AI Data centre expansion is being limited by electricity and site selection, and network transmission delays are beginning to become new constraints. Relativity Networks announced the completion of $22 million in SAFE financing and stated that it had received a follow-up order of $40 million from a large, super-large cloud manufacturer to attempt to enter the market with an empty core fiber.

Financing synchronized with orders

This financing was provided by agencies such as Rhapsody Venture Partners, Bell Ventures Inc. and Fast Than Glass LLC. SAFE is an early and common financing tool for start-ups, which is usually converted into shares in subsequent pricing wheels.

In addition to financing, the company received an order of $40 million. The next line is described as a head super-massive cloud servicer, but the report does not disclose a specific name.

Main empty core fibre-optic scheme

Relativity Networks develops empty core fibre. Such techniques are not common and, unlike fibre optics, which traditionally transmit light signals via glass, they allow the light to be transmitted in the cavity of the centre of the fibre, and are therefore closer to the PV theoretical upper limit.

According to the company, such a programme would allow data to be transmitted about 30 per cent faster than traditional fibre optics. Based on these measurements, the time of signal transmission 1 km is approximately 5 microseconds in the traditional fibre-optic; when replaced with an empty core fibre, it can be reduced to about 3.5 microseconds.

  • Traditional fibre optics: about 5 microseconds/ km
  • Empty core fibres: about 3.5 microseconds/ km
  • Time-delayed: about 30%

Aim for cross-campus, AI, interconnectivity.

With the expansion of the AI training and reasoning system, computing is no longer limited to single GPU racks. Today, a data centre park may span hundreds of acres, spread over dozens of buildings, and the physical distance between GPUs increases.

Relativity Networks believes that opportunities are mainly deployed in multi-campus settings. That is to connect the original data centre to operate as a unified system. This has been done against the background that developers often need to decentralize the deployment of computing power according to existing power conditions and available room locations, rather than concentrating all equipment in one location.

Jason Eisenholz, the CEO of the company, stated that the larger AI system was currently spreading the algorithm to a number of parks to take advantage of existing power resources, but that the facilities still needed to work together like a synchronized machine.

Under this logic, if the transmission is reduced by about 30 per cent, the acceptable physical distance will be increased by about 30 per cent before the developer reaches the same time extension threshold. This means that the geographical elasticity of the data centre may be increased and partially relieves the siteing pressure of AI infrastructure expansion.