- Fictitious charges burn up to 45% of energy only to maintain stable power levels
- Skeleton Graphenegpu offers up to 40% more computing with the same hardware
- IA’s workloads are killing energy efficiency, Graphenegpu softens the drawing in real time
As artificial intelligence systems become more demanding, many data centers have been consuming almost twice the energy they technically need.
This excessive use is not due to obsolete system or hardware failures, it is rooted in how the GPU behaves, since its energy demand can be drastically balanced in seconds, from the complete accelerator to the inactive.
To cope, operators often implement fictitious loads, wasted from deliberate energy, to maintain a stable power raffle, but these data centers deliberately slow down the performance of tens of thousands of GPU to avoid power cuts
Fictitious loads mean a massive waste of energy
While this avoids damage and blackouts, it means that up to 45% of energy is lost as heat, it does not have a useful calculation.
Skeleton Technologies now states that it can have a more efficient alternative, one that allows GPUs to function at full capacity without overwhelming the network.
The company Estonia developed Graphenegpu, a peak razor system that uses patented curved graphene supercaders.
Unlike lithium -based systems, these capacitors can respond in just 10 microseconds, absorbing energy during periods of inactivity and downloading it instantly when the GPU is loaded.
The result, according to Skeleton, is the ability to maintain a performance consisting of the GPU without stressing the network or resorting to strangulation.
Their tests suggest that the system can deliver up to 40% more failures (floating point operations per second, using the same GPU, simply eliminating performance sanctions associated with thermal rating and power instability.
“Graphenegpu offers up to 40% more computing with the same energy footprint, while reducing capital and operation costs by reducing network update needs, energy waste and cooling,” said Taavi Madiberk, CEO of Skeleton Technologies.
“Driven by our patented cured graphene, this is a fundamental change in how AI infrastructure can climb, sustainably and economically.”
The company also reports up to a 44% reduction in the energy capacity that data centers must book from the network.
The central unit, the GRAPHENGPU PCS 50, offers up to 80 kW of maximum power in a standard 1ou factor, compatible with the existing and cooled infrastructure or liquid.
It is important to highlight that it avoids lithium completely, using the material based on patented Skeleton in place.
According to Skeleton, this technology has been tested under rigorous hyperscalero grade GPU workloads with positive results. However, it has not been tested independently for the performance and durability of the real world.
The first shipment of this technology will begin in Germany in June 2025. The company also has an American production site planned for the beginning of 2026.
