The AI-Energy Nexus: Solar's Breakout Moment Converges with AI's Insatiable Appetite for Power

The artificial intelligence revolution is reshaping industries and economies, but it comes with a voracious appetite for energy. As AI capabilities expand exponentially, a critical question emerges: Is there an energy bottleneck that could impede further AI advancements?

This Skywert Analysis explores the dynamic interplay between AI’s energy consumption and the race to develop more efficient technologies and energy sources.

The Growing Energy Appetite

Since the release of ChatGPT in late 2022, AI has experienced a meteoric rise in capabilities and adoption. Large language models (LLMs) now increasingly power customer service chatbots, complex financial analysis, and much more.

The AI market is projected to grow at a compounded annual growth rate of 37.7% from 2023 to 2030, reaching $1.8 trillion by the end of the decade.

This explosive growth comes at a significant cost. A single ChatGPT query requires 2.9 watt-hours of electricity, nearly ten times that of a Google search. To put this into perspective, 2.9 watt-hours is approximately enough to charge a modern smartphone from about 15% to 30% battery capacity or to run a 10-watt LED bulb for about 17 minutes.

Goldman Sachs Research estimates that data center power demand will grow 160% by 2030, with AI representing about 19% of this demand by 2028.

The implications are stark. Data centers currently consume 1-2% of global electricity, but this could rise to 3-4% by 2030. In the US and Europe, this surge is expected to drive electricity growth unseen in a generation. According to analysts of Goldman Sachs, US utilities need to invest around $50 billion in new generation capacity solely to support data centers, while Europe faces a €1.65 trillion investment need in grid infrastructure and renewable energy sources.

Energy Constraints

Several prominent figures in the AI industry have highlighted energy constraints as a potential barrier to further AI progress – most notably Mark Zuckerberg.

In the Dwarkesh Podcast, Mark Zuckerberg acknowledged the difficulty in planning around exponential growth curves, stating that “one of the trickiest things in the world to plan around is an exponential growth curve.” This uncertainty hasn’t deterred major investments, with companies willing to spend “$10Bs or $100B+” on infrastructure, betting on continued progress in AI.

While GPU production constraints have eased, Zuckerberg identifies energy as the next major hurdle:

“I actually think before we hit [capital constraints] that, you’re going to run into energy constraints. I don’t think anyone’s built a gigawatt single training cluster yet.”

To put this in perspective, a gigawatt-scale data center would consume energy equivalent to a “meaningful nuclear power plant” or a solar farm (based on today’s solar technology) the size of 3,800 to 5,300 football fields – solely for training AI models.

The energy sector is furthermore heavily regulated, introducing significant delays:

“Getting energy permitted is a very heavily regulated government function [...] If we wanted to stand up some massive facility, powering that is a very long-term project.”

Mark Zuckerberg estimates that building large-scale energy infrastructure could take “many years of lead time,” potentially slowing AI advancements.

Because of that, Zuckerberg speculates on potential shifts in AI training methodologies: “It seems quite possible that in the future, more of what we call training for these big models is actually more along the lines of inference generating synthetic data to then go feed into the model.”

This approach could potentially alter energy requirements and distribution of computing resources.

AI Improvements

We believe that not only will a shift in AI training methods lead to greater energy efficiency, but it is likely that - following historical exponential developments - we will see a parallel revolution in energy efficiency across AI hardware as well as alternative and much more energy efficient AI approaches beyond LLMs.

Specialized AI chips are at the forefront of energy efficiency improvements. Nvidia's A100 GPUs, designed for AI workloads, already offer significant performance-per-watt gains over previous generations. This trend will continue. Startups like Cerebras Systems are pushing boundaries with wafer-scale engines that dramatically reduce energy consumption for AI tasks.

Beyond traditional deep learning, researchers and companies are exploring alternative approaches to traditional deep learning models which address the growing energy consumption concerns. Neuromorphic computing and spiking neural networks (SNNs) stand out as promising contenders in the quest for more energy-efficient AI.

Neuromorphic computing, which mimics the architecture and behavior of the human brain, integrates memory and processing units, allowing for better parallel processing. This approach has shown potential for significant energy savings compared to traditional von Neumann architectures. For instance, Intel's Loihi neuromorphic chip has demonstrated up to 1,000 times better energy efficiency for certain AI workloads compared to conventional processors.

Spiking neural networks, a key component of neuromorphic systems, process information using discrete events or "spikes," similar to biological neurons. This event-driven approach can lead to substantial energy savings, as computations occur only when necessary, rather than continuously. Research suggests that SNNs could achieve energy efficiencies closer to that of the human brain, which operates on a mere 20 watts.

Other alternative approaches include quantum computing for AI, which, while still in its early stages, promises exponential speedups for certain algorithms, potentially leading to more energy-efficient computations in the long term.

While these alternative approaches show great promise they are still in various stages of development and adoption. The energy consumption benefits of these technologies are expected to be significant, but real-world implementations at scale are needed to fully validate their efficiency claims.

Energy-AI Symbiosis

From first principles, we must consider the parallel exponential trajectories of AI’s energy demand and the improvements in energy production and efficiency.

While AI’s appetite for power is growing at an unprecedented rate, the renewable energy sector is experiencing its own revolution.

Solar power costs have plummeted by 85% since 2010 and 99.7% since 1975, with global capacity increasing 2 million-fold. This rapid advancement is driven by positive feedback loops: as deployment increases, costs decrease, spurring further adoption.

Nevertheless, photovoltaic materials are still too expensive and inefficient to replace coal and gas completely and battery technology – to store solar technology until needed – isn’t yet cost effective enough.

The real problem we are facing today is that – while the laws of physics suggest that massive improvements are possible – scientists have made little progress exploring the enormous range of chemical possibilities.

This is where AI comes in. AI is already accelerating innovations in energy technology, optimizing solar cell design, batteries, and grid management by evaluating billions of chemistries in simulation. This will accelerate dramatically.

Ray Kurzweil explained that “in all of history until November 2023, humans had discovered about 20,000 stable inorganic compounds for use across all technologies. Then, Google’s GNOME AI discovered far more, increasing that figure overnight to 412,000.”

As we’ve described in our Skywert Solar Technology Report, once AI becomes significantly smarter and more capable, it will be able to discover and select optimal materials enabling a quantum leap in solar and battery technology making solar energy abundant and virtually free.

Not only in solar technology, in the realm of nuclear fusion – long considered the holy grail of clean energy – AI is playing a crucial role in managing the complex plasma reactions necessary for fusion to occur. The recent achievement of fusion ignition at the National Ignition Facility in California brings this technology closer to reality, potentially heralding an era of abundant, cheap energy that could power the next generation of AI systems.

The nexus between artificial intelligence (AI) and energy is not merely a unidirectional relationship of increasing demand, but rather a profoundly interconnected and mutually reinforcing feedback loop poised to revolutionize both sectors simultaneously.

As AI's voracious appetite for power intensifies, so too does its transformative potential to reshape the energy landscape.

This symbiotic relationship manifests in myriad ways. As AI drives down the cost of solar energy, concurrent advancements in robotics drastically reduce labor costs, which in turn diminish the expenses associated with raw material extraction. Simultaneously, AI algorithms are identifying innovative methods to replace scarce and costly rare-earth elements with abundant and easily extractable alternatives. These developments unfold against the backdrop of exponentially improving advanced manufacturing capabilities. In a remarkable self-reinforcing cycle, today's cutting-edge AI chips, despite their energy intensity, are being harnessed to optimize designs for the next generation of even more powerful and efficient processors.

This virtuous cycle, characterized by the simultaneous exponential advancement of multiple technologies, is catalyzing both AI development and the transition to clean energy at a pace that far exceeds what either could achieve in isolation.

Beyond the core technologies themselves, AI's unparalleled ability to optimize complex systems becomes increasingly vital as energy grids grow more intricate with the integration of renewable sources and distributed resources. As the share of intermittent wind and solar energy rises, the challenges of balancing supply and demand, managing grid stability, and optimizing energy storage become more acute. AI emerges as an indispensable tool for navigating this complexity, enabling the creation of intelligent, adaptive, and resilient energy networks.

Skywert Perspective

While the race between demand and supply might appear tight on the surface, we believe this is due to an intuitive linear bias assuming exponential progress in one field (AI) and linear progress in others (energy, material science, robotics, quantum computing, etc.).

However, the convergence of multiple exponential technologies – including AI and solar – suggests that technological advancements may well keep pace with, or even outstrip, the growing energy needs of AI systems.

First, the AI infrastructure is on an exponential trajectory to become more energy efficient. We're witnessing a Moore's Law-like progression in AI chip design. NVIDIA's H100 GPU, for instance, offers up to 30 times faster inference than its predecessor, the A100, while improving energy efficiency. Google's TPU v4 demonstrates a 2.7x improvement in performance/Watt compared to TPU v3. This rapid advancement in chip architecture is just the beginning. As we move from general-purpose GPUs to more specialized AI ASICs, we anticipate even greater leaps in energy efficiency. The ultimate frontier – quantum computing – looms on the horizon, promising computational power that could revolutionize AI processing while potentially reducing energy consumption by orders of magnitude.

Second, significantly more energy-efficient AI systems will meet exponentially improving solar technology, which is delivering more and cheaper energy with every passing day. Since 1975, solar cells have become 99.7% cheaper per watt of capacity, and worldwide capacity has increased by around 2 million times. This isn't merely linear progress; it's exponential growth that mirrors the advancements in AI.

Consider the symbiotic relationship: AI is now being employed to optimize solar cell designs, predict energy output, and improve overall system efficiency. This creates a positive feedback loop that will dramatically accelerate the adoption and efficiency of solar energy.

Moreover, the convergence of AI and energy technologies extends beyond just hardware improvements. AI-driven smart grids are emerging as a powerful tool for optimizing energy distribution and consumption. For example, in 2016 Google's DeepMind AI reduced cooling energy for data centers by 40% – a feat that showcases the potential for AI to not just consume energy more efficiently, but to actively participate in energy conservation.

From a first principles perspective, we must consider the fundamental limits of computation and energy conversion. While current technologies are far from these theoretical limits, the exponential progress we're observing suggests we're on a trajectory to approach them. As we do, we may find that the energy demands of AI systems plateau, while our ability to generate and distribute clean energy continues to grow exponentially.

Historically, we've seen that when two or more exponential technologies converge, the result is often a step-change in capability that defies linear projections. The convergence of AI, advanced materials science, and renewable energy technologies could create a similar inflection point in our energy landscape. Just as the convergence of miniaturization, lithium-ion battery technology, and mobile software led to the smartphone revolution – a development that would have seemed implausible based on linear extrapolations of 1990s mobile phone technology – we may be on the cusp of an energy revolution that will render current concerns about AI energy consumption obsolete.

The Road Ahead

While the race between AI's energy appetite and technological innovation may appear daunting, we at Skywert see this challenge not as an insurmountable obstacle, but as a catalyst for innovation that will drive both AI and energy technologies to new heights of efficiency and sustainability.

While current projections paint a picture of potential energy scarcity, a first-principles analysis through the lens of historical exponential trends presents a far more optimistic outlook than linear projections suggest.

Our Skywert Solar Technology Forecast model, when combined with AI energy demand projections, paints a more nuanced picture. By 2030, we anticipate global installed solar capacity to reach 33 TW, capable of generating over 34,000 TWh annually - more than enough to cover 100% of projected global electricity demand. This exponential growth in solar capacity, driven by continuous efficiency improvements and cost reductions, is poised to outpace even the most aggressive AI energy demand scenarios.

Pessimistically assuming AI data centers consume 10 times more energy than traditional ones, as some studies suggest, we could see AI-related energy demand in Europe reach 400-500 TWh by 2030. However, this linear projection fails to account for the exponential improvements in AI hardware efficiency. The rapid evolution of specialized AI chips, like Google's TPUs and Nvidia's H100 GPUs, consistently delivers order-of-magnitude improvements in performance per watt. We expect this trend to accelerate, potentially reducing AI's energy footprint by 50-70% every 2-3 years.

Moreover, the decentralized nature of solar technology offers a compelling counterpoint to concerns about concentrated power demand. Unlike traditional energy infrastructure, solar installations can be widely distributed, mitigating the need for massive grid upgrades in specific regions. This decentralization aligns perfectly with the distributed nature of edge AI computing, potentially leading to a more resilient and efficient energy ecosystem.

Battery technology is set to play a crucial role in this energy transformation. We project exponential improvements in energy density and cycle life, with solid-state batteries potentially offering 2-3 times the energy density of current lithium-ion technologies by 2030. This advancement will not only support grid stability for intermittent solar generation but also enable AI data centers to operate more efficiently, leveraging energy storage for load balancing and peak shaving.

The synergy between AI, solar, and battery technologies opens up innovative possibilities for massive energy efficiency gains. AI-optimized smart grids could dynamically balance supply and demand, while AI-driven predictive maintenance could maximize the efficiency of solar installations. Furthermore, AI could enable real-time optimization of data center workloads based on solar energy availability, creating a symbiotic relationship between energy supply and demand.

By 2030, we envision a world where abundant, cheap solar and nuclear fusion energy not only meets but exceeds the needs of even the most energy-intensive AI applications.

Opportunities

The convergence of exponential advancements in solar technology, artificial intelligence, and energy storage presents an unprecedented opportunity for astute investors and entrepreneurs to capitalize on the transformative potential of the Energy-AI Nexus. As the cost of solar energy plummets and AI capabilities soar, new multi-trillion dollar markets are emerging that have the potential to reshape entire industries and create immense value.

One of the most promising areas is the development of AI-optimized, decentralized energy systems that seamlessly integrate solar generation, storage, and consumption at the local level.

At Skywert we expect to see hyper-local solutions, where every building becomes both an energy producer and a computing node. A business idea in this space would be to develop a modular, AI-powered "Energy+Compute" unit that combines high-efficiency solar panels, advanced energy storage, and edge AI or crypto mining processors. These units could be installed on commercial and residential buildings, creating a distributed network of energy production and computing power. This network could dynamically allocate resources between selling excess energy back to the grid during peak hours and offering computing power for AI tasks when demand is high or crypto mining, if grid pricing is unattractive and AI computing demand low.

Investors should also pay close attention to startups developing next-generation solar materials and advanced manufacturing techniques. Perovskite solar cells, for instance, have the potential to dramatically boost efficiency and slash costs, but are still in the early stages of commercialization. With solar on track to become the dominant energy source globally, perovskites and other advanced materials could yield 10-100x returns for early investors.

Another overlooked opportunity lies in the application of AI to optimize renewable energy financing and deployment. The solar industry still relies heavily on cumbersome, manual processes for customer acquisition, system design, underwriting, and asset management. Startups like Aurora Solar and Raptor Maps are using AI to streamline these workflows, reducing soft costs and enabling solar to scale faster and more profitably.

Finally, investors cannot afford to ignore the potential for AI to accelerate the development of transformative clean energy technologies like nuclear fusion, advanced geothermal, and green hydrogen production. While still nascent, these areas are attracting increased attention and capital from visionary investors and entrepreneurs. Helion Energy, for example, has raised over $570M to develop a fusion power plant that could provide virtually limitless, on-demand clean electricity. If successful, Helion and others in the fusion space could deliver the ultimate energy breakthrough - and generate enormous wealth for backers.

Conclusion

The exponential growth of AI is set to drive a significant increase in energy demand over the coming years. However, our Skywert perspective suggests that the long-term impact may be less severe than linear projections indicate.

The convergence of exponential advancements in AI infrastructure efficiency, solar technology, and energy storage solutions is poised to mitigate the growing energy footprint of AI systems. As these technologies evolve in tandem, they create a virtuous cycle of innovation, with each driving progress in the others.

Nevertheless, valid concerns remain in the short term. Over the next 6 months to 2-3 years, the rapid proliferation of AI applications and the scaling of AI models will likely outpace efficiency gains. This could lead to significant strains on energy grids and a surge in electricity consumption, particularly in regions with high concentrations of data centers and AI-driven industries. The short-term imbalance between AI energy demand and the maturity of clean energy solutions underscores the urgency of accelerating investments in renewable energy infrastructure, energy efficiency technologies, and grid modernization.