The Undersea Cables of Thought: How Physical Geography Shapes What AI You Can Access (and How Fast)

You type a prompt and hit enter. The response appears in a second or two. You don't think about the path it took. But your words travelled further than you can imagine: from your device to a local exchange, across continents via fibre optic cables buried under oceans, through routers and switches, to a data center thousands of miles away, processed by servers, then back again. That distance is not abstract. It's physical. And it determines how fast you get your answer.

This is the geography of AI: the undersea cables, data centers, and fibre routes that shape who can access which models, and how quickly. If you live near a major data hub, your prompts fly. If you live far from the network, you wait. The speed of thought is constrained by the speed of light and the layout of the ocean floor.

Let's map this hidden infrastructure. By the end, you'll understand why your AI feels fast or slow, why geography matters more than you think, and what it means for the future of equitable access.

The Physical Path of a Prompt
Your prompt doesn't travel through "the cloud." It travels through solid, physical infrastructure.

The Journey:

Your device sends the prompt to your local router.

Your ISP routes it to a regional exchange.

Backbone networks carry it across land and under sea.

A data center receives the prompt and processes it on a server.

The response travels the same path back.

The Speed Limit:
The fastest possible speed is the speed of light in fibre, about 200,000 km/s (slower than light in a vacuum). Even at that speed, distance matters.

The Math:

A prompt from New York to a data center in Virginia: ~500 km round trip → ~5 ms latency.

A prompt from Sydney to a data center in California: ~24,000 km round trip → ~120 ms latency.

Add switching, processing, and queueing delays: the difference can be half a second or more.

A Contrarian Take: Latency Is Not the Only Limit, but It's the Most Democratic.

Wealth can buy you a faster subscription tier, a better device, a dedicated fibre line. But it cannot buy you a shorter path to the data center. Physics is the great equalizer.

If you live in a region with poor connectivity, no amount of money will make your prompt travel faster than the speed of light. The undersea cables are the same for everyone. The distance is the distance.

This is both a limitation and an opportunity. The geography of AI is not a hierarchy of wealth. It's a geography of proximity.

The Undersea Cable Network
Most international data travels through undersea fibre optic cables.

The Map:

Hundreds of cables span the ocean floor, connecting continents.

Major hubs: Virginia (US), London (UK), Singapore, Tokyo, Sydney.

The cables are owned by consortia of telecoms and tech companies.

The Gaps:

Africa is underserved. Many countries rely on a single cable, making them vulnerable to outages.

South America has limited connectivity to Asia and Africa.

Small island nations are often at the end of long, slow links.

The Implications for AI:

If your nearest data center is across an ocean, your latency will be high.

If your region has few cables, a break can cut you off entirely.

Data Center Geography
Where are the AI models actually running?

The Clusters:

North America: Virginia, California, Oregon, Canada.

Europe: Ireland, London, Frankfurt, Amsterdam.

Asia: Singapore, Tokyo, Seoul, Mumbai.

Oceania: Sydney, Melbourne.

South America: São Paulo (limited).

Africa: Cape Town, Johannesburg (very limited).

The Gap:
Most of the world's AI compute is concentrated in a few regions. If you're not near one of these clusters, your prompts travel further and wait longer.

The Trend:
Cloud providers are building more data centers in emerging regions, but the distribution remains highly uneven.

A Contrarian Take: The Data Center Map Is Not the End of the Story.

The major cloud providers place data centers where there is demand, reliable power, and good connectivity. This creates a feedback loop: demand attracts data centers, which reduces latency, which attracts more demand.

Regions without data centers are locked out of the loop. Their latency remains high, so local users and businesses are less likely to use AI, so there's less demand, so no data center is built.

Breaking this cycle requires investment, policy, and a recognition that AI access is becoming a competitive necessity.

The Real‑World Impact
Latency is not just a technical metric. It affects experience, productivity, and opportunity.

User Experience:

A 50 ms delay is barely noticeable.

A 200 ms delay feels sluggish.

A 500 ms delay disrupts flow.

Application Constraints:

Real‑time applications (voice, video, interactive AI) require low latency.

If you're far from a data center, some applications may be unusable.

Economic Consequences:

Businesses in high‑latency regions are at a competitive disadvantage.

AI‑powered tools may be less effective, leading to lower productivity.

Example:
A customer service AI in Lagos, Nigeria, may have to route through Europe or North America. The latency is noticeable. The interaction feels less fluid. The technology is less effective than it could be.

What You Can Do
You can't move the undersea cables. But you can make choices.

1. Choose a Nearer Data Center
   If your AI provider offers regional endpoints, select the closest one.

2. Use Edge Computing
   Some AI models can run on local devices. For simple tasks, edge computing eliminates network latency.

3. Batch Your Queries
   If latency is high, reduce the number of round trips. Combine multiple prompts into one.

4. Advocate for Local Infrastructure
   Support policies that encourage data center investment in underserved regions.

5. Monitor Your Latency
   Measure it. Understand it. Choose providers that are transparent about their infrastructure.

The Future of AI Geography
The undersea cable network is not static. New cables are laid every year.

Near Term:

More cables to Africa and South America.

More data centers in emerging regions.

Lower latency for more of the world.

Medium Term:

Edge AI will reduce reliance on distant data centers.

Local inference on devices and local servers will become common.

Long Term:

The geography of AI may become less relevant as processing moves to the edge.

But for large models, data centers will remain essential.

The Unequal Map
The undersea cables are the hidden backbone of AI. They determine who gets fast answers and who waits. They are not a conspiracy. They are the legacy of decades of investment, concentrated in wealthy regions.

But they are also a map of inequality. The same cables that carry your prompt in milliseconds leave others waiting seconds. The geography of AI is the geography of power.

The next time you get a fast response, consider the path it took. The fibre under the ocean, the servers in Virginia, the routers in between. And consider the person on the other side of the world, whose prompt is still travelling.