Why the Future of AI Depends on Sustainable Energy

The future of artificial intelligence seems limitless but the energy required to build this future is anything but. During a panel discussion in San Francisco featuring Constantijn van Oranje, investors, chip designers, and entrepreneurs, one theme became painfully clear: AI scales faster than our energy grid can handle.

The Energy Challenge of Scalable Intelligence

AI models are not only becoming larger but also smarter. We are moving from generative AI (producing text and images) to reasoning models systems that make decisions, execute tasks, and ultimately control physical actions. Think of robots, self-driving cars, or surgical assistants. However, this shift requires three to five times more computing power than current foundation models.

And computing power means energy. A lot of energy.

A projection mentioned during the panel indicated that if OpenAI continues its current growth trajectory, by 2033 the company will require as much energy as the nation of India. This is not a dystopian exaggeration it is a realistic scenario if we continue to scale using our current methods.

Chips, Photonics, and Memory near the Processor

The hardware world is feeling the pressure. As one of the speakers noted: “Most energy isn't spent on the calculation itself, but on moving data.” Consequently, work is underway on new architectures where memory is placed closer to the processing core, and on optical interconnections via photonics to make data traffic faster and more energy-efficient.

Innovations are also occurring on the cooling side: from liquid cooling to new materials that dissipate heat more effectively. Nevertheless, the consensus is that these improvements are incremental and that something fundamentally different must happen to truly break through the energy demand.

The Energy Bubble: The New Ceiling of Innovation

Fabrizio del Maffeo, CEO of Axelera AI, painted a different picture: it is not the AI bubble, but the energy bubble that will burst. In Taiwan, for example, the energy supply is reaching its limit while demand for data center capacity grows exponentially. In the US, data centers now consume more than 5% of national electricity a percentage that is rising rapidly.

AI, therefore, does not just have a compute problem; it has an energy ceiling.

New Architectures, Old Laws

Still, there is optimism. New chip architectures, such as the Mamba model and quantization techniques (where calculations are performed at lower precision), are driving massive efficiency gains. While Nvidia’s older GPUs worked with 32-bit calculations, modern AI chips already run on 8-bit or lower, with minimal loss of quality.

This means that models can become smaller and more energy-efficient without their performance declining. Furthermore, new hardware categories are emerging, specifically tailored for tasks from edge devices to specialized AI accelerators.

Solar Energy, Batteries, and Direct Current

Entrepreneur and investor Sid Sijbrandij of Gitlab and Kilo Code put an important point on the table: technology alone will not solve the energy issue. The energy infrastructure itself must be redesigned.

His vision: the future is solar and batteries. Fusion reactors are too far off; nuclear energy is too slow to build. Solar energy does offer scalability, provided we approach it intelligently.

His plan:

• Day/night cycle: buffering via batteries.
• Seasonal cycle: overproducing in summer and making that energy available in winter via storage or conversion (for example, into fuels).
• New infrastructure: switching from alternating current (AC) to direct current (DC), directly from solar panels to batteries and AI accelerators without conversion losses.

According to him, the key lies not in more energy, but in less waste.

What's Next?

The race to scale artificial intelligence is a contest between software innovation, hardware architecture, and energy efficiency.
The future of AI will be determined not just by who trains the smartest models, but by who can feed them the most sustainably.

The conclusion from San Francisco: “The biggest breakthrough in AI will not come from Silicon Valley, but from a power plant.”

June 2026 update: more transparency, uneven distribution

Two recent publications sharpen the picture of AI's environmental load.

UN report on the environmental cost of AI (June 2025). Researchers from a United Nations think tank conclude that AI already accounted for roughly one-fifth of data centre electricity use last year, and that AI water consumption is substantially higher than earlier estimates by the International Energy Agency. Around 90% of global AI compute capacity sits in China and the United States, while countries in South America and Africa mainly bear the downsides through raw material extraction and climate damage. Making AI more energy-efficient only helps so much: the Jevons effect lowers prices and pushes usage back up. Source: NOS on the UNU report.

Amazon discloses its water use for the first time (June 2026). Amazon's global data centres consumed 2.5 billion gallons of water (≈ 9.5 billion litres) in 2025, at a rate of 0.12 litres per kWh a 2% drop from 2024 despite continued expansion. Amazon says its data centres rely on air cooling about 90% of the time, switching to evaporative water cooling only during the hottest hours. Important caveat: this figure excludes indirect water use at the power plants supplying electricity and at construction of new data centres. Source: The Verge Amazon's data centers used 2.5 billion gallons of water last year.

The through-line: more openness from tech vendors helps, but the absolute environmental load keeps growing. Picking models deliberately, asking suppliers for water and CO2 figures, and looking beyond the meter inside the data centre all remain essential.

// About the author

Remy Gieling

Mede-oprichter, AI-expert & bestseller-auteur

Tech-expert (1988) gespecialiseerd in kunstmatige intelligentie en mede-oprichter van ai.nl, The Automation Group, Proxies en eBrain.ai. Oud-hoofdredacteur van diverse zakenmerken en daardoor een geoefend verteller op het podium en in de media. Verzorgt jaarlijks 150+ AI-keynotes in binnen- en buitenland en is gastdocent aan Nyenrode. Co-auteur van zeven boeken, waaronder 'Handboek AI Strategie' en 'AI Agents', en bekend als presentator op radio en RTL Z. Reist langs de labs van OpenAI, Nvidia en Tencent en vertaalt de nieuwste doorbraken naar inzichten die leiders direct kunnen toepassen.

LinkedIn