JUPITER

JUPITER is a trans-Pacific submarine cable with five landing points: Maruyama and Shima in Japan, Hermosa Beach and Cloverdale in the United States, and Daet in the Philippines. Ready for service in 2020, the 14,557 km system was built to carry hyperscaler traffic between North America and East Asia, with a branching unit extending the reach into Southeast Asia. It is one of the newest, densest fiber pipes across the Pacific.

What makes JUPITER interesting right now is not the marketing datasheet. It is what our measurements show when we ping across it in both directions — and the two numbers do not match at all.

A 200-millisecond mystery

Our monitor pings from a probe near the Hermosa Beach landing (Los Angeles, California) toward a Philippine target close to the Daet landing. Over the past 30 days we collected 60 clean samples. The minimum was 118.6 ms, the average 125.9 ms, and the distribution is almost flat — on most days every sample sits within a fraction of a millisecond of the minimum. The path traverses eight IP hops from probe to target. It is, by every measurable standard, the "textbook" behaviour of a well-engineered submarine route.

Then we ping the reverse direction — from a probe close to Daet toward a Hermosa Beach target. Eighteen clean samples came back with a minimum of 205.3 ms, an average of 304.9 ms, a maximum near 770 ms, and a standard deviation of 117 ms. The route takes twelve to sixteen hops. Same two landings, same physical ocean, nothing like the same latency.

Westbound (Hermosa Beach → Daet)

Eastbound (Daet → Hermosa Beach)

Physics floor: how fast can light possibly travel?

Glass slows light down. In a single-mode submarine fibre the group index is roughly 1.467, so the speed of a signal is about 204,500 km/s — about a third lower than vacuum. That lets us compute a hard lower bound on any ping that uses JUPITER as its only long-distance segment.

The full JUPITER system is 14,557 km, but Hermosa Beach to Daet does not traverse every kilometre — it uses the main Pacific trunk plus the Philippine branch, skipping the Cloverdale and Shima legs. Working backwards from our measurement, 118.6 ms of round-trip delay corresponds to roughly 12,130 km of glass. That is almost exactly what you would expect for "Hermosa Beach → branching unit → Daet": the great-circle distance across the Pacific plus a few thousand kilometres of spur.

In other words, the westbound minimum is the physics floor. There is no congestion budget, no routing detour, nothing that could make it faster without rewriting physics. JUPITER is carrying that traffic exactly the way it was designed to.

So where does the extra 200 ms go?

The eastbound picture is different. Twelve to sixteen hops is simply too many for a ping that should be crossing a single cable plus a few access-network routers. Something else is happening between our Daet-side probe and the Hermosa Beach target.

Three things are consistent with what we see:

• A different cable. The return route may be leaving the Philippines on a separate system — for example a Hong Kong or Singapore transit hop — then crossing the Pacific on another cable entirely. The extra 87 ms minimum is in the right range for an Asia-via-hub → US path.
• Provider asymmetry. BGP policy often sends outbound packets along a cheaper or peer-preferred route that has nothing to do with the return path. Carriers on each side optimise independently; the customer sees a "hot potato" on one side and a carefully groomed private line on the other.
• Middleboxes on the return leg. The 770 ms spikes and 117 ms standard deviation are not physics. They look like a router somewhere is queueing during peak hours. Sub-millisecond stability on one direction and 500+ ms tails on the other is a near-perfect fingerprint for that.

We cannot tell from outside which of the three dominates. What we can tell is that JUPITER itself is working beautifully; the asymmetry lives in the routing decisions wrapped around it.

Why hyperscalers keep building these

JUPITER is part of a broader shift in how trans-Pacific capacity is financed. For three decades most large international cables were paid for by telecom carriers, who then resold capacity wholesale. Starting in the mid-2010s, the operators of planet-scale data services — the companies running search, video, social, and cloud at unprecedented scale — began co-investing as primary sponsors alongside traditional carriers.

The economics are straightforward. A hyperscaler that moves multiple terabits per second between its US and Asian data centres can amortise a share of a new cable faster than it could rent equivalent capacity on the open market. Equally important, direct ownership lets the operator control the fibre pair — which means dedicated capacity at predictable latency, tuned to the operator's own traffic engineering rather than a wholesale price list.

JUPITER's consortium includes both hyperscaler sponsors and regional carriers. The cable is managed by a joint operating agreement; individual owners take fibre pairs for their own use and share the common infrastructure (power feed, landing stations, repeaters). It is the same pattern we have documented in other articles on Marea and Equiano: a private-sector cable built for one or two big customers, opened for wholesale lease only as a secondary function.

12-fibre-pair architecture

JUPITER's trunk was built with twelve fibre pairs — twice the count of most earlier trans-Pacific systems. Each pair runs through the same chain of optical amplifiers (erbium-doped repeaters powered by high-voltage direct current sent from shore), so adding pairs does not multiply the cost linearly: once the cable body and power-feed equipment exist, extra fibres are comparatively cheap. The result is a step-change in capacity. JUPITER's design capacity, with modern coherent transponders, is in the hundreds of terabits per second across the system — more than an entire generation of 2000s-era cables put together.

For the end user this shows up nowhere directly. What it produces is the kind of number we measured: 118.6 ms, every day, within 0.1 ms, across an ocean.

What our data proves

Thirteen thousand kilometres of open ocean, 60 measurements over 30 days, and a minimum that matches the theoretical limit to within instrument noise. JUPITER works. The cable does precisely what a trans-Pacific fibre pair built in 2020 is supposed to do.

The reverse direction is a different story — not a story about JUPITER, but a story about the routing politics of the wider internet. The physics is identical; the economics is not. Pings are one of the few tools that can tell you, from the outside, that the two directions of a connection do not share the same path. In this case the gap is almost 200 ms and plainly visible.

JUPITER is the floor. Everything above 118.6 ms westbound — and everything above 205.3 ms eastbound — is somebody's routing choice. The cable itself has no opinion on the matter.

Try it yourself

See the live data we collect on this cable on the JUPITER cable page, and explore asymmetric routing between any two cities on our route calculator. Our open measurement data is updated every two hours.