AI data centers face an energy bottleneck as Lektra tests distributed GPU model

Competition in artificial intelligence is often measured by the size of models, the availability of chips and the speed at which companies introduce new services. Behind this race, however, lies the physical infrastructure that determines how far the industry can develop: data centers and the energy required to power them.

Every AI service requires computing resources. Training increasingly complex models, running inference at scale and supporting enterprise applications can require specialized servers, graphics processing units and cooling systems. As demand grows, so does the need for facilities capable of hosting these resources and ensuring continuous operations.

The problem is that expanding computing capacity depends on more than access to capital and hardware. Land, electricity connections, permits, grid infrastructure and thermal-management systems are also required. In some markets, these factors are becoming decisive in determining how quickly the sector can expand.

Computing Demand Meets the Limits of the Power Grid

Data centers are energy-intensive facilities. Their expansion can place additional pressure on local electricity grids, particularly when large campuses are concentrated in areas already experiencing high demand.

A review article published in Energies in December 2025 examines the technical challenges associated with integrating AI data centers into electricity systems, including growing power loads, grid planning and demand management. Scientific source.

The problem cannot be solved simply by generating more electricity. Energy must be available where it is needed, grid infrastructure must be capable of delivering it, and new connections must be built within a reasonable timeframe.

A project may have financing and hardware in place but still face delays because an adequate electricity connection is unavailable. Likewise, a site with access to energy is not automatically suitable for a data center: connectivity, security, cooling systems, operational reliability and technical requirements must also be compatible with the intended workloads.

This distinction is important for understanding the economics of AI infrastructure. The cost of a computing service depends on more than the price of GPUs. It must also account for electricity, installation, maintenance, networking, cooling, equipment depreciation and the actual utilization of available capacity.

When some of these costs rise, the competitive advantage may shift toward operators capable of deploying resources quickly, making better use of their facilities and maintaining high utilization rates.

Construction Timelines Are Becoming a Competitive Factor

Large data centers offer significant advantages, including economies of scale, concentrated technical expertise, centralized management and the ability to host substantial amounts of hardware. However, building new facilities can take considerable time, particularly when new electricity connections or permits are required.

For the AI market, the time needed to bring additional capacity online can become just as important as the investment cost. Companies developing models, software and digital services need access to computing resources that match their requirements, without incurring excessive costs for unused capacity or waiting too long for infrastructure expansion.

This is driving interest in alternative models, including specialized cloud providers, distributed infrastructure and computing facilities installed closer to locations where energy resources and usable connections are already available.

Distribution, however, is not a universal solution. An infrastructure made up of multiple nodes must address coordination, maintenance, security, hardware availability and connectivity. Moreover, not all workloads can be distributed interchangeably: some applications require low-latency networks or concentrated computing capacity.

The potential advantage therefore depends on how well the workload matches the type of infrastructure, GPU configuration and characteristics of each site.

Integrating Renewable Energy and Computing Capacity

Another development path involves bringing energy generation and computing demand into closer alignment. Sites equipped with renewable energy facilities and storage systems may offer opportunities to deploy computing infrastructure, provided that suitable technical and economic conditions are in place.

The principle is to make better use of existing resources, rather than treating every increase in computing capacity as a project that must begin with the construction of a large, purpose-built campus.

This approach may provide greater flexibility, but it requires careful assessment. Renewable generation can fluctuate, energy storage has costs and limitations, and AI workloads may require continuous operations. Systems must therefore balance energy availability, reliability and service costs.

The relationship between renewable energy and artificial intelligence also needs to be assessed without oversimplification. Powering a data center with renewable electricity does not automatically eliminate the environmental impact of hardware, construction or overall energy consumption. The outcome depends on the source of electricity, how the facility operates and the actual efficiency of its processes.

The industrial question is whether these models can deliver competitive computing capacity while shortening deployment timelines or improving the utilization of energy resources.

The Computing Business Beyond Major Cloud Providers

The AI infrastructure market creates opportunities at several levels of the value chain. Alongside hyperscalers investing in large campuses, specialized providers operate GPU rental services and companies are developing platforms to coordinate distributed computing resources.

Commercial potential depends on the ability to meet specific needs: rapid GPU availability, competitive pricing, dedicated configurations, geographic distribution or specialized services for particular industries.

For emerging operators, however, commercial demand must translate into operational capacity. A list of potential customers is not equivalent to signed contracts, just as demand for GPUs does not generate revenue until the requested resources are available and being used.

Relevant indicators include the number of GPUs actually installed, utilization rates, the cost per unit of computing, recurring revenue, operating margins and the capital required to expand capacity. It is also essential to distinguish between resources already in operation and projects still under development.

Lektra and the Distributed Infrastructure Model

Lektra, a US startup, operates in this context by developing a platform designed to deploy AI computing nodes at distributed energy sites, particularly infrastructure associated with solar power and storage systems.

According to the profile published on UCapital Dealflow, the model involves partners providing energy, space and connectivity, while Lektra supplies GPUs, cloud coordination and customer-facing services for businesses requiring computing capacity.

The company states that it has already launched an initial operating site in Kearny, New Jersey, and reports demand exceeding the GPU capacity available at the time. The profile also lists a $25 million fundraising target and trailing twelve-month revenue of $4.6 million. These figures are reported in the company's profile and should be distinguished from future results that have yet to be verified.

The project's growth will depend on its ability to install additional GPUs, expand operating sites and meet demand without undermining the economics of the service. It will also be important to assess power reliability, actual hardware utilization and the costs required to keep the platform competitive.

The issue extends beyond a single company. As demand for artificial intelligence increases, competitive advantage may increasingly depend on the ability to combine hardware, energy and infrastructure into economically sustainable models. The challenge is not simply to build more data centers, but to make the right computing capacity available in the right place at the right cost.

Lektra — Energy-native AI infrastructure platform