Bezos plan for solar powered datacenters is out of this world… literally

Amazon founder and executive chair Jeff Bezos says that, within two decades, gigawatt-scale datacenters powered by a continuous stream of photons from the sun will fill Earth’s orbit.

Bezos’ prediction might sound like he’s auditioning for the role of Hugo Drax in the inevitable Moonraker reboot, but the billionaire businessman really believes that spaceborne datacenters are going to be a thing and may eventually outperform their terrestrial counterparts.

“One of the things that’s going to happen in the next – it’s hard to know exactly when, it’s 10 plus years, and I bet it’s not more than 20 years – we’re going to start building these giant gigawatt datacenters in space,” he said during a fireside chat with Ferrari and Stellantis chair John Elkann, at Italian Tech Week.

Of course, to make this dream a reality, we’re going to need heavy lift rockets capable of shuttling the various components to and from orbit, which Bezos’ Blue Origin just so happens to be building.

However, even if we can make putting bit barns in orbit economical, there is no shortage of challenges to overcome. Datacenters are massively complex facilities. Even with an abundant supply of power, an orbital datacenter would still need a way to reject a gigawatt of thermal energy through radiation.

For reference, the ISS’s radiators are capable of rejecting about 70 kilowatts of thermal energy. Bit barn operators will also have to contend with bit-flips on a fairly regular basis unless the hull can be sufficiently hardened against charged particles from the sun and cosmic rays from the outer reaches of space — standard ECC probably isn’t going to cut it.

Many of these challenges are already being tackled at a smaller scale. HPE has been working on the problem for several years now with Spaceborne and Spaceborne-2. These systems have made several trips to the International Space Station already and have tested things like edge compute storage and recovery on long-term space missions.

The first unit, launched in 2017, experienced several failures during its 615-day mission aboard the ISS. As HPE’s Dr Eng Lim Goh told The Register at the time, one of the machine’s four redundant power supplies and nine of its 20 SSDs encountered issues.

Axiom Space has also launched a similar compute prototype to the ISS. In August, the startup’s shoebox-sized Data Center Unit One (AxDCU-1) arrived on the space station. However, all of these designs have been tiny power-optimized designs and are nothing like the kinds of data stations Bezos is describing.

Orbital datacenter advocates expect full scale facilities will require high degrees of automation, including the extensive use of robots to perform maintenance or upgrades.

But while most of the challenges facing spaceborne datacenters are engineering and economic in nature, there’s one that can’t easily be overcome, at least not without bending the laws of physics: latency.

We think about the speed of light being practically instantaneous, but it’s very much not. Depending on how high up these datacenters are parked, access latencies will be on the order of 20-40ms for low Earth orbit and upwards of 600ms for Geostationary satellites. That’s an eternity compared to terrestrial datacenter networks.

• All eyes on markets for AI Bubble Watch: Is it a Floater or a Popper?
• Microsoft CTO says he wants to swap most AMD and Nvidia GPUs for homemade chips
• OpenAI ropes in Korean giants Samsung and SK Hynix to feed its AI megaproject
• Alibaba unveils $53B global AI plan – but it will need GPUs to back it up

While these facilities may be untenable for certain workloads, for other less-latency-sensitive, more power-hungry workloads, they may be ideal, Bezos argues.

“These giant training clusters, those will be better built in space, because we have solar power there, 24/7 … There are no clouds and no rain, no weather,” Bezos said. “We will be able to beat the cost of terrestrial datacenters in space in the next couple of decades.” ®

Amazon founder and executive chair Jeff Bezos says that, within two decades, gigawatt-scale datacenters powered by a continuous stream of photons from the sun will fill Earth’s orbit.

Bezos’ prediction might sound like he’s auditioning for the role of Hugo Drax in the inevitable Moonraker reboot, but the billionaire businessman really believes that spaceborne datacenters are going to be a thing and may eventually outperform their terrestrial counterparts.

“One of the things that’s going to happen in the next – it’s hard to know exactly when, it’s 10 plus years, and I bet it’s not more than 20 years – we’re going to start building these giant gigawatt datacenters in space,” he said during a fireside chat with Ferrari and Stellantis chair John Elkann, at Italian Tech Week.

Of course, to make this dream a reality, we’re going to need heavy lift rockets capable of shuttling the various components to and from orbit, which Bezos’ Blue Origin just so happens to be building.

However, even if we can make putting bit barns in orbit economical, there is no shortage of challenges to overcome. Datacenters are massively complex facilities. Even with an abundant supply of power, an orbital datacenter would still need a way to reject a gigawatt of thermal energy through radiation.

For reference, the ISS’s radiators are capable of rejecting about 70 kilowatts of thermal energy. Bit barn operators will also have to contend with bit-flips on a fairly regular basis unless the hull can be sufficiently hardened against charged particles from the sun and cosmic rays from the outer reaches of space — standard ECC probably isn’t going to cut it.

Many of these challenges are already being tackled at a smaller scale. HPE has been working on the problem for several years now with Spaceborne and Spaceborne-2. These systems have made several trips to the International Space Station already and have tested things like edge compute storage and recovery on long-term space missions.

The first unit, launched in 2017, experienced several failures during its 615-day mission aboard the ISS. As HPE’s Dr Eng Lim Goh told The Register at the time, one of the machine’s four redundant power supplies and nine of its 20 SSDs encountered issues.

Axiom Space has also launched a similar compute prototype to the ISS. In August, the startup’s shoebox-sized Data Center Unit One (AxDCU-1) arrived on the space station. However, all of these designs have been tiny power-optimized designs and are nothing like the kinds of data stations Bezos is describing.

Orbital datacenter advocates expect full scale facilities will require high degrees of automation, including the extensive use of robots to perform maintenance or upgrades.

But while most of the challenges facing spaceborne datacenters are engineering and economic in nature, there’s one that can’t easily be overcome, at least not without bending the laws of physics: latency.

We think about the speed of light being practically instantaneous, but it’s very much not. Depending on how high up these datacenters are parked, access latencies will be on the order of 20-40ms for low Earth orbit and upwards of 600ms for Geostationary satellites. That’s an eternity compared to terrestrial datacenter networks.

• All eyes on markets for AI Bubble Watch: Is it a Floater or a Popper?
• Microsoft CTO says he wants to swap most AMD and Nvidia GPUs for homemade chips
• OpenAI ropes in Korean giants Samsung and SK Hynix to feed its AI megaproject
• Alibaba unveils $53B global AI plan – but it will need GPUs to back it up

While these facilities may be untenable for certain workloads, for other less-latency-sensitive, more power-hungry workloads, they may be ideal, Bezos argues.

“These giant training clusters, those will be better built in space, because we have solar power there, 24/7 … There are no clouds and no rain, no weather,” Bezos said. “We will be able to beat the cost of terrestrial datacenters in space in the next couple of decades.” ®

Amazon founder and executive chair Jeff Bezos says that, within two decades, gigawatt-scale datacenters powered by a continuous stream of photons from the sun will fill Earth’s orbit.

Bezos’ prediction might sound like he’s auditioning for the role of Hugo Drax in the inevitable Moonraker reboot, but the billionaire businessman really believes that spaceborne datacenters are going to be a thing and may eventually outperform their terrestrial counterparts.

“One of the things that’s going to happen in the next – it’s hard to know exactly when, it’s 10 plus years, and I bet it’s not more than 20 years – we’re going to start building these giant gigawatt datacenters in space,” he said during a fireside chat with Ferrari and Stellantis chair John Elkann, at Italian Tech Week.

Of course, to make this dream a reality, we’re going to need heavy lift rockets capable of shuttling the various components to and from orbit, which Bezos’ Blue Origin just so happens to be building.

However, even if we can make putting bit barns in orbit economical, there is no shortage of challenges to overcome. Datacenters are massively complex facilities. Even with an abundant supply of power, an orbital datacenter would still need a way to reject a gigawatt of thermal energy through radiation.

For reference, the ISS’s radiators are capable of rejecting about 70 kilowatts of thermal energy. Bit barn operators will also have to contend with bit-flips on a fairly regular basis unless the hull can be sufficiently hardened against charged particles from the sun and cosmic rays from the outer reaches of space — standard ECC probably isn’t going to cut it.

Many of these challenges are already being tackled at a smaller scale. HPE has been working on the problem for several years now with Spaceborne and Spaceborne-2. These systems have made several trips to the International Space Station already and have tested things like edge compute storage and recovery on long-term space missions.

The first unit, launched in 2017, experienced several failures during its 615-day mission aboard the ISS. As HPE’s Dr Eng Lim Goh told The Register at the time, one of the machine’s four redundant power supplies and nine of its 20 SSDs encountered issues.

Axiom Space has also launched a similar compute prototype to the ISS. In August, the startup’s shoebox-sized Data Center Unit One (AxDCU-1) arrived on the space station. However, all of these designs have been tiny power-optimized designs and are nothing like the kinds of data stations Bezos is describing.

Orbital datacenter advocates expect full scale facilities will require high degrees of automation, including the extensive use of robots to perform maintenance or upgrades.

But while most of the challenges facing spaceborne datacenters are engineering and economic in nature, there’s one that can’t easily be overcome, at least not without bending the laws of physics: latency.

We think about the speed of light being practically instantaneous, but it’s very much not. Depending on how high up these datacenters are parked, access latencies will be on the order of 20-40ms for low Earth orbit and upwards of 600ms for Geostationary satellites. That’s an eternity compared to terrestrial datacenter networks.

• All eyes on markets for AI Bubble Watch: Is it a Floater or a Popper?
• Microsoft CTO says he wants to swap most AMD and Nvidia GPUs for homemade chips
• OpenAI ropes in Korean giants Samsung and SK Hynix to feed its AI megaproject
• Alibaba unveils $53B global AI plan – but it will need GPUs to back it up

While these facilities may be untenable for certain workloads, for other less-latency-sensitive, more power-hungry workloads, they may be ideal, Bezos argues.

“These giant training clusters, those will be better built in space, because we have solar power there, 24/7 … There are no clouds and no rain, no weather,” Bezos said. “We will be able to beat the cost of terrestrial datacenters in space in the next couple of decades.” ®

Amazon founder and executive chair Jeff Bezos says that, within two decades, gigawatt-scale datacenters powered by a continuous stream of photons from the sun will fill Earth’s orbit.

Bezos’ prediction might sound like he’s auditioning for the role of Hugo Drax in the inevitable Moonraker reboot, but the billionaire businessman really believes that spaceborne datacenters are going to be a thing and may eventually outperform their terrestrial counterparts.

“One of the things that’s going to happen in the next – it’s hard to know exactly when, it’s 10 plus years, and I bet it’s not more than 20 years – we’re going to start building these giant gigawatt datacenters in space,” he said during a fireside chat with Ferrari and Stellantis chair John Elkann, at Italian Tech Week.

Of course, to make this dream a reality, we’re going to need heavy lift rockets capable of shuttling the various components to and from orbit, which Bezos’ Blue Origin just so happens to be building.

However, even if we can make putting bit barns in orbit economical, there is no shortage of challenges to overcome. Datacenters are massively complex facilities. Even with an abundant supply of power, an orbital datacenter would still need a way to reject a gigawatt of thermal energy through radiation.

For reference, the ISS’s radiators are capable of rejecting about 70 kilowatts of thermal energy. Bit barn operators will also have to contend with bit-flips on a fairly regular basis unless the hull can be sufficiently hardened against charged particles from the sun and cosmic rays from the outer reaches of space — standard ECC probably isn’t going to cut it.

Many of these challenges are already being tackled at a smaller scale. HPE has been working on the problem for several years now with Spaceborne and Spaceborne-2. These systems have made several trips to the International Space Station already and have tested things like edge compute storage and recovery on long-term space missions.

The first unit, launched in 2017, experienced several failures during its 615-day mission aboard the ISS. As HPE’s Dr Eng Lim Goh told The Register at the time, one of the machine’s four redundant power supplies and nine of its 20 SSDs encountered issues.

Axiom Space has also launched a similar compute prototype to the ISS. In August, the startup’s shoebox-sized Data Center Unit One (AxDCU-1) arrived on the space station. However, all of these designs have been tiny power-optimized designs and are nothing like the kinds of data stations Bezos is describing.

Orbital datacenter advocates expect full scale facilities will require high degrees of automation, including the extensive use of robots to perform maintenance or upgrades.

But while most of the challenges facing spaceborne datacenters are engineering and economic in nature, there’s one that can’t easily be overcome, at least not without bending the laws of physics: latency.

We think about the speed of light being practically instantaneous, but it’s very much not. Depending on how high up these datacenters are parked, access latencies will be on the order of 20-40ms for low Earth orbit and upwards of 600ms for Geostationary satellites. That’s an eternity compared to terrestrial datacenter networks.

• All eyes on markets for AI Bubble Watch: Is it a Floater or a Popper?
• Microsoft CTO says he wants to swap most AMD and Nvidia GPUs for homemade chips
• OpenAI ropes in Korean giants Samsung and SK Hynix to feed its AI megaproject
• Alibaba unveils $53B global AI plan – but it will need GPUs to back it up

While these facilities may be untenable for certain workloads, for other less-latency-sensitive, more power-hungry workloads, they may be ideal, Bezos argues.

“These giant training clusters, those will be better built in space, because we have solar power there, 24/7 … There are no clouds and no rain, no weather,” Bezos said. “We will be able to beat the cost of terrestrial datacenters in space in the next couple of decades.” ®