Apricot

APRICOT is a hyperscaler submarine cable lit for service in 2025, connecting eight landings across five countries and one US territory — Minamiboso in Japan, Toucheng in Taiwan, Tuas in Singapore, Batam and Tanjung Pakis in Indonesia, Agat in Guam, Baler and Davao in the Philippines. Co-owned by Google, Meta, NTT, PLDT, and Chunghwa Telecom, APRICOT is a recent entrant in the Asian intra-regional mesh. Its design spec is unusually modern: 12 fibre pairs running a combined 290 Tbps across the system.

APRICOT is one of the first of a new generation of Asian cables built primarily by hyperscalers rather than by telecom consortia. That has two practical consequences for how it gets used — and, as our measurements show, for the latency personality that emerges in its first year of operation.

Measured RTT: a stable ~86 ms

Our monitor measures APRICOT on the Japan↔Indonesia leg — between Minamiboso and Tanjung Pakis. Recent samples against a real Indonesian target give the following picture:

The Tanjung Pakis → Minamiboso samples are remarkably tight — a 0.6 ms spread across five measurements over three days. Forward-direction data is thinner (only one post-switch sample so far), but it lands within half a millisecond of the reverse. Both directions use 16 hops, suggesting the same physical path.

The great-circle distance from Minamiboso to Tanjung Pakis is about 5,600 km. Light in submarine fibre (with a refractive index of ~1.467) makes that round-trip in a theoretical minimum of 54.8 ms. We measure 85.8 ms — that is 1.57× the physics floor. For comparison, that sits between EIG's 1.13× (15,000 km, closer to straight) and Marea's 1.95× (Atlantic). 1.57× is a solid number for a first-year Pacific-Asia cable.

The 85.8 ms figure will be the baseline for future measurements. If it drifts upward, something has changed on the carriers' routing policy. If it stays flat for the next decade, APRICOT is doing exactly what a Japan-Indonesia fibre pair is supposed to do.

290 Tbps on 12 pairs

APRICOT's 290 Tbps design spec works out to roughly 24 Tbps per fibre pair. Using current-generation coherent transponders, that is about 80 wavelengths at 300 Gbps each — close to the practical ceiling for a C-band submarine system. Extended C+L band, if retrofitted at the landing stations, could push per-pair capacity past 40 Tbps without laying an additional metre of fibre.

For comparison: EIG (2011) was designed for 3.84 Tbps on two pairs — 1.92 Tbps per pair. APRICOT's per-pair capacity is 12.6× higher than EIG's original design. Most of that gain did not come from the glass, which has improved only modestly in fifteen years. It came from the electronics at each end: coherent modulation, forward error correction, high-order QAM constellations, and multi-symbol signalling collectively pushed per-wavelength capacity from ~10 Gbps in 2011 to 300–400 Gbps in 2025. Submarine cables age gracefully because the parts that matter are on dry land and can be upgraded without touching the ocean.

Eight landings, five countries

The topology is useful to read. APRICOT connects the major intra-Asian internet economies (Japan, Taiwan, Philippines, Singapore, Indonesia) and uses Guam as a hub island. Tuas in Singapore and Toucheng in Taiwan are already among the most-landed cable stations in Asia — APRICOT plugs into existing high-capacity ecosystems. Baler in the Philippines and Tanjung Pakis in Indonesia are less-crowded stations, which means APRICOT's fibre pairs have more end-to-end exclusivity for their dedicated owners at those points.

The Guam landing at Agat is worth noticing. Guam has become the second-most-landed island territory in Pacific submarine cables after Hawaii, with four major systems adding landings in 2024–2025. Small islands that happen to sit on useful great-circle routes are being upgraded from relay stations to first-class content cache locations — APRICOT's Agat landing is part of that broader trend.

Hyperscaler traffic on day one

Unlike traditional carrier consortium cables — where dozens of telecoms pool their capital and then slowly migrate production traffic onto the new system over 12–24 months — a hyperscaler-sponsored cable like APRICOT has its biggest customers on the inside. Google, Meta, and NTT each hold dedicated fibre pairs. Their own traffic — cloud backend replication, content delivery to edge caches, interactive session flows between data centres — moves onto the cable the day it is lit.

That changes the early-life latency profile in two ways. First, the cable is used at meaningful volume from day one, so commercial routing decisions on each side of the Pacific have real traffic to optimise. Second, the owners are motivated to route traffic symmetrically through their own infrastructure — carrying both ingress and egress across the same fibre pairs — rather than tromboning the return through a cheaper competitor's cable. That is why our Japan-to-Indonesia and Indonesia-to-Japan numbers are both close to 86 ms rather than one direction being twice the other.

Contrast with an older carrier-consortium cable like EIG, where Mumbai-to-Sesimbra runs at physics floor through the cable but the reverse path takes the long way around the world via Asian transit. EIG's asymmetry is economic — Indian operators get a better commercial deal on the Asia-via-Singapore route than on EIG's direct westbound path. APRICOT's symmetry is also economic, just in the other direction: its owners are making the routing decisions and they want the fastest path, not the cheapest.

Why the new Asian mesh matters

APRICOT joins BIFROST, Echo, and a handful of other recent systems in building out a denser Asian intra-regional internet. For the last two decades, the bulk of Asia-to-Asia traffic relied on legacy trunks like APG and SJC (commissioned 2013 and 2017 respectively), with latencies determined by where those systems landed. The new generation of cables lands in more places and runs more fibre pairs per cable, giving regional operators shorter and more diverse paths between the same source-destination pairs they have always needed to serve.

For a traffic engineer at a Singapore-based streaming service, for example, the practical outcome is that Indonesian users can be served from cache instances in Jakarta over a shorter path with more redundant capacity. For a Japanese cloud operator, Southeast Asian replicas can sync backend state over a latency budget that used to be reserved for North America or Europe. APRICOT is one piece of an infrastructure re-layout that the end user never sees but that measurably reshapes what is possible at the application level.

What our data proves

• APRICOT is delivering ~86 ms Japan↔Indonesia round-trip. Five recent samples on the reverse direction land in a 0.6 ms window — stable physics in the first year of service.
• Both directions use the same 16-hop path. Consistent with hyperscaler owners routing their own traffic symmetrically across their own fibre pairs.
• 1.57× physics floor for the 5,600 km link is typical for a new Pacific-Asia cable — between EIG's 1.13× (established, near-straight path) and Marea's 1.95× (Atlantic).

Try it yourself

Live data on the APRICOT cable page. For contrast see JUPITER (trans-Pacific hyperscaler with dramatic 2× asymmetry), BIFROST (southern-corridor Asia-to-Americas), and EIG (2011 consortium, 12.6× less per-pair capacity).