Scaling AI innovations is increasingly being hyped as the path toward global prosperity, enterprise value, and even sustainability. And there are merits to the argument that overcoming limitations in chips, AI models, and data quality might help us achieve these things faster. However, Caspar Herzberg, CEO at industrial software firm AVEVA, argues that the biggest constraint today is energy, specifically the wiring up of the grid required to scale AI infrastructure – at least in the West.
This is less of an issue with China for various structural and investment reasons. Against the backdrop of plans to rapidly build multiple gigawatts of new data centers to meet the expected demand for AI services, most theories for powering them suffer from fundamental flaws. Long grid queues; shortages and long delays for carbon-emitting gas turbines, diesel generators & transformers; reliability challenges with wind and solar; disposal issues for nuclear innovations; etc.
China seems to be rapidly scaling the grid infrastructure needed for more and larger data centers, not to mention electric vehicles and electrified industrial processes powered by solar and wind. In contrast, Herzberg argues that Europe has been measuring success related to the grid and its supporting role in empowering the economy (not to mention data centers) in the wrong way:
We tend to celebrate theoretical measures of capacity, such as projects announced, but that doesn’t tell you whether industry can actually plug in and get affordable power. We need performance-based metrics that represent the real-world context of industrial power use.
For various reasons, the pace of electrification in Europe has plateaued over the last decade. China, meanwhile, has had a clean energy and electrification boom, exemplified by a massive surplus of high-voltage power lines, the expansion of high-speed rail, electric cars, batteries, solar and wind infrastructure, and associated products. The Chinese economy accounted for roughly 40% of all clean energy investments in 2024. From a business perspective, this has also led to new challenges for Chinese companies and shakeups through a process of consolidation. Herzberg says:
It’s true that Chinese makers of solar panels and batteries are now facing the pinch of oversupply, however, another way of looking at the situation with China and clean energy is that the extensive supply available has been just what the doctor ordered. It has helped to drive down costs that make the economic case for investment in renewables increasingly a no-brainer for industry.
Herzberg believes there is also a bigger problem that China and other countries face in the rush to sustainability:
The bigger issue regarding renewables is not overcapacity, but intermittency. Digital tools help grid operators keep the system stable, smooth out peaks and generally get more out of the existing grid. In other words, we need intelligent infrastructure, and thanks to digitalization, we finally have the tools to do that.
Herzberg acknowledges there are also adjacent issues with scaling the physical equipment for transforming the electrical properties of wind, power, and battery systems into the form required for maintaining a reliable, resilient, and cost-effective grid:
No matter how you slice it, new generation assets will take time to develop, and as electricity demand soars due to AI, the grid will come under pressure.
The US Department of Energy predicts that US data center use could triple, accounting for 6.9-12% of total energy consumption by 2028. It's not entirely clear that the Trump “Energy Dominance” strategy, which champions coal, gas, and nuclear at the expense of renewables, will keep pace with this growth. At least not in ways that disincentivize consumer use by imposing higher costs, which is not playing out well for building community support for new data centers.
Technocratic challenges
Here is a funny paradox to sit with for a moment. In late 2025, Europe was pricing carbon at €78/ton across the economy, compared to €7/ton in China, but only for power generation. China also only gradually introduced carbon levies on steel, aluminum, and cement in 2024. Europe’s tougher stance makes sense on one level, but guess who is making faster progress toward net zero and by a wide margin? In 2023, China had installed more solar than the rest of the world combined, while also exporting two-thirds of the panels it produced. Depending on your frame of mind, this was either a smarter de-carbonization strategy or an unfair burden on European renewable innovation.
China also benefits from increased consolidation in its energy generation and grid businesses, which might run afoul of anti-competition laws in the West. China has also developed a more systemic statewide planning and permitting process, which might raise concerns about policy overreach, not to mention fears of aggressive eminent-domain overreach associated with securing property for essential infrastructure in the West.
Herzberg argues that it's important to come back to the fundamentals in working out the best policies and frameworks:
At the end of the day, people want power that’s there when they need it, at a price they can afford, and with a low carbon footprint. China is building a ton of capacity, and like Europe, faces issues around grid integration and energy storage. That said, it faces fewer bureaucratic hurdles than the West in bringing renewables online. China has state-owned grid operators and centralized planning, and has taken a highly strategic, long-term approach. Having just returned from China, I met with several industrial customers, and the level of innovation is impressive.
The picture is more complex in Europe, and especially in the UK. For example, wholesale electricity prices are set based on natural gas prices using “marginal cost” or “last increment” models needed to meet full demand. These keep prices artificially high, especially when fossil-fuel price shocks hit markets. As a result, industrial firms in Europe pay about 50% more for electricity than their rivals in China. Herzberg says:
Europe needs to accelerate capacity build-out, along with these larger, more strategic technology investments around grids and storage. Industrial intelligence can help with the energy transition in both the Chinese and European contexts – software analytics to help manage renewable intermittency and curtailments, for example, as well as engineering and digital twin solutions that allow for better, more streamlined design and construction of energy assets – what I call ‘fast energy’ infrastructure.
Europe is indeed making some progress, but despite the war in Ukraine resulting in energy shocks, it's not advancing nearly as quickly as China. For example, in 2024, renewables accounted for 59.4% of electricity production, which is impressive, but that only represented a 2.3% increase from 2024. For context, this number does not necessarily reflect progress in decarbonizing traditionally non-electrical processes such as steel, cement, and chemical production.
Herzberg argues that a fundamental problem in Europe has been the over-focus on crude, quasi-theoretical measures of capacity, such as gigawatts awarded or installed. But this is not indicative of the actual power industry can access or its affordability. What’s needed is aligning grid modernization policy and strategy with a basket of performance-based metrics that reflect the real-world context of industrial power use. European regulators are making some progress toward system-level, performance-based metrics, including electrification rates, curtailments, congestion costs, how well countries are interconnected, and how widely digital tools are deployed.
These broader metrics could also help steward the growth of a more resilient grid. For example, in the wake of the recent blackout on the Iberian Peninsula, most observers pointed to issues around voltage and frequency instability. A broader view would take into account key system-level stressors, such as real-time congestion status. Similarly, at the micro-level, firms need greater intelligence about electrical systems and grid context to make the right investment and energy management decisions.
Opening silos for grid intelligence
However, implementing these more systemic policies requires creating a shared infrastructure for grid intelligence. At one level of abstraction, this means better data gathering, aggregation, and feedback. But it also faces many challenges, including legacy grid infrastructure, cultural inertia, market mechanics, and risk management. The vision is that as these data silos open up and information is shared up and down the value chain, every player across the ecosystem can access the same source of truth. This can boost proactive decision-making to better predict grid operations, optimize power flows, balance loads, and support maintenance.
Herzbers says the biggest issue towards realizing this vision is fragmentation:
Utilities operate a diverse range of SCADA and other relevant systems that create a highly fragmented grid architecture. As we move toward more distributed energy resources (DERs), this is getting even more challenging. Many operators also rely on manual and batch data processes, meaning they can’t track, forecast, manage or dispatch efficiently. So, the role of connected infrastructure, time-series data, analytics, digital twins and energy management tools becomes very clear. Only through digitalization of the entire grid system can we expect to see growth at scale and true resilience of the new system.
He suggests the first step is to optimize existing operations. For example, many efficiency gains can be achieved by better use of existing assets and by improving real-time visibility into power generation capacity flowing into the network at scale, rather than building new physical infrastructure. The data silos where asset information is stored often hide early signs of equipment degradation and process inefficiencies, which can be more costly to fix later.
Another step is to develop tools to better model scenarios that account for uncertainty. This can help companies to build leaner, more agile operations that respond to geopolitical, supply chain and weather fluctuations in real time. For example, when operators can see what would happen if wind dropped by 60% for a week, they can make informed decisions about resource allocation and investment priorities. They have the flexibility to shift as conditions change. For executives, this visibility across the production network highlights new opportunities and provides planning support.
Herzberg walks through a few examples of how this might work in practice:
It’s kind of like when you choose to run your dishwasher at night. Electricity rates are lower when most people are asleep, which saves on your household bill. It’s basic supply and demand, but here it is scaled up to an industrial level. Non-core loads can be shifted to times where there is less competition for grid resources. But this requires a lot of intelligence and the capability to execute changes in a timely manner.
Battery storage (or hydrogen) can act as a buffer, providing incremental power when needed. Strategic load management should not negatively impact utilization; it’s really about optimizing for high utilization of those most critical assets and processes to support a high return on capital. Octopus Energy has started to do this at scale in the UK, with the growth of prosumer energy set to further accelerate this trend.
More sophisticated measures of grid performance are also benefiting transmission and distribution system operators. For example, Enel, one of the largest renewable energy companies, is using industrial intelligence to monitor and optimize its geothermal fleet. This has helped them reduce unexpected outages that would otherwise force fossil-fuel backup generation online.
Similarly, ONS, the national grid system operator in Brazil, has adopted advanced forecasting, real-time dispatch analytics, and asset-performance modelling to improve how the grid manages renewable intermittency. Instead of dispatching based on static schedules, ONS leverages more granular grid-state insights, such as line ratings, regional congestion, and weather-driven variability, to cut dispatch times from hours to minutes. This has raised renewable penetration, improved frequency stability, and lowered system operating costs.
Herzberg explains that digital capabilities are critical across these use cases:
Improved intelligence and situational awareness enable us to make better predictions about grid operations, optimizing power flows, load balancing and maintenance. Take dynamic line ratings (DLRs), which use real-time data like temperature, wind and sun to accurately see how much power it can safely carry in that moment, without causing an outage. That lets you squeeze more out of the existing infrastructure and plug in additional renewables without compromising reliability.
Software-defined grid
One policy problem is that an overnight end-to-end overhaul of physical infrastructure to support net-zero goals is not realistic. The European Court of Auditors has estimated that achieving net-zero goals by 2050 based on current strategies could require €1.994 to €2.294 trillion, exceeding existing planning budgets of €1.871 billion. Herzberg believes we could make up for this shortfall in the short term by harnessing time-series data on grid operations, digital twins for real-time simulation, and AI-powered predictive analytics at scale.
Herzberg says one promising approach for doing this at scale might be improving the standards and implementation of a ‘software defined grid’ akin to existing progress in software defined networks, cloud infrastructure, and OpenRAN for mobile networks:
The complexity of integrating renewables, storage, distributed resources and variable demand simply cannot be managed with copper and concrete alone. The software-defined grid of the future needs a governed, network-aware data foundation which allows grid operators to squeeze far more value from existing assets, accelerate renewable ramp-up, manage distributed energy resources and avoid blackouts or overloads, all while delivering more affordable, reliable power.
Herzberg previously worked in the networking world at Cisco, where he was inspired by the examples set by telecom service providers and enterprise networks. Taking inspiration from these examples, a software-defined grid would abstract the control logic from physical devices, with a focus on virtualization, disaggregation from hardware, vendor neutrality, and infrastructure programmability. This enables greater automation and agility.
Aveva’s parent company, Schneider Electric, recently announced the One Digital Grid Platform, which closely mirrors the idea of software-defined networking but in an energy distribution context. One essential starting point lies in developing a unified data fabric. This could create a flexible orchestration layer that can adapt to new market schemes, new technologies, and the demands of electrified industry and AI, without forcing a rip-and-replace of physical assets. Herzberg explains:
This is really the beauty of software-defined automation applied to the grid. It means we can transform the grid with less capital-intensive hardware (and do so faster). With greater system-level intelligence, we can also ‘sweat the asset,’ and get more value from existing infrastructure. Software increases flexibility, it doesn’t detract from it. I see software-defined automation as being the factor that really unlocks many of these energy tech innovations and allows us to create scale.
My take
The notion of a software-defined grid has merit in accelerating the transition to a net-zero economy, not to mention powering all those theoretical new AI data centers everyone is planning to build very soon and very big. AVEVA and Schneider, its parent company, seem to be making interesting progress toward realizing this vision, not to mention competitors like Siemens, GE Vernova, ABB, and others.
At the same time, I was recently struck by how some European transmission and distribution operators are looking beyond these vendors in their grid planning and automation strategies, focusing on open-source alternatives. This speaks to the need for broader industry collaboration on interoperability, integration, and openness to more fully realize the vision of software-defined grids.
Also, in researching this, I was struck by the significant differences in carbon-pricing markets between China and Europe, and relative progress towards net-zero goals. On the one hand, I can see how suggesting that pricing carbon more cheaply in Europe might be perceived by many stakeholders as not taking the threat seriously enough. But I can't escape the feeling that it's one of the fundamental reasons that China is making far faster progress towards actually realizing sustainability goals rather than just talking about it.
I think it's important to appreciate that the path to net-zero, with or without all the proposed AI data centers, is going to be painful no matter how you slice and dice it. Existing business models and land use will have to be reshaped, and a lot more carbon will be generated as we learn to build new infrastructure more efficiently and sustainably. Many of these pains will be eased to the extent that vendors and the open source community collaborate to enhance open source standards and develop more flexible business models for industry collaboration on software defined grids to accelerate this process.