Starlink Surpasses 10 Million Users as SpaceX Scales Gigawatt AI and Readies Orbital Compute

The satellite network's dominance, direct-to-phone expansion, and space-based AI infrastructure plans point to a tightly integrated system built for both global connectivity and frontier intelligence workloads.

Starlink's active user base crossed 10.3 million by the end of Q1 2026 after more than doubling in the prior year. Its constellation now accounts for roughly three-quarters of every active maneuverable satellite in orbit, while coverage spans more than 164 countries and territories serving over three billion people. At the same time, the company has stood up gigawatt-scale AI training infrastructure on the ground and is advancing plans to place portions of that compute in orbit, powered by sunlight and cooled by radiation.

Key Takeaways

• Starlink reached approximately 10.3 million users by the end of Q1 2026, reflecting more than 100 percent year-over-year growth from roughly 4.4 million users at the end of 2024.
• The network now represents about 75 percent of all active maneuverable satellites currently in orbit.
• Service is live across more than 164 countries and territories, providing coverage to populations exceeding three billion people.
• Adoption is accelerating in aviation with frequent new airline integrations, while enterprises increasingly use Starlink as primary or backup connectivity for operational resilience.
• Starshield operates as a dedicated constellation delivering secure capabilities tailored to U.S. government needs.
• The first-generation direct-to-device service, activated with 650 satellites and partnerships involving 30 mobile network operators, has already demonstrated coverage for 1.9 billion people and confirmed real demand.
• A second-generation direct-to-device system is designed to deliver full 5G performance directly to standard phones, targeting connectivity for the remaining three billion people currently outside reliable networks.
• Colossus Two has been deployed as the world's largest coherent supercomputer, incorporating NVIDIA GB300 processors as the first gigawatt-scale training cluster along with the first gigawatt-scale Megapack battery installation and early large-scale use of both GB200 and GB300 hardware.
• Orbital AI compute capacity is under active development for deployment within the next couple of years, leveraging solar power generation and radiative cooling in the space environment.
• Proven satellite subsystems already in production—including ion propulsion thrusters, inter-satellite laser links, flight computers, and reaction wheels—remove key technical barriers to placing AI workloads in orbit.
• Vertical integration across launch, satellites, connectivity, and AI creates a self-reinforcing loop: better infrastructure improves model quality and lowers token costs, which in turn funds further capacity expansion.

Rapid Scaling of the Starlink User Base

User growth has followed a steep trajectory. The network closed 2024 with roughly 4.4 million subscribers. By the end of the following year that figure had doubled to approximately 8.9 million. The first quarter of 2026 added further momentum, bringing the total to 10.3 million active users. That performance equates to sustained expansion above 100 percent year-over-year.

The underlying constellation provides the foundation for this growth. With roughly 75 percent of all active maneuverable satellites in orbit, Starlink maintains both dense global coverage and the ability to route traffic dynamically through inter-satellite links. This orbital density supports consistent performance even as demand rises across consumer, enterprise, and mobility segments.

Expanding Use Cases Beyond Consumer Broadband

While residential service drove early adoption, newer segments are contributing meaningful volume. Fixed-site deployments now frequently position Starlink as either the primary link or a resilient backup when terrestrial fiber or other infrastructure experiences outages. Organizations that treat uptime as a core operational requirement have incorporated the network into standard planning.

Aviation has emerged as one of the fastest-moving verticals. Airlines continue to announce integrations on a regular cadence, equipping aircraft to maintain high-performance connectivity over transoceanic routes and remote regions where legacy systems deliver limited or no coverage. The same reliability characteristics that appeal to airlines also support broader mobility and IoT applications.

Government customers operate on a parallel track through Starshield. This separate constellation is purpose-built to meet official requirements for secure, resilient communications that can function independently of commercial infrastructure.

Direct Satellite Links to Everyday Phones

Direct-to-device capability represents a distinct layer of the connectivity strategy. The first-generation implementation functioned as a working proof of concept. A deployment of 650 satellites activated service with approximately 30 mobile network operators worldwide. Coverage already reaches 1.9 billion people, and usage patterns have validated both technical feasibility and market interest in satellite-augmented mobile service.

The second-generation system targets full 5G performance delivered to ordinary handsets. Because it removes the need for specialized user hardware beyond a compatible phone, this step has the potential to extend modern connectivity to the roughly three billion people who remain outside reliable terrestrial networks. The approach builds directly on the satellite manufacturing and launch cadence already operating at scale.

Gigawatt-Scale AI Training on the Ground

The same vertical integration approach applied to connectivity has been extended to AI infrastructure. Ownership of the full stack from launch through satellite operations created the conditions to develop and train frontier models internally while capturing the complete value chain.

Colossus Two now operates as the largest coherent supercomputer deployed to date. The facility uses best-in-class NVIDIA GB300 processors and established several milestones at once: the first gigawatt-scale training cluster, the first gigawatt-scale battery installation built around Megapack systems, and early large-scale deployment of both GB200 and GB300 generations. A nameplate power capacity of one gigawatt provides the headroom required for training increasingly large models while the co-located battery systems help stabilize energy draw and support operational continuity.

Extending Compute into Orbit

The next phase moves portions of AI workloads off the ground entirely. Orbital placement offers structural advantages for power and thermal management. Satellites generate electricity directly from sunlight, and radiative cooling dissipates heat efficiently in the vacuum of space without consuming water or requiring mechanical chillers.

Data movement between orbital nodes and ground users can leverage the existing Starlink architecture, which has already demonstrated high-capacity global distribution. Once positioned, the marginal operating cost remains extremely low because the primary energy source is continuous solar input rather than grid electricity or fuel.

Progress relies on satellite subsystems that are already flying in volume. Ion propulsion maintains orbital positioning, laser terminals handle high-bandwidth inter-satellite communication, flight computers manage onboard processing, and reaction wheels control attitude. These mature components eliminate many of the integration challenges that would otherwise push orbital AI timelines further out. Development is now ramping, with meaningful capacity targeted within the next couple of years.

The Integrated Infrastructure Loop

Connectivity and AI capabilities reinforce each other through a deliberate operating model. Superior network reach and capacity provide the distribution layer needed to deploy and scale AI systems efficiently. Improved model performance then raises the quality of intelligence available across applications while simultaneously lowering the cost per token for inference. Revenue from both connectivity services and AI workloads funds the next round of infrastructure expansion.

Because launch cadence, satellite production, and orbital operations already function at high throughput, iteration cycles stay compressed compared with organizations that must coordinate multiple external suppliers for each layer. The result is a system in which advances in one domain accelerate progress in the others, creating compounding advantages in both global connectivity and computational intelligence.

This combination of terrestrial gigawatt-scale training resources and upcoming orbital capacity positions the overall architecture to address power, cooling, and land constraints that increasingly affect conventional data center expansion while extending high-performance connectivity to populations that traditional networks have left behind.