Why Subsea Cable Capacity Is Data Centers’ Next Bottleneck

On September 21, 2026, Meta announced it would build the world’s first petabit-class transoceanic subsea cable. Called Petal, the 7,000-kilometer system will run from the U.S. to France’s Atlantic coast, carrying 1 petabit per second — roughly double the capacity of today’s best transatlantic cables — and won’t enter service until 2029. It’s an eye-catching engineering milestone. But the more useful number for anyone planning data center capacity isn’t the petabit figure. It’s this: Google, Meta, Microsoft, and Amazon now consume close to 90% of trans-Atlantic submarine cable bandwidth and over 80% of trans-Pacific bandwidth, up from essentially zero in 2010. Power and land have dominated the capacity-planning conversation for the past two years. The ocean floor is quietly becoming the next constraint, and it’s one that far fewer buyers are tracking.

What Petal actually signals

Petal isn’t a one-off. Meta says it has now invested in more than 20 subsea cable systems worldwide, and it built the cable with NEC as systems integrator, Sumitomo Electric Industries producing the fiber, and Orange supporting the French landing. The headline technical achievement is multi-core fiber deployed at transoceanic distance for the first time — a way to push more capacity through a cable without a proportional increase in repeaters, power, or physical infrastructure. Meta’s own description calls it “the single largest generational increase in capacity of any transoceanic subsea cable system, ever.”

That framing matters because Petal is arriving alongside a wave of comparable projects. Meta’s Waterworth system spans 50,000 kilometers across the Atlantic and Indian Oceans. Africa-focused 2Africa runs 45,000 kilometers with 180 Tbit/s of capacity across 46 landing points. A newer system called Fastnet is targeting more than 320 Tbit/s ahead of a 2027–2028 completion. Industry investment in subsea systems is projected to hit roughly $13 billion over 2025–2027, close to double the 2022–2024 period, according to submarine cable market trackers. Subsea cables already carry an estimated 99% of intercontinental data traffic — satellite constellations remain a rounding error by comparison — so this is the literal physical layer that AI training runs, inference traffic, and cross-region replication all depend on.

Four companies, most of the ocean

The capacity buildout is real, but so is the concentration behind it. Google now holds stakes in more than 30 cable systems. Meta is involved in around 20. Microsoft and Amazon combined appear in roughly 10 more. Altogether, the four hyperscalers now have a stake in more than 60 of the roughly 600 cable systems operating or planned globally — a minority of cables that nonetheless carries the large majority of the traffic that matters for AI and cloud workloads. That’s a sharp reversal from fifteen years ago, when subsea cables were built and owned almost entirely by telecom consortia, with content and cloud companies renting capacity as tenants.

The manufacturing side is just as concentrated. Three companies — Alcatel Submarine Networks (France), SubCom (U.S.), and NEC (Japan) — control 92% of global cable-laying capacity between them, with China’s HMN Technologies at roughly 8%. That’s a narrow supplier base building the physical infrastructure that an even narrower set of buyers is reserving for itself years in advance. For anyone who has watched how a handful of GPU foundries and grid interconnection queues have already reshaped data center site selection, the pattern here should feel familiar: the resource that used to be commoditized and easy to lease is turning into a strategic asset that gets built and consumed vertically, by the same handful of companies, before it ever reaches an open market.

This is the same dynamic we’ve tracked in other parts of the supply chain — see our analysis of how $68 billion in data center projects got blocked in 2026 by bottlenecks buyers didn’t budget for. Network capacity is shaping up to be the same kind of blind spot.

Why this matters if you’re not a hyperscaler

If your organization buys colocation, leases wholesale capacity, or operates a multi-region SaaS platform, you’re not laying your own cable — but you’re still exposed to this dynamic in three concrete ways.

First, wholesale and IRU (indefeasible right of use) capacity on new routes is increasingly presold to the companies financing construction before a cable even reaches final investment decision. When Google, Meta, Microsoft, and Amazon are absorbing 80–90% of a route’s bandwidth, the spare capacity available to everyone else on that specific path shrinks, even as headline system capacity grows. A bigger cable doesn’t automatically mean more available capacity for a mid-market buyer; it can mean the same four tenants took a bigger slice and the market-facing remainder barely moved.

Second, latency and route diversity decisions are getting harder to make independently of hyperscaler infrastructure choices. If the newest, lowest-latency routes on a given corridor are hyperscaler-controlled and reserved primarily for internal traffic, a buyer optimizing for latency-sensitive workloads (real-time inference, financial trading, multiplayer or streaming platforms) may find their best options are indirect: leasing through a cloud provider’s network rather than through an independent carrier with its own diverse cable stakes.

Third, concentration is a resilience risk, not just a pricing one. Subsea cables get damaged more often than most buyers realize — anchor drags, fishing trawlers, and seismic activity account for the large majority of faults, and repair ships are a limited, globally shared fleet. When a small number of owners control most of the capacity on a given corridor, an outage or a deliberate act against a chokepoint (the Baltic Sea cable-cutting incidents of 2024–2025 are the reference case regulators now cite) has a proportionally larger effect on whoever’s traffic that owner happens to be carrying.

What it means for site selection and colocation strategy

The practical response isn’t to panic about subsea cables the way buyers panicked about GPU lead times in 2024. It’s to add network topology to the same due-diligence checklist you already use for power and grid interconnection. A few things worth asking any colocation or cloud provider right now: which specific subsea systems and landing stations does their backbone actually touch, and are those systems hyperscaler-majority-owned or carrier-neutral; what’s their route diversity on the corridors that matter for your workload, meaning do they have genuinely separate physical paths, not just separate contracts; and how do they handle capacity during a cable fault, given that repair times on damaged subsea systems can run from days to several months depending on ship availability and weather.

This is also starting to show up in site selection itself. Facilities near major cable landing stations — historically a nice-to-have for latency — are becoming more contested and more expensive, echoing what’s already happened with power-rich sites near substations. Markets like Northern Virginia, Marseille, Singapore, and Mombasa are seeing colocation and interconnection-focused development cluster specifically around landing points, not just around power availability. If your capacity plan for 2027–2029 assumes cheap, abundant network bandwidth as a given, it’s worth stress-testing that assumption the same way most buyers have already learned to stress-test power availability.

The counterargument worth taking seriously

None of this means hyperscaler-funded subsea investment is bad for the market. It’s genuinely true that Meta, Google, Microsoft, and Amazon are underwriting risk that traditional telecom consortia had grown unwilling to take on, and that new capacity — even majority-reserved capacity — eventually loosens pricing on older, fuller cables as traffic migrates to newer systems. Wholesale capacity prices on mature routes have fallen for over a decade even as hyperscaler ownership rose, and some of these systems do sell surplus capacity to carriers and enterprises once their own ramp-up period passes. The honest picture is not “hyperscalers are locking everyone else out of the ocean.” It’s that the buyers best positioned to benefit are the ones who understand the ownership structure well enough to negotiate for it, rather than assuming bandwidth is a commodity that shows up automatically wherever there’s power and rack space.

Subsea capacity used to be invisible to anyone outside network engineering — a utility you assumed was there. That’s no longer a safe assumption for buyers planning multi-year data center or colocation commitments in regions served by a small number of corridors. The teams that will handle this well are the ones who start asking their providers about cable ownership and route diversity with the same seriousness they now apply to grid interconnection timelines. Start by reviewing your own capacity and site-selection checklist against this risk with the TechInfraHub Data Center Buyer’s Toolkit.

Written by

Raajeev Ratra

Data Center Infrastructure Expert | 15+ Years in DC Design, Operations & Project Management

Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.

Connect on LinkedIn →