Your satellite provider quoted you 200 Mbps. You're getting 11. The vessel is pitching in moderate seas south of New Zealand, and your systems just stalled because the connection dropped. Again.
This is the gap between satellite marketing and satellite reality. If you're considering AI on your vessel (or honestly, any technology that depends on internet connectivity)you need to understand what's actually happening with your link. Because your whole strategy depends on it. (Spoiler: the answer is not to depend on the cloud at all.)
Three Types of Satellite, Three Different Realities
Satellite services sit at three different altitudes, and each makes a fundamentally different trade-off.
Traditional satellites (GEO), Companies like Inmarsat and SES. The satellite sits 35,000+ km above the equator and doesn't move relative to you. The good news: stable connection, no handover interruptions. The bad news: signal has to travel 70,000+ km round-trip, so you're looking at 550–700 millisecond delays on every request. Throughput is typically 2–20 Mbps. This is what most of the commercial fleet still runs, reliable but slow.
Mid-orbit satellites (MEO), SES's O3b mPOWER, orbiting at ~8,000 km. Delay drops to 125–175 ms. Throughput can be substantial. The catch: fewer satellites, coverage thins out at high latitudes, and you're paying premium prices for premium performance.
Low-orbit satellites (LEO) (Starlink, OneWeb, soon Amazon's Kuiper. This is where the hype is. Latency drops to 25–60 ms, close to what you'd get on land. Starlink Maritime advertises up to 220 Mbps. The constellation is massive)over 6,000 satellites. But these satellites cross your sky in about 4 minutes, so your terminal is constantly switching between them. And coverage depends on how close you are to a ground station.
Those are the physics. Here's what they actually mean when you're at sea.
What You're Actually Getting
Starlink Maritime in the North Atlantic or Gulf of Mexico (close to ground stations, dense satellite coverage overhead)delivers 80–150 Mbps fairly consistently. That's real. It's also the best-case scenario.
Move to the central Pacific, 1,500 nautical miles from the nearest ground station, and that drops to 15–40 Mbps. Head to the Southern Ocean below 55°S and coverage gets genuinely spotty. There will be periods with no coverage at all.
Legacy VSAT (what most of the fleet still runs)gives you 3–10 Mbps on a standard plan. Predictable, though. The connection doesn't vanish every 4 minutes.
The number that matters: sustained throughput over 24 hours, including weather, handovers, and congestion. For Starlink in good coverage areas, that's roughly 40–80 Mbps. In remote waters, 8–25 Mbps. That's your real number. That's what you plan around.
The Gaps Nobody Mentions in the Sales Meeting
Handovers. LEO satellites move fast. Your terminal switches satellites every 15–90 seconds. Each switch causes a brief interruption (usually a fraction of a second, sometimes longer. For browsing the web or checking email, this is invisible. For any system that needs a sustained, uninterrupted connection)video calls, real-time data sync, cloud-based AI, these micro-interruptions add up.
Rain. Heavy rain degrades satellite signals at the frequencies Starlink and most providers use. In the tropics during monsoon season, you can lose hours of usable bandwidth per day. Older C-band VSAT handles rain better, which is one reason it hasn't disappeared despite the terrible latency.
The "connected but empty" problem. Sometimes the satellite overhead can see your terminal but can't reach a ground station. Your dashboard shows "connected." Your throughput is zero. Your systems think they're online but nothing is getting through. This is worse than being disconnected, because at least a disconnected system knows to fall back to local operation.
Coverage geometry. LEO constellations thin out near the equator and at extreme latitudes. If your vessels transit the Southern Ocean, equatorial Pacific, or high Arctic routes, you will have dead zones. Full stop.
What This Means for AI: and Everything Else
Most people design their technology as if satellite is just a slightly worse version of shore-side broadband. It's not. It's a fundamentally different medium with different failure modes.
Shore broadband fails in binary, it works or it doesn't. Satellite fails in gradients. You get 80 Mbps, then 30, then 8, then 0 for 12 seconds, then 45, then 2 during a rain cell, then back to 60. Your technology needs to work usefully across that entire range.
This is why any AI system that depends on a cloud connection is the wrong architecture for a vessel. It's not about whether satellite is "good enough." It's about the fact that it's unpredictable, and anything mission-critical needs to work regardless of what the link is doing.
The right approach: run AI locally on the vessel. Use the satellite link for updates, sync, and communication, things that can tolerate interruptions. But your AI assistant, your monitoring, your operational tools? Those run on hardware aboard the ship. They work in perfect coverage and in zero coverage. They don't care what your satellite link is doing.
Should You Run Two Providers?
Starlink plus a traditional VSAT backup is becoming a common setup. Starlink handles the bulk throughput; VSAT provides the fallback that works everywhere, even if it's slow. The cost is real, two terminals, two subscriptions, two sets of hardware. For vessels where connectivity matters operationally, it's worth it. For a bulk carrier running basic monitoring, probably not.
Stop Planning for the Brochure
The satellite landscape is better than it's ever been. LEO constellations have changed what's possible at sea. But "possible" and "reliable" are different words.
Plan for the connection you'll actually have, the 2 AM throughput in the South Pacific during a rain squall, not the demo on a calm day in the English Channel. Build your systems to work without connectivity, and treat satellite as a bonus when it's available.
If you're planning a vessel deployment and want to understand what your actual connectivity picture looks like, reach out to our team. We've been through this with enough vessels to know what works and what doesn't, and we build AI systems designed for the reality of life at sea, not the marketing brochure.