enabling global AI applications.

The inference era isn't defined by who owns the most GPUs.

It will be defined by who can move data faster, connect users more efficiently, and deliver AI where it's needed most.

Every AI response begins long before the GPU. It begins with the network.

Sources

  1. Deloitte โ€” Why AI's Next Phase Will Likely Demand More Computational Power, Not Less โ€” TMT Predictions 2026, inference share of compute
  2. Deloitte โ€” The AI Infrastructure Reckoning: Optimizing Compute Strategy in the Age of Inference Economics โ€” Tech Trends 2026
  3. Equinix โ€” AI Infrastructure: Scale Up or Scale Out? โ€” training vs. inference infrastructure placement
  4. Mayfield โ€” Inference Is the New Battleground for Scaling AI โ€” inference economics and infrastructure investment
  5. Vast.ai โ€” The Future of AI Inference in 2026: Key Trends Shaping AI Infrastructure โ€” compute demand growth and adoption data

Explore NJFX's Connectivity Ecosystem

See how NJFX's carrier-neutral subsea and terrestrial network connects to the broader AI infrastructure landscape.

Explore Subsea Page Get in Touch

Training to Inference: The Infrastructure Shift Reshaping AI Read More ยป

Hello!

Login to your account

small_c_popup.png

Let's have a chat

Learn how we helped 100 top brands gain success.