The role of software engineers in the AI datacentre fabric
When software engineers think about AI, the conversation almost always gravitates toward the compute side: GPUs, model architecture, parallelism frameworks and inference optimisation.
But the reality is far bigger, argues Melton Chang, EVP, for Power Systems at Schneider Electric.
Chang guest posts for the Computer Weekly Developer Network to explain how software application development engineers just might be the unexpected heroes in solving AI’s looming energy crisis.
Telling us that AI performance now depends not only on GPUs and training software, but equally on the datacentre facility and the systems that keep that compute powered, cooled and reliable, Chang insists that AI represents a completely new category of energy demand. It changes how datacentres are planned and built.
Instead of worrying about how many racks can fit in a room, operators now run against limits in power supply and grid stability.
Chang writes in full as follows…
This shift elevates the importance of software that manages the full energy ecosystem, including volatile server workloads, power quality & distribution, as well as onsite generation, storage and grid integration. The world needs smarter energy systems, not just more capacity. Software innovation is now critical to the future of the entire AI ecosystem.
Breaking the silos
Datacentres bring together highly specialised teams: electrical engineers, IT architects, network and security specialists and operations staff – each often using different tools and data models. This fragmented structure makes it difficult to form a unified view of the facility. Software engineers play a critical role in connecting these disciplines by creating the integration layer that consolidates their systems into a single, continuously updated digital twin.
[As readers will know] a digital twin is a virtual model of the datacentre’s electrical infrastructure that reflects how the real system behaves over time. Platforms such as ETAP aggregate live data across power and facility systems, giving operators a unified view, predictive insights and the ability to explore “what‑if” scenarios that guide capacity planning and reduce risk.
Software engineers enable this by integrating Building Management Systems (BMS), Energy Management Software, protection relays and other operational technologies into one coherent environment.
Replacing assumptions with simulation
Before the electrical infrastructure was installed, the digital twin enables cross-disciplinary co-design that was previously impossible. For power systems, software engineers contribute to the design phase with cutting-edge software to create detailed digital twin models enabling comprehensive analysis and simulation of the datacentre’s entire electrical infrastructure for short circuit analysis, load flow analysis, arc flash studies and energy usage.
The most powerful capability of the digital twin is the ability to test scenarios that cannot be tested on a live production system. This is where the software engineer becomes the indispensable bridge between domain experts and actionable insight.
The power of data management
Datacentres now generate vast streams of real-time operational data from switchgear, transformers, UPS systems and cooling equipment. AI models built on these data streams evaluate equipment health, predict failures and reduce downtime. Software engineers strengthen this ecosystem by developing data pipelines, anomaly detection models, diagnostic tools and visualisations that operators rely on to run facilities efficiently. Better software translates directly into higher accuracy, lower energy waste and greater resilience.
As grid constraints grow, datacentres increasingly need flexible operating strategies to reduce their impact during peak demand periods. AI-driven facilities now behave like large, fast-changing electrical loads: GPU clusters can swing from idle to overload within milliseconds, creating sharp, unpredictable fluctuations in power demand that traditional grid and facility controls were not designed to handle. At the same time, the International Energy Agency (IEA) highlights that AI‑driven datacentres are increasingly integrating higher shares of renewable energy with renewables expected to supply about 50% of all datacentre electricity by 2030 as wind and solar capacity expands. This shift underscores how software-driven coordination across energy assets can help operators improve sustainability while managing rapidly growing compute demand.
As operators add renewable generation, battery systems and onsite power, datacentres increasingly function as microgrids. Managing these hybrid environments requires advanced control software that can balance rapid swings in load and generation. Engineers skilled in AI, distributed systems and cybersecurity build orchestration layers that optimise energy flows and coordinate renewable and backup assets. These software-driven capabilities are becoming essential for next‑generation AI datacentres to remain resilient, sustainable and aligned with evolving grid stability requirements.
Power systems are evolving like software
Imagine if a software developer were told that every time an application needed a bug fix, a physical server had to be replaced. Those updates required travel, manual rewiring and weeks of testing, all while users stayed online with zero tolerance for failure. That is effectively how today’s hyperscale electrical systems still operate. Modern datacentres routinely exceed 100 MW, functioning as private medium‑ and low‑voltage grids.
Chang: Software engineers play a critical role in connecting [technology] disciplines by creating the integration layer that consolidates systems into a single, continuously updated digital twin.
Now imagine applying the same principle that transformed computing: virtualisation. Instead of protection and automation logic living inside individual relays and controllers, that intelligence is abstracted from the hardware that enacts it. Protection schemes, load priorities and automation workflows become software components. They can be deployed remotely, version‑controlled, validated, updated and rolled back just like any other critical application.
Cybersecurity patches that once rolled out site‑by‑site over months can now be applied consistently across an entire fleet in hours. This shift fundamentally changes how electrical systems are built.
During construction, the complete behaviour of the power network can be modelled and exercised in a high‑fidelity digital environment before any physical equipment is installed. Faults are injected. Islanding transitions are executed. Busbar failures are simulated. Load‑shedding strategies are stressed and refined. By the time switchgear arrives on-site, the logic controlling it is already proven. Commissioning becomes a deployment exercise rather than a live‑equipment experiment, dramatically compressing MV timelines and reducing startup risk.
A structural change is needed
Operations benefit even more. Testing no longer means interacting with energised gear. New strategies can be evaluated safely. Failure scenarios can be rehearsed repeatedly. Control logic can evolve continuously without introducing instability. The electrical system becomes observable, testable and improvable, behaving less like fixed infrastructure and more like a managed software platform. The result is a structural change in how power systems scale.
Electrical infrastructure is no longer a rigid constraint defined by hardware limitations. It becomes a flexible, software‑managed foundation that accelerates datacentre build schedules, reduces human exposure and operational risk, enables participation in grid‑flexibility programs and adapts to rapidly changing load profiles. Just as virtualisation unlocked the cloud era, this abstraction of power‑system intelligence lays the groundwork for the next generation of compute. In an AI‑driven world where density, speed and resiliency define competitiveness, power systems must evolve beyond hardware‑bound logic and start thinking like software.
AI productivity opportunity, if energy scales
AI promises unprecedented advances in productivity, industrial capability and scientific discovery. But energy agencies, policymakers and economists agree – AI’s future depends on energy systems that can scale as quickly as compute demand grows. We’ve reached a point where energy challenges can’t be tackled by one group alone. The next big steps will come when software developers, grid engineers and operations teams pool their strengths to create smarter, more adaptable, software-led energy systems built for the demands of an AI‑driven world.