Table of Contents
Undersea cables — the 1.3-million-kilometer grid of fiber on the ocean floor that carries roughly 95% of all international data — just became the AI industry’s next infrastructure frontier, and the moves are measurable: Meta’s Petal becomes the first petabit-capacity transoceanic cable (7,000 km from the US to France, delivering about a million gigabits per second — roughly 125 terabytes every second), Microsoft’s consortium announced I-2SEA, a 3,600-kilometer cable engineered for AI workloads linking India to Malaysia and Singapore by late 2029, and both land in a market where India’s data-center capacity is projected to quintuple to about 7 gigawatts by 2030. This guide explains why the AI buildout is buying the ocean floor: what petabit fiber actually changes, who owns which routes now, the strategic logic of hyperscalers becoming cable operators, and what it means for Southeast Asia — including the Philippine position on the map.
All figures are dated and attributed throughout — capacity numbers to the cable operators’ own announcements, market projections to the named research houses — because this is a story where the plumbing is verifiable even when the marketing is not.
Key Takeaway
- 🌊 The floor of the ocean is the AI industry’s next bottleneck: about 95% of international data traffic rides undersea fiber; satellite capacity cannot match it, and the AI buildout’s cross-campus data flows are rewriting the demand curve.
- 🚀 Petal is a generational jump: the first petabit-class transoceanic system (US-France, ~7,000 km, online 2029) — built on Sumitomo’s dual-core fiber that doubles capacity without doubling power, the largest single capacity increase in undersea history.
- 🗺️ AI-named routes are the new lane: I-2SEA (Microsoft, Singtel, Tata Communications, NEC and partners; 3,600 km India→Malaysia/Singapore, service from Q4 2029) is engineered explicitly for AI and hyperscale traffic — India’s 1.4→7 GW capacity surge needs it.
- 🧭 Ownership is the strategy: Meta and Microsoft now build cables themselves (Petal; the Waterworth 50,000-km ring; I-2SEA) because owning the pipe controls timing, maintenance, capacity priority and cost for AI traffic no carrier can guarantee.
- 🇵🇭 The Philippine position: the country sits on the corridor between Asia’s two fastest-growing AI markets — and every regional route decision (landing stations, transit capacity, pricing peering) is a question for PLDT-class carriers, regulators and the OFW telecom workforce that builds and maintains the landing infrastructure.
What undersea cables are (and why AI strains them)
AI’s undersea cables are bundles of glass fiber strands, thinner than human hair, carrying data as pulses of light across thousands of kilometers; optical repeaters boost the signal every 50-100 km, and wavelength-division multiplexing packs multiple terabits per second into a single fiber pair. About 95% of international data traffic travels over these systems — video calls, market trades, streaming, backups, everything — a dependence TeleGeography’s research leadership states flatly in every cable-market briefing. Satellites cannot compete: the capacity class is orders of magnitude apart, and latency physics favors fiber.
AI strains the grid differently than the human internet did. Training a frontier model shuttles enormous datasets between data centers (checkpoint sync, distributed training across regions); inference at global scale replicates models across campuses; and the hyperscalers’ internal “cloud region to AI campus” flows now move petabyte-scale sets on schedules. The old internet was content delivered to people; the AI era is weights, datasets and state synchronized between machines — traffic that concentrates on corridors between a handful of giant campuses. That is why the new cables get AI names on the specification sheet: they are built for machine traffic patterns, not human browsing.
Petal: the first petabit cable, and what dual-core fiber changes
Meta’s Petal announcement (September 2026) set the new bar for undersea cables: the first subsea system designed for petabit capacity, spanning about 7,000 kilometers between the United States and France, built with NEC Corporation and Sumitomo Electric Industries, and expected online in 2029. The enabling technology is Sumitomo’s specialized dual-core fiber: each strand carries two cores with optical signals traveling in opposite directions, performing with crosstalk eliminated to near-measurement — one physical strand delivering what conventionally requires a pair of single-core fibers, with no proportional increase in power. NEC engineering leadership called achieving petabit-per-second across a transoceanic system “a major technological milestone in the history of global telecommunications,” and Meta’s own statement framed the scale: double the capacity of today’s most advanced transoceanic cables — “the single largest generational increase in capacity of any transoceanic subsea cable system, ever.”
The numbers make the scale vivid: at up to a million gigabits per second, Petal moves roughly 125 terabytes every second — about 25,000 five-gigabyte HD movies each second, or, in Meta’s framing, enough for 75% of the world’s population to stream music simultaneously. But the AI logic matters more than the party trick: petabit-class links halve the effective latency-weighted cost of syncing training state between continents, and Petal is a piece of Meta’s multi-billion-dollar infrastructure push — the same program that includes Project Waterworth, the planned 50,000-kilometer ring connecting five continents that would be the longest subsea system ever built.
I-2SEA and the India-Southeast Asia AI corridor
The second class of AI undersea cables is regional and AI-named. The I-2SEA system, announced July 2, 2026 by a consortium of Microsoft, Singtel, Tata Communications, Lightstorm (backed by infrastructure investor I Squared), Singapore’s ASEAN Cableship and Japan’s NEC, spans 3,600 kilometers from a landing station in Machilipatnam (Andhra Pradesh, India) through a second Indian landing at South Chennai to Singapore and Kuala Lumpur — with the Chennai-Singapore leg targeted for Ready-for-Service in the fourth quarter of 2029. It exists for a measurable demand wave: Macquarie Equity Research projects India’s operational data-center capacity rising from about 1.4 gigawatts today to roughly 2.8 GW by 2027 (on projects already under construction) and toward 7 gigawatts by 2030 if planned projects land — a five-fold jump served today by just 17 active submarine cables with about 960 terabits of maximum potential capacity, with at least 10 more cable systems publicly announced.
The consortium’s stated purpose is explicit: I-2SEA will connect to Lightstorm’s AI fabric for low-jitter, loss-optimized transport across data centers and cloud hubs, expanding the company’s connected AI and cloud zones in India from 19 to 29. The landing points tell the strategic story — Machilipatnam sits where Meta and Alphabet have announced data centers — and Lightstorm’s chief executive framed the system as engineered for AI and hyperscale workloads. This is the India lane of a pattern our India compute analysis and SEA infrastructure scoreboard track: capacity, power, land and now ocean fiber move together.
Why hyperscalers are becoming cable operators
The pattern behind Petal and I-2SEA is ownership of the undersea cables, and it has three legs. Timing: carrier-operated undersea cables install AI capacity when the carrier’s roadmap says; an owned cable delivers when the AI roadmap needs it (deployment timing, maintenance windows and upgrade priorities all move in-house). Cost at scale: the AI traffic between owned campuses is intercompany — owning the pipe prices it at construction amortization instead of retail transit, and the 2030s traffic curves make that arbitrage large. Control: reroute authority, repair priority, capacity partitioning — the operational levers that determine whether a training run finishes tonight or next week. The counterside deserves its own honesty: private cables concentrate choke-point risk (two hyperscalers now own pieces of the world’s critical transit), raise regulator questions about net-neutrality-adjacent capacity allocation, and create single-owner failure domains — the same concentration concerns the national compute discussion raises for chips and cloud. The world’s answer so far: consortium structures (I-2SEA’s six-party build) as the hedge against single-owner routes.
The map that matters: routes, chokepoints and the Southeast Asian position
Global geography gives the undersea cables map its pressure points: the Malacca Strait corridor (the shortest path between the Indian Ocean and East Asia’s campuses), the Luzon Strait and the South China Sea routes that connect the ASEAN metros, and the trans-Pacific and trans-Atlantic trunk lanes where hyperscalers concentrate capacity. For the Philippines, the position is structural: the archipelago sits directly on the corridor between Southeast Asia’s two growth poles (Singapore-Johor to the south, Taiwan/Hong Kong/Japan to the north), and regional route planning — new Malaysia-Singapore-India systems like I-2SEA, petabit trans-Pacific builds — will price Philippine landing, transit and peering for a decade. The local decisions that follow: which new systems carry Philippine landing stations, how the carriers’ AI-traffic pricing tiers get set, and how quickly the country’s data-center corridors (our water-and-power analysis) connect to international capacity at regional-competitive rates. For OFW engineers and tech workers, the hiring side is real: cable landing engineering, subsea network operations and data-center interconnection are the skilled trades the buildout employs — most concentrated in the countries that own the landing infrastructure.
WorldNgayon Insight: who owns the pipe owns the priority
What changed: undersea cables moved from telecom procurement to hyperscaler capital projects — the world’s first petabit cable and the first AI-named regional systems moved undersea capacity from telecom procurement to hyperscaler capital projects — the ocean floor joined the compute buildout sheet. Why now: machine-to-machine traffic (weights, checkpoints, replicated inference) outgrew carrier planning, and dual-core fiber made a capacity doubling affordable within the same power budget. Why it lasts beyond the news cycle: cables take 3-5 years from announcement to service (Petal and I-2SEA both 2029), so every announced system is a dated future event — named update triggers through decade’s end. Who is affected: carriers and their regulators, hyperscalers building private transit, countries pricing landing rights — and every AI team whose cross-region training times depend on route capacity. What to do next: read the cable announcements as capacity futures — who lands where, who owns the fiber, and which corridors the AI campuses will load. Risks, opportunities, costs, skills: the risk is choke-point concentration in private hands; the opportunity is regional compute markets finally unconstrained by transit; the cost is measured in multi-billion multi-year builds; the skills gap runs through subsea network engineering and interconnection design. Connection to work and money: the cable layer is the bottom rung of the sovereign AI stack — and the first one nations can share without surrendering sovereignty, which is why the consortium shape keeps winning.
What are undersea cables and how much traffic do they carry?
Bundles of glass fiber on the ocean floor carrying data as light across thousands of kilometers, boosted by repeaters every 50-100 km. Roughly 95% of all international data traffic travels over them; satellite capacity is orders of magnitude smaller and high-latency, so the cables are effectively the internet’s only intercontinental transport.
What is the Petal undersea cable?
The world’s first petabit-capacity subsea cable: Meta’s ~7,000-kilometer US-France system (built with NEC and Sumitomo Electric, online 2029), delivering about a million gigabits per second — double today’s most advanced transoceanic capacity — using first-of-its-kind dual-core fiber that doubles throughput without doubling power.
Why is Meta building its own undersea cables?
Control and cost: Meta’s AI infrastructure moves petabyte-scale datasets between its own campuses, and owning the cable sets deployment timing, maintenance priority, capacity partitioning and intercompany pricing without carrier dependence — part of a multi-billion push that also includes the planned 50,000-kilometer Waterworth ring across five continents.
What is the I-2SEA cable?
A 3,600-kilometer AI-workload submarine cable from India to Malaysia and Singapore, announced July 2, 2026 by a Microsoft, Lightstorm, Singtel, Tata Communications, ASEAN Cableship and NEC consortium, landing at Machilipatnam and South Chennai, with the Chennai-Singapore leg ready for service targeted for Q4 2029 — built explicitly for AI, cloud and hyperscale traffic.
How fast is India’s data-center capacity growing?
Macquarie Equity Research projects India’s operational capacity rising from about 1.4 gigawatts to roughly 2.8 GW by 2027 on projects under construction, and toward 7 GW by 2030 if planned projects proceed — a five-fold expansion that India’s 17 existing submarine cables (about 960 terabits potential) cannot serve alone, hence at least 10 announced new systems.
Can satellites replace undersea cables for AI traffic?
No — the capacity class is orders of magnitude apart, latency is physics-unfavorable, and AI workloads concentrate petabyte-scale transfers on fixed corridors; satellites serve remote and backup niches. The cables’ ~95% share of international traffic reflects what the physics allows, which is why AI-era cable buildout is accelerating rather than slowing.
What are the chokepoints in the undersea cable network?
The geographic narrows where many systems converge: the Malacca Strait between the Indian Ocean and East Asia, the Red Sea/Egypt corridor between Europe and Asia, and key trans-Atlantic and trans-Pacific landing regions — plus the concentrated landing stations and cable stations themselves. Chokepoint concentration is why consortium ownership and reroute capacity matter strategically.
How do cable projects affect Southeast Asia and the Philippines?
The AI era’s routes price the region’s landing stations, transit and peering for a decade; the Philippines sits on the corridor between Southeast Asia’s growth poles, so its local decisions (which new systems land, how AI-traffic tiers price, how fast data-center corridors connect to international capacity) follow directly from the Meta-, Microsoft- and India-led announcements now on the table.
What jobs does the AI undersea cables buildout create?
Landing-station engineering, subsea network operations and maintenance, interconnection and peering design, and the data-center-side network engineering the cables feed — plus the marine and survey trades the cable ships employ. The work concentrates where landing infrastructure lives, which makes regional buildouts a regional jobs map.
Financial Disclaimer
This article is editorial analysis of publicly reported prices, contracts and market data, not investment, financial or procurement advice, and no outcome is promised. Posted rates, contract discounts and market projections change; verify current terms with the named providers and institutions. Compute, technology and capacity markets carry substantial risk.
Sources: Livescience — Petal’s engineering and scale (dual-core fiber, 1M Gbps, the Meta/NEC statements); Reuters — the I-2SEA consortium announcement (members, landings, the 2029 target, TeleGeography’s traffic and capacity figures); Indoneo — India’s capacity trajectory (Macquarie projections) and the cable’s strategic read; Inside AI News — Meta’s Petal announcement and the Waterworth context. Company capacity figures are the operators’ own announcements; market projections are the named research houses’. Landing-station and peering implications for specific Philippine routes reflect public corridor planning, not announcements of new Philippine systems.
