Bifrost

BIFROST is the longest submarine cable we monitor. Nineteen thousand eight hundred and eighty-eight kilometres of fibre, stretched from Jakarta across the Indonesian archipelago, through Davao and Manado in the Philippines, via Tuas in Singapore, out to Alupang in Guam, and then across the full width of the Pacific Ocean to three North American landings — Grover Beach in California, Winema in Oregon, and Rosarito in Mexico. It is one of the first high-capacity cables to connect Southeast Asia directly to the Americas on a new southern route that avoids the congested Japan and Taiwan corridors. Ready for service in 2025, with a Meta-backed consortium behind it.

Because the cable is so long — and because the two ends of our BIFROST measurement sit on opposite sides of the Pacific, in countries an ocean apart — this is the best cable in our dataset to demonstrate one very stubborn fact about the internet: the path from A to B is almost never the same as the path from B to A.

The two directions do not agree

Our monitor has two probes active near BIFROST landings. One is close to Jakarta; the other is close to the Rosarito cable station in Baja California, Mexico. Both probes ping toward the opposite landing, and the measurements we collect look like they come from two completely different cables.

That 70-millisecond gap between the two minimums is enormous. It is larger than many complete transatlantic crossings. And the hop counts are just as striking: the eastbound path typically takes more than twenty IP hops, while the westbound path settles at sixteen. These are not the same route.

Physics floor: the 20,000-kilometre problem

BIFROST's full trunk length is 19,888 km. At a fibre signal speed of about 204,500 km/s (glass refractive index ~1.467), the round-trip minimum through the entire length of the cable is 194.4 ms. That is the absolute floor — the number below which, regardless of how well the cable is built, no ping can go.

Our measured Jakarta → Rosarito minimum is 198.3 ms. That is 1.02× the physics floor. If you are looking for the cleanest evidence we have ever published that a submarine cable is doing exactly what it was designed to do, it is sitting in this one statistic. For a 20,000-kilometre path, we are measuring within 2% of the theoretical best-case. No routing detour, no wasted milliseconds — the packet traverses essentially the full BIFROST trunk and arrives at precisely the time glass physics allows.

But our Rosarito → Jakarta minimum is 128.6 ms. That is below the physics floor for the full BIFROST length.

Below the floor means a different path

A 128.6 ms round trip corresponds to a one-way fibre distance of about 13,156 km. The great-circle distance from Rosarito to Jakarta is approximately 13,800 km. Close match. That is consistent with a direct trans-Pacific path across the shortest route between the two coasts — which BIFROST does not take. BIFROST swings south through the Indonesian archipelago, picking up multiple landings, before crossing to the Americas; its effective path length between Jakarta and Rosarito is the full ~20,000 km.

In other words, the Rosarito → Jakarta return path is probably not using BIFROST at all. It is using a shorter cable — one of the direct US-to-Japan or US-to-Singapore systems like FASTER, JUPITER, or Southern Cross NEXT — that takes a straighter route across the Pacific and terminates in Asia via a different set of cables.

The forward path (Jakarta → Rosarito) is clearly using BIFROST: the minimum matches the physics floor for the cable's full length to within 2%, and the sample count is higher and more stable in exactly the way one would expect from a direct, single-cable route.

Three trans-Pacific cables in one measurement

What the BIFROST data actually captures, then, is not a single cable in operation. It is a routing policy — the emergent behaviour of multiple cables cooperating in an asymmetric way.

The Jakarta-side carrier (the operator running probe 6681) seems to prefer the long BIFROST route for eastbound traffic. That might be because BIFROST is the cheapest option for them, or because they own a pair on BIFROST and cannot send traffic on competitors' cables, or because their best-peered upstream happens to sit on a BIFROST-connected hub. Every one of these reasons is consistent with what we see.

The Rosarito-side carrier (the operator behind probe 125) takes a completely different approach for westbound traffic. Its path is 5,000 km shorter in effective fibre length, almost certainly because it uses a direct Pacific cable rather than the BIFROST trunk. That operator apparently has better reach, a cheaper contract, or more capacity on the shorter route, and it moves traffic that direction.

This is not a malfunction. It is how BGP — the internet's global routing protocol — handles cost and policy at scale. Each direction is optimised independently by the carrier on its side. The result is a measurement that looks like two different cables because, from a traffic-engineering perspective, it is.

Why BIFROST matters on a map

BIFROST joined a small set of cables that connect the "southern corridor" between Asia and the Americas. Historically, the overwhelming majority of trans-Pacific capacity passed through Japan or Taiwan and landed on the US West Coast somewhere between Los Angeles and Oregon. Political and geographic risk concentrated in those hubs: earthquake zones along the Ring of Fire, dense fishing activity, strategic chokepoints.

The southern corridor takes the cable further south: via Indonesia, Singapore, and Guam, landing in California and Mexico. That keeps capacity flowing when the northern corridor has issues, and it reduces the combined failure risk of the transpacific internet.

Three other cables sit on this corridor — BIFROST's siblings — Echo, Apricot, and the existing SEA-US, with a handful more on the drawing board. They collectively add tens of terabits per second of new Pacific capacity that does not transit the same countries as the older Japan-California trunks.

Mexico joins the trans-Pacific map

Rosarito is worth a second look. The town sits on the Baja California coast, just south of Tijuana and the US border. Before BIFROST, Mexico had essentially no trans-Pacific cable connectivity: its transpacific traffic routed north into the US and then across via cables from Oregon or California. With BIFROST, Mexico has a direct fibre link to Singapore and Indonesia for the first time.

This changes the network map. Latin American traffic destined for Southeast Asia no longer needs to detour through US infrastructure. For operators south of the border, a direct landing on their own coast reduces transit fees, speeds up BGP convergence, and removes a significant source of geopolitical risk from their network design.

What our data proves

Three concrete observations come out of the BIFROST measurement:

• The cable works. Jakarta → Rosarito minimum RTT of 198.3 ms is within 2% of the theoretical physics floor for a 19,888 km path. BIFROST is delivering its advertised performance.
• The return route is a different cable. Rosarito → Jakarta minimum RTT of 128.6 ms is physically impossible on BIFROST (below the light-speed floor for its length), so the return path is using a shorter trans-Pacific system.
• BGP is asymmetric. Both observations are stable over the full 30-day measurement window. This is not a transient routing glitch — it is the steady-state policy of the two sides' operators.

BIFROST is one of the newest, longest, and most geographically important cables in our dataset. What the data shows is that building a long cable is only the first step — the harder, more interesting problem is the messy coordination of transit policy, peering, and BGP across thousands of operators that actually decides where your packet will travel on any given day.

Try it yourself

Live measurements are on the BIFROST cable page. For more on routing asymmetry, see our companion piece on JUPITER (another trans-Pacific cable where the two directions measure as 118 ms westbound and 305 ms eastbound). For another example of a long trans-oceanic cable hitting near-physics-floor latency, see Marea (1.95× floor on the Atlantic).