Andreessen Horowitz's a16z Speedrun fund is backing Orbital Inc., a startup that plans to launch AI data centers into low-Earth orbit powered entirely by solar energy, bypassing the growing electricity crisis constraining artificial intelligence infrastructure on Earth. The company, led by co-founder and Chief Executive Euwyn Poon, closed on funding of an undisclosed amount to finance its first test mission, called "Orbital-1," which will validate the concept of running GPU-powered inference workloads 1,200 miles above the planet's surface .

Why Is AI's Energy Consumption Becoming a Critical Problem?

The artificial intelligence industry's explosive growth has created an unprecedented strain on terrestrial power grids. Data centers housing thousands of graphics processing units (GPUs) and other advanced chips consume electricity at rates that are outpacing the deployment of new energy sources. This energy crisis has become so acute that major technology companies, including Google, Microsoft, and Amazon, recently met with U.S. President Donald Trump at the White House to discuss mitigation strategies and commit to bringing new energy sources online .

Orbital's founders believe the solution lies not in building more power plants on Earth, but in moving compute infrastructure into space entirely. By positioning satellites in sun-synchronous orbits, the company can harness continuous solar energy without the limitations imposed by Earth's day-night cycle. Additionally, the vacuum of space provides a natural cooling mechanism, allowing heat generated by GPU operations to dissipate directly into the atmosphere through radiative cooling, eliminating the need for energy-intensive cooling systems .

How Will Orbital's Space-Based Data Centers Actually Work?

Orbital is designing a constellation of independent satellite nodes that will distribute AI inference workloads across massive orbital clusters of Nvidia-powered servers. Each satellite will function as a decentralized processing unit, capable of handling discrete inference requests without the latency problems that might plague other space-based computing approaches. The company is targeting inference workloads specifically, not training workloads, because inference requests are stateless and can be distributed across clusters without compounding errors .

One of the primary concerns about space-based data centers is latency, or the delay in transmitting data to and from orbit. However, Poon noted that at altitudes of 500 to 600 kilometers, round-trip latency measures only 20 to 40 milliseconds, comparable to a fiber connection between Los Angeles and Denver. Starlink has already demonstrated commercially that sub-50-millisecond latency is achievable from space to ground-based users, validating the feasibility of this approach .

Steps to Understanding Orbital's Technical Challenges and Solutions

• Radiation Hardening: Cosmic radiation in low-Earth orbit can cause "bit flips," corrupting data stored in GPU memory. Orbital has developed radiation-hardening technologies and designed the Orbital-1 test mission specifically to measure these effects under real conditions and validate protective systems.
• Satellite Lifecycle Management: Rather than attempting the nearly impossible task of repairing or upgrading satellites in orbit, Orbital designs each satellite with a defined lifecycle and plans for controlled deorbit and complete burn-up on reentry, making the approach more environmentally friendly than terrestrial data centers that generate massive e-waste streams.
• GPU Testing and Validation: The company conducts extensive ground-based testing of GPUs and systems before launch to ensure reliability in the harsh space environment, which is characterized by extreme temperature fluctuations and high-radiation exposure.

Poon explained the company's approach to managing these challenges:

"Our approach focused on radiation hardening, and the test mission is designed to specifically measure these effects in real conditions. We're also fortunate that our target workload is inference, which is stateless. That means each request is discrete, so a corrected error doesn't compound across runs the way it would in AI training," he stated.

Euwyn Poon, Co-founder and Chief Executive at Orbital Inc.

What's the Timeline for Orbital's First Mission?

Orbital is moving with remarkable speed. The company plans to launch its first satellite aboard a SpaceX Falcon 9 rocket in April 2027, just 12 months from the time of this announcement. The Orbital-1 mission will serve as a proof-of-concept, validating the company's ability to sustain GPU operations in a high-radiation environment and support commercial AI inference workloads at scale. If successful, Orbital intends to scale up significantly and build a massive constellation of orbital compute nodes .

To support this ambitious timeline, Orbital is opening a dedicated research and development facility in Los Angeles called Factory-1, where the company will begin manufacturing its first specialized compute satellites .

Andrew Chen, a General Partner at a16z, emphasized the fund's commitment to backing founders tackling the hardest problems.

"The harder the problem, the better. Orbital is taking on AI's biggest constraint with a bold and radical idea," he noted.

Andrew Chen, General Partner at Andreessen Horowitz

If Orbital succeeds in solving the technical challenges of operating GPUs reliably in space, the company could fundamentally decouple AI progress from the resource constraints facing terrestrial data centers. This would enable a future in which cloud computing is literally performed in the sky, far above the clouds, powered by the sun's unlimited energy and cooled by the vacuum of space.