TPU

TPU: A New Digital Route Across the Pacific Ocean

TPU is one of the most technologically fascinating next-generation submarine cables. Owned by Google, it connects the western coast of the United States with Guam, Taiwan, the Philippines, and the Northern Mariana Islands.

The name TPU is derived from the three main directions of the system: Taiwan - Philippines - United States. The total length of the cable is approximately 13,470 km, and its design capacity reaches up to 260 Tbps. This is not an experimental line or a small regional project but a major trans-Pacific system designed to handle massive volumes of cloud, corporate, and consumer traffic.

The story of TPU is intriguing not only because of its technical specifications. The cable was built against the backdrop of the growth of cloud services, Google's increasing role in global telecommunications infrastructure, and heightened U.S. focus on the resilience of connections with Taiwan and countries in the western Pacific region.

Quick Facts

🗺 Show TPU on the interactive cable map

How the Route is Structured

The main TPU trunk crosses the Pacific Ocean between the northern coast of California and the western Pacific. Separate branches extend to Taiwan, the Philippines, and Guam. Later, a connection to the island of Tinian was added to the system.

The initial configuration consisted of four main segments:

• a trans-Pacific trunk approximately 12,500 km long;
• a Taiwan branch approximately 320 km long;
• a Philippines branch approximately 520 km long;
• a Guam branch approximately 130 km long.

In total, this adds up to the published length of approximately 13,470 km. The extension to Tinian came later, so the actual length of the final configuration may be slightly longer.

The project also included unused underwater branching units. This is an important engineering detail: the cable was originally built with the potential for future expansion. Additional islands or new systems can be connected without laying another full line across the Pacific Ocean.

Project History

The public history of TPU began in May 2023, when Google's subsidiary, GU Holdings, filed an application with U.S. regulators to construct and operate the cable. Initially, marine installation was planned to begin in March 2024, with the system launching in May 2025. For this reason, TPU was often referred to in early publications as the "2025 cable."

In reality, the timeline shifted. One reason was the lengthy U.S. licensing process. In July 2023, the U.S. Department of Justice requested a pause in the application review to complete an assessment of national security, foreign participation, and telecommunications infrastructure protection. The review lasted over a year.

In January 2025, Google reached an agreement with U.S. government agencies to meet special security requirements. Subsequently, on February 26, 2025, the U.S. Federal Communications Commission issued the final permit for the construction and operation of TPU. Before obtaining a permanent license, Google was allowed to build and test the U.S. sections of the system under a temporary permit-at the company's own risk.

Thus, the difference between 2025 and 2026 is straightforward: May 2025 was the original planned date, while 2026 became the actual year the system was ready for operation.

Who Owns the Cable

TPU is commonly referred to as a Google cable, but legally, different parts of the system are owned by different companies within the group: the U.S. sections are owned by GU Holdings; international waters and the Philippine section are managed by Google Singapore; the Taiwan section is owned by Google Taiwan. This structure is used to comply with the laws of various countries and territories.

TPU is not a traditional consortium system, where multiple operators jointly fund construction and share fibers. Google controls the cable independently.

The system is registered as a non-common carrier. This means Google is not obligated to provide access to it for all operators at public and uniform rates. The company can use the capacity for its global network and enter into individual commercial agreements with selected partners.

Why Google Needs Its Own Cable

The TPU example illustrates how the economics of submarine connectivity have changed. Previously, new systems were typically built by international consortia of telecommunications operators. Today, the largest consumers of international bandwidth-Google, Microsoft, Meta, and Amazon-invest in cables themselves.

For Google, TPU addresses several objectives:

• connects Google's U.S. and Asian Cloud infrastructure;
• enables data replication between regions;
• supports YouTube, search, advertising, and other services;
• reduces reliance on third-party operators;
• allows control over route quality and load;
• creates a reserve for future growth of cloud and AI services;
• enables independent selection of landing points and backup routes.

This does not mean that TPU exclusively carries Google data. The company may sell or lease individual capacity blocks on a long-term basis. However, the system's primary architecture was designed to meet the needs of Google's own network.

Why Choose Eureka

The U.S. landing point for TPU is located near Eureka and Arcata in northern California. This is an unusual choice. Most older trans-Pacific cables landed closer to major telecommunications hubs in California, such as Los Angeles, San Francisco, or the central coast.

The Eureka landing provides geographic diversity. Damage to infrastructure, an earthquake, or a major incident in one of the traditional cable clusters would not necessarily affect the northern California route.

The cable station is located at an EdgeConneX facility in Arcata. This site was previously used by a company involved in agricultural feed production and was later acquired and converted into a data center and cable station.

The same facility also hosts another Google cable-Echo. This enhances Eureka's importance as a new trans-Pacific hub but simultaneously creates a single point of risk. Two different oceanic routes are not entirely independent if they share a station, power supply, or terrestrial transport channel.

Guam: A Cable Crossroads and Risk Point

Guam is one of the most important telecommunications hubs in the western Pacific. The island hosts cables connecting the U.S., Japan, Australia, and Asian countries. TPU lands at Tanguisson Point and utilizes AT&T's cable station.

Guam's significance is defined by its geographic location. The island lies between major Pacific traffic routes and enables data to be switched between multiple regional and intercontinental systems.

However, the presence of many cables does not guarantee absolute resilience. If multiple systems use the same island, nearby shore corridors, shared stations, or power supplies, they could be simultaneously affected by a severe typhoon, earthquake, terrestrial infrastructure damage, power system failure, multiple cable cuts in one marine corridor, or a military or political crisis.

This is why Google's strategy involves not just connecting to Guam but forming a network of links between Guam, Tinian, Hawaii, Japan, Taiwan, the Philippines, and the U.S.

Taiwan: A Direct Route to the U.S.

In Taiwan, TPU lands at Dawu on the southeastern coast. A new station was built for the cable, owned by Chunghwa Telecom. The location on the Pacific side of the island creates a route distinct from lines concentrated near the western coast and the Taiwan Strait.

This is particularly important for Taiwan. The island's economy heavily depends on international data transmission, with major semiconductor manufacturers, tech companies, cloud platforms, and financial organizations operating there.

TPU provides an additional direct link between Taiwan and the U.S. However, it is incorrect to claim that the cable was built solely as a contingency for a potential conflict involving the island. The official goals of the project are capacity growth, reliability, and cloud infrastructure development. Geopolitical resilience is an important part of the overall picture but not a proven sole reason for the construction.

The Philippines: A New Point Outside Manila

The Philippine branch of TPU lands in Claveria on the northern coast of Luzon Island. The cable station is owned by Innove Communications, part of the Globe group. Claveria should not be seen as a pre-existing major international internet hub. Its significance arose specifically due to the construction of the TPU station and its connection to Luzon's terrestrial infrastructure.

The northern location offers several advantages:

• reduces the branch's length to the main trunk;
• is relatively close to Taiwan;
• creates an alternative to traditional points near Manila;
• expands the geography of international landings in the Philippines;
• reduces dependence on a limited number of coastal clusters.

However, true resilience will depend not only on the submarine cable. Equally important are independent terrestrial routes from Claveria to Manila and other major cities in the country.

Why Tinian Was Added to TPU

Tinian is part of the Commonwealth of the Northern Mariana Islands. Before Google's new projects, the territory's international connectivity heavily depended on Guam. This posed an obvious risk: a major incident on Guam could disrupt the primary external route for the entire archipelago.

In April 2024, Google announced the expansion of TPU to Tinian. At the same time, the company introduced the Proa system, which is set to connect Japan, Tinian, and Guam. Together, TPU and Proa form a new route: continental U.S. - Tinian - Guam - Japan. Tinian thus transforms from a terminal island branch into an intermediate element of a larger network.

For the Northern Mariana Islands, the project has not only technical significance. Direct international cables could enhance the territory's attractiveness for data centers, government systems, telemedicine, distance education, and digital business. However, the mere presence of a cable landing does not guarantee affordable internet for the population. This requires local competition, operator access to capacity, a developed terrestrial network, and reasonable pricing policies.

Design Capacity: What 260 Tbps Means

The design capacity of TPU is up to 260 Tbps. The main trunk features 16 pairs of optical fibers. The branches to Taiwan and the Philippines are specified to have 20 pairs each. The initial engineering estimate was approximately 13 Tbps per pair.

It is important to understand that 260 Tbps is not the volume of traffic constantly passing through the cable. Submarine systems distinguish between design, installed (equipped), activated, sold capacity, and actual traffic usage.

The cable may be physically capable of 260 Tbps but initially utilize only a small portion of this potential. Additional capacity is activated as demand grows and new terminal equipment is installed. Furthermore, capacity can increase over the system's lifespan due to advancements in optical transponders and modulation methods-without replacing the submarine cable itself.

Multi-core Fiber: The Key Technological Feature

TPU is described as the first commercial submarine project to use multi-core optical fiber. In a standard fiber, a single light-conducting core is used. In multi-core fiber, several independent cores are placed within a single shared cladding. Each core can carry its own optical signal. This allows for an increase in the number of parallel channels without a proportional increase in cable size.

However, it is important not to overstate the scale of the technology's application. According to published data, multi-core fiber is used on the branches to Taiwan and the Philippines, not on the entire 12,500-kilometer trans-Pacific trunk. This choice makes sense: shorter branches have lower optical losses and are less complex environments for the first commercial implementation of the new technology. The exact number of cores, channel configuration, and spectral plan for TPU have not been publicly disclosed.

Underwater Optical Switching

TPU's branching units employ optical switching. A branching unit is a seabed device that splits the trunk into multiple directions. In the simplest version, fibers are connected to specific branches during manufacturing, and the configuration cannot be changed after installation.

Optical switching allows for more flexible distribution of fiber paths between directions. For example, some resources can be reallocated between the main route and a specific branch without retrieving the underwater device to the surface. This is particularly useful for a system that is expanding and interconnecting with other Google cables. However, publicly available data does not reveal TPU's exact switching matrix, the number of available configurations, or the possibility of dynamically switching individual spectral bands.

The Physics of Latency

Based on the full length of TPU-13,470 km-the minimum possible signal propagation time is approximately 66 ms one way and 132 ms round trip. This calculation is based on the speed of light in optical fiber, approximately 204,000 km/s.

However, the measured Eureka - Claveria direction is shorter than the full cable length. Its calculated length is approximately 10,594 km, with a physical minimum of about 103.7 ms RTT.

Over the past 30 days, 22 measurements were recorded for this direction: a minimum of 191.0 ms, an average of 192.7 ms, a maximum of 194.2 ms, a standard deviation of 0.85 ms, and a median hop count of 8. The minimum latency was approximately 1.84 times higher than the calculated physical limit.

How to Interpret 191 ms

A latency of around 191-194 ms appears very stable. The difference between the minimum and maximum is only about 3.2 ms, and the standard deviation is less than one millisecond. This indicates a stable route without noticeable short-term congestion.

However, stability does not prove that traffic is passing directly through TPU. The approximately 87 ms difference between the physical limit and the measured RTT is too large to be explained solely by transponder and router operations. Several explanations are possible:

1. IP traffic is currently routed through another active cable.
2. The route detours through Japan, Guam, or another Asian hub.
3. The specified destination address is physically located away from the cable station.
4. A long terrestrial backhaul is in use.
5. BGP policy selects a commercially preferred path rather than the shortest one.
6. The measurement reflects the availability of a network node associated with TPU, not the cable segment itself.

Thus, the phrasing should be cautious: this is a measurement between points associated with Eureka and Claveria, but not a proven measurement of the TPU submarine segment's latency. To confirm the route, traceroutes, IP geolocation of interfaces, autonomous system data, and BGP analysis are needed.

Why a 0.9 ms Value is Impossible

Our data for the Eureka - Claveria direction does indeed include a 0.93 ms value recorded on June 12, 2026. Physically, this is impossible: even in a vacuum, a signal cannot traverse the Pacific Ocean round trip in 0.9 ms. Meanwhile, the average value for the same direction is about 191 ms, meaning this is a single outlier rather than a systematic discrepancy.

This result cannot be explained by ICMP response rate limiting either: rate limiting leads to packet loss, delayed responses, or no response, but it cannot make a signal move faster than light. The most likely causes are a response from an intermediate or local device instead of the final destination, hitting the nearest node instead of the stated address, a target identification error, or a data processing defect.

The correct practice is to automatically exclude such values from statistics as physically impossible. This is now implemented: measurements below the physical limit of the corresponding segment are not included in cable performance calculations.

What the March 29, 2026 Incident Means

The database contains a warning opened on March 29, 2026. A review of its details shows that it was not a cable failure, for several reasons.

First, the alert was recorded for the address 198.41.0.4-an anycast address of one of the root DNS servers. Such an address, by definition, responds from the node closest to the probe, so measurements to it do not characterize a specific submarine cable.

Second, the alert's numbers are internally contradictory: the baseline value was 47.6 ms, the current value was 44.3 ms, meaning the current latency was LOWER than the baseline, even though the increase was recorded as 296 percent.

Third, the system itself noted the absence of confirmations: the route verdict was "no_data," the consensus was "insufficient_witness_context," and the number of independent witnesses was zero. The alert automatically closed on April 4, 2026, after three consecutive successful checks.

To confirm a physical incident, simultaneous latency increases across multiple independent probes, packet loss, traceroute changes, BGP route disappearance, traffic switching to a backup system, and operator reports would be required. None of these were recorded, so the event is correctly described as a monitoring anomaly rather than a cable failure.

Redundancy: It's Not That Simple

There are indeed many alternative cables along the TPU route. Guam is connected to Japan, Australia, and Asia; Eureka is used by the Echo system; Taiwan and the Philippines have other international lines. But the presence of a neighboring cable does not automatically ensure redundancy.

For an alternative route to work effectively, the operator needs pre-purchased capacity, configured BGP routing, an independent landing station, a separate terrestrial backhaul, sufficient spare capacity on the backup system, and an agreement with the alternative cable's owner. If two cables land at the same station or pass through the same shore corridor, they could be affected simultaneously.

TPU is best viewed not as a standalone line but as part of Google's larger Pacific network, which includes Echo, Proa, Taihei, Tabua, and inter-island connections in the Pacific.

What Incidents Are Most Dangerous for TPU

Fishing and Shipping. In shallow waters, cables are more often damaged by anchors and fishing gear. Approaches to shore stations, where the cable crosses the continental shelf, are particularly vulnerable.

Earthquakes and Underwater Landslides. Taiwan, the Philippines, Guam, and the Northern Mariana Islands are in a tectonically active region. A strong earthquake could trigger an underwater landslide and damage multiple cables simultaneously.

Typhoons. While the cable at great depths is usually unaffected by storms, typhoons can destroy coastal infrastructure, power supplies, terrestrial routes, and shallow sections.

Volcanic Activity. The western Pacific is part of the Pacific Ring of Fire. Volcanic processes can cause earthquakes, changes in seabed topography, and localized underwater landslides.

Geopolitical Risks. TPU connects territories near potential conflict zones. However, without confirmed data, specific outages cannot be attributed to sabotage or military activity.

Common Points of Failure. Multiple cables may share a station, data center, power input, or terrestrial route. In such cases, diversity in marine routes does not ensure full fault tolerance.

How Important is TPU

TPU is significant not only because of its 260 Tbps capacity. It demonstrates the transition from standalone cables to software-managed oceanic networks. Google is building several interconnected systems that can distribute traffic based on load, availability, and data purpose.

Three features of TPU are particularly important. First, private ownership: Google controls the system's technical and commercial model. Second, expandable architecture: unused branching units allow for adding new directions. Third, new technologies: multi-core fiber and optical switching increase the density and flexibility of cable infrastructure.

At the same time, TPU traverses one of the world's most challenging regions-seismically active, typhoon-prone, and at the center of strategic competition among major powers.

Conclusion

• TPU is a Google-owned trans-Pacific system approximately 13,470 km long with a design capacity of up to 260 Tbps.
• The cable connects California, Guam, Taiwan, the Philippines, and Tinian. The main trunk uses 16 fiber pairs, while the branches to Taiwan and the Philippines use 20 pairs each. The supplier is NEC.
• The initial launch was planned for May 2025, but prolonged licensing and project changes delayed the launch to 2026.
• The main technological feature is the commercial use of multi-core fiber on the Asian branches.
• The architecture with underwater switches and spare branching units shows that TPU was designed with future expansion in mind.
• Measurements for Eureka - Claveria show stable 191-194 ms latency, approximately 1.84 times higher than the physical limit of the segment, but do not yet prove traffic is passing directly through TPU.
• Physically impossible results, such as 0.93 ms, are automatically excluded from statistics.
• The March 29, 2026 warning was a monitoring anomaly, not a failure: it was recorded for an anycast address of a root DNS server, the current latency was below the baseline, there were no independent witnesses, and the alert closed automatically.
• TPU should be viewed not as a single cable but as a key element in Google's developing Pacific network.