What will happen to data centers when physical limits are reached?

When PUE drops below one, and each new rack goes into the negative? The answer is already beginning to emerge. The next round of data center evolution is not about increasing density or increasing channels. It is about changing the paradigm itself: a data center is not a building, but an environment, an ecosystem, an organism. In the article, I presented three scenarios for the future of data centers.

Spoiler: none of the concepts described below are fiction. They are all logical continuations of processes already underway.

1. Industrial “green” data center (2030-2035)

In the next 10–15 years, several powerful trends will influence the development of data centers. First of all, the rapid growth of edge and AI loads. More and more data is generated and consumed locally, and calculations are increasingly taking place “at the edge” rather than in central clouds.

Add to this the increasingly stringent sustainability requirements (ESG), the shortage of space in large cities, and the constant race for energy efficiency. As a result, we will get a completely new look for the data center — compact, modular, almost autonomous , and as “green” as possible.

▍ What will it be like?

The new data center will be designed around the principles of sustainability, modularity, and autonomy. Energy efficiency will be a requirement rather than a desirable feature, and the architecture will be adaptive rather than static.

Such data centers will be located right next to generation sources: solar, wind, or tidal power plants. Scenarios are possible where a small modular nuclear reactor (SMR) will act as a backup energy source — compact, stable, and local. Electricity will be supplied directly and distributed across microgrids, and a storage system (BESS) will smooth out peak loads.

According to calculations, such a scheme will reduce the carbon footprint several times over. Offshore and underwater data centers in China are already testing these approaches. By 2035, they will be the norm rather than the exception. Hybrid schemes will be used for uninterrupted operation: microreactors, batteries, and “smart” load distribution systems.

The data center architecture will be micromodular . The modules can be added or moved like elements of a construction set. This will allow building distributed data center networks — from small edge nodes at 5G/6G stations to full-fledged clusters in remote regions.

The computing density will increase by an order of magnitude. Up to 100 kW per rack, including AI accelerators, GPUs, TPUs, and new-generation MLUs. Traditional CPUs will begin to leave the market. The racks themselves will no longer be “dry” — they will be completely immersed in a dielectric liquid that provides uniform, quiet, and extremely efficient cooling.

Natural ventilation and, where possible, direct water exchange with natural sources — rivers, seas, oceans, and underground reservoirs — will be added to the immersion. In coastal or underwater locations, the temperature will be stably maintained at 20–25 °C. This will eliminate the need for bulky air conditioning systems and make the entire circuit energy-neutral.

The usual “corridor with servers” will also no longer exist. The entrance area will turn into an information terminal with an AI interface, through which you can request the status of the data center or visualize its digital model. Live metrics on the wall: temperature, generation, PUE < 1.1, energy balance, forecasts.

▍ How to manage it?

Infrastructure management will be completely digitalized. Each data center will have its own digital twin — a real-time model synchronized with physical equipment. It will not just display the state, but actively manage it: predict failures, reconfigure routes, launch preventive maintenance.

The work will be coordinated by a neural network that optimizes energy consumption, cooling, and load in real time. Human intervention will be minimized. The operator will become an observer — the last line in the decision-making chain.

This is what a new-generation industrial data center will be like: modular, distributed, energy-neutral, and smart. It will become the basis of the digital world, in which data is generated, processed, and stored not somewhere in the “cloud,” but right at the source. There will be fewer and fewer people in it, and more and more intelligence.

2. Space data center (2035–2040)

In the second half of the 2030s, the first wave of orbital data centers will appear — autonomous computing modules located beyond the Earth. They will be powered by solar energy, cooled by a vacuum, and ensure continuous operation regardless of terrestrial conditions. In essence, we are talking about creating a new infrastructure shell around the planet — a distributed “computing ring” independent of climate, politics, and energy crises.

Orbit is not just an original location. It is a fundamentally new level of reliability, autonomy, and scalability. The key advantage is absolute physical isolation. Neither fires, nor floods, nor wars, nor cyberattacks can reach an orbital data center. This was the argument for Lonestar Data, which was the first company to send storage devices with digital archives (including the US Constitution) to the surface of the Moon.

Tech giants are also working on the space direction. Microsoft Azure Space integrates terrestrial clouds with satellite systems, and IBM and Lumen Orbit are exploring ways to reduce AI’s carbon footprint by moving computing off-planet.

▍ What will it be like?

Space data centers will be located in geostationary or sun-synchronous orbit, as well as on the surface of the Moon.

The design of the orbital data center will consist of a chain of modular satellites – rectangular blocks of 12x3x3 meters, covered with ribbed radiators and solar panels, hanging above the equator at an altitude of about 600 km.

Inside the data center are radiation-resistant modules on ARM and specialized AI chips. Each will consume 100-200 kW, a block of ten modules – up to 2 MW, and a full section of five satellites – about 10 MW. This is enough to replace a medium-sized ground data center.

The main engineering asset of such data centers is cooling. The temperature of the vacuum of space at -270.45 °C turns the space into an ideal radiator, and the panels remove heat using infrared radiation – without fans, freons and other terrestrial solutions.

Energy will come from solar panels. Up to 90% of the generation will immediately go to the servers, the rest – to superconducting batteries. There will be no generators or fuel in this data center – only photons, silicon and calculations. In turn, buffering will allow the system to operate for up to 30 days in autonomous mode.

Communication with orbital data centers will be built on optical channels and lasers. The main traffic will go along laser lines to receivers on Earth. However, there will also be a backup path – through the satellite systems Starlink, SES and similar ones.

▍ How to manage it?

These data centers will have an AI circuit on board that diagnoses and reconfigures operations in real time. The ground control center (NOC) will only observe and adjust strategy. All operations — from load balancing to temperature and power control — will not require human intervention.

Based on trends, the space data center will cease to be a concept by 2040. It will become a necessity. Everything that seemed like science fiction — autonomous server satellites, laser communications, radiation-hardened AI accelerators, cryogenic cooling — will become the new norm.

3. Hybrid bioquantum data center (2040–2060)

When silicon starts to slip and energy costs reach their limit, biological, quantum and neuromorphic computing will enter the scene. In one complex – in a single “living” data center.

The growth of computing needs, especially for AI, is already pushing engineers to reconsider architectures. For example, the Swiss startup Final Spark is working on biochips based on organoids – tiny clumps of living neurons grown from stem cells. The idea is that such processors will be able to perform cognitive tasks with a thousand times less energy consumption.

According to the company’s founder Fred Jordan, in 10-15 years we will have the first biocomputer-servers – not as a lab experiment, but as part of the IT infrastructure.

Quantum computing is developing in parallel. Projects like the Quantum Data Centre of the Future in the UK are already integrating photonic quantum chips into conventional data centers. It’s still experimental, but by mid-century the technology could be the basis for hybrid architectures.

Add to that neuromorphic chips—for example, Qualcomm’s Zeroth semiconductors, which work like the brain. They can already combine storage and computing in a single node, eliminating expensive data transfers.

▍ What will it be like?

The hybrid bioquantum data center of the future looks more like a laboratory biofactory than a technology park. From the outside, it will resemble a laboratory aquarium — translucent, with internal lighting that changes depending on the activity of the modules. Inside the data center, there will be capsules with organoids, multilayer optics, cryostat blocks, and bioreactors.

The biomodules will be capsules with organoids, their temperature will be about 37 °C, because the nutrient medium circulates inside. Each such block will be able to process patterns, understand contexts, and generalize — much like an animal’s brain.

Quantum nodes will be placed in cryostats, where the temperature drops to 10–15 mK. Their tasks include quantum logic, encryption, generating ultra-precise predictions, and working with complex probabilistic models.

There will also be neuromorphic units — silicon and polymer chips built on memristors that imitate synapses and neurons. These elements will be able to adapt to input data in real time, essentially learning “on the fly.” According to experts, their operation requires a temperature of up to 50 °C, so active cooling will not be required.

There will no longer be classic servers with a rigid hierarchy. The architecture is modular, decentralized, and each computing unit is simultaneously memory, a processor, and a router. Data will be transmitted via optical channels and nanohighways built into the very structure of the case. Data will become a single part of the computing process, and RAM in the usual sense will no longer be needed.

It is implied that cooling in such data centers is three-circuit. The helium cycle supports quantum modules, liquid channels serve organelles, and air cooling with natural convection is distributed across the neuromorphic part.

▍ How to manage it?

It will not be possible to manage a hybrid data center, because the principle of “monitoring metrics, twiddling alerts, changing configs” does not work here. The entire system will be controlled by a distributed AI trained on logs, failures and behavior patterns of previous generations of the same data centers.

This AI will become the nervous system of the data center — it will itself identify zones of overheating, underload, deformation or fatigue of materials and reconfigure flows without operator intervention. However, there will still be a place for a person — his role will be to be an observer with a neural interface and control the viability of the data center “organism”.

This concept is a kind of “cyborg data center”, where programming begins at the level of materials and cells, and the digital “super leg” is controlled by AI.

Conclusion

The data center of the future will no longer be just a building with servers, because it will be part of an ecosystem – technological, energetic, biological. Everything that we considered basic – silicon, cold air, square meters – will fade into the background. Autonomous modules, self-learning systems, living neuroprocessors and off-planet computing will take their place.

And if today’s data centers are data factories, then the data centers of tomorrow are intelligent organisms built into the fabric of the environment, the planet and even space.