The Age of AI Colonization
How DeepSeek’s Breakthrough Fuels a Compute Boom... and a Climate Reckoning
This article is an expanded take on a LinkedIn post I shared last week in response to DeepSeek’s breakthrough—an innovation that promises to make model training cheaper, faster, and more efficient. Many thought this would lessen AI’s voracious demand, but as we’ll explore below, the real story couldn’t be more different.
Last week, the markets reacted to DeepSeek’s news: training massive AI models is about to become cheaper, faster, and more efficient. Some analysts speculated this innovation might undermine GPU demand, even causing a dip in Nvidia’s valuation. But to assume that lower cost means slowing demand is to ignore one of the most enduring ideas in economics: Jevons’ Paradox. When something becomes more efficient and cheaper, we don’t use it less. We use it more.
In the case of AI, DeepSeek’s efficiency breakthrough won’t reduce the size or number of models being trained, it will further fuel an AI gold rush. We’re about to witness a hyper-proliferation of specialized AI, from internal business units building domain-specific language models to national labs and institutions rolling out custom AI for strategic advantage.
Far from signaling a lull, this is the spark that will ignite the next AI industrial revolution.
Jevons’ Paradox in Action
The principle behind Jevons’ Paradox is straightforward. If the cost of producing or consuming a resource plummets, demand often rises dramatically.
Industrial coal usage in the 19th century is a classic example: cheaper coal lit the fuse on the global industrial boom. The same is playing out with AI:
Cheaper AI Training: DeepSeek makes training costs tumble, removing barriers to experimentation. Expect organizations to train more models in parallel, not fewer.
Broader Adoption: Sectors that were on the fence, due to compute expense, energy costs, or data center constraints, will now jump in.
Runaway Demand: Once AI becomes significantly cheaper, enterprises will find more and more processes to automate or optimize, compounding the demand for compute.
Result: We’re about to see a massive surge in AI experimentation and deployment, from R&D labs to the C-suite.
Corporate AI Models Become the Norm
Imagine a Bespoke Executive Agent (BEA) that has ingested every major decision, product pivot, and marketing strategy your company has ever undertaken. It’s intimately familiar with your brand, your sales data, and even the intangible nuances of your internal culture. When the CFO or CMO wants to test a new proposal, the dedicated AI model can simulate hundreds of scenarios, instantly drawing from decades of proprietary data.
Tailored Organizational Knowledge
Every brand, every business unit within that brand, and quite possibly every product line, will train and deploy its own specialized model. “Horizontal” models like ChatGPT might still exist, but they’ll be overshadowed by an explosion of purpose-built AI that thrives on deep, proprietary knowledge.AI Co-Leadership
We’re already seeing how generative AI can inform human decision-making. As these specialized models become more accessible, the entire decision-making framework of the C-suite will shift toward AI-human collaboration. “Executive agent” isn’t hyperbole—it’s the future, where AI becomes a strategic partner, co-piloting major decisions with near-real-time insights.Localized Data Centers & Micro-Scale Compute
As data sovereignty and control become paramount, companies will increasingly look to build or co-locate their own data centers. It won’t be a race to build a handful of 100-acre server farms but rather a proliferation of smaller, strategically located data centers—some on corporate campuses, some off-grid, each powered by dedicated, reliable energy sources.
The Climate Question: Energy and Water Demands
It’s impossible to discuss a coming wave of AI data centers without addressing the energy and resource implications. Today’s data centers are already energy-hungry behemoths, with a growing reliance on water for cooling. According to some estimates, global data centers consume more than 2% of the world’s electricity (IEA Electricity Report, 2024). That figure is only set to rise with thousands of new, smaller AI installations on the horizon.
Power Intensification
Even if DeepSeek reduces the energy needed per training run, the sheer proliferation of models will drive up net energy use. If you can do something at half the cost, you might be tempted to do it twice as often, or more. Furthermore, it becomes even more habitual and integration into workflow accelerates.Micro-Grids and Smaller Scale Data Centers
We’ll see new modes of data center design, deployment, and operation. Instead of a handful of mega-facilities, you might have distributed networks of smaller, more specialized computing clusters—powered by onsite renewables or local utility partnerships. Efforts like OffGridAI are leading the charge, advocating for off-grid energy solutions that combine solar, batteries, and efficient computing infrastructures to minimize emissions and reduce strain on public utilities.Water Cooling Considerations
With more data centers popping up, especially in arid regions or places where water is already a scarce resource, water usage will become a bigger issue. Innovations in cooling, whether through liquid immersion or other advanced methods, will be essential to mitigate environmental impacts.
National AI Boom: A New (Compute) Arms Race
Nations and large governmental bodies aren’t waiting on the sidelines. They recognize that large-scale AI is a strategic asset, and they’re moving quickly to deploy specialized computing capabilities for defense, energy, healthcare, education, and beyond. DeepSeek’s innovation will only expedite this:
National Labs: Governments will ramp up specialized AI labs, each with unique domain focuses (e.g., climate modeling, advanced materials, cybersecurity).
Secure Infrastructure: Expect a rise in secure, high-power computing hubs designed to meet national security standards and data sovereignty requirements.
We’re effectively looking at a compute arms race, where the ability to deploy advanced AI becomes both an economic and geopolitical advantage.
Systems Thinking: AI as a Self-Reinforcing Ecosystem
From a systems thinking perspective, it’s crucial to see AI’s growth not as a linear progression but as a feedback loop. As AI gets cheaper:
More Data: More organizations adopt AI, generating more data in the process.
Better Models: More data drives better, more efficient models, which fosters more AI adoption.
Infrastructure Expansion: Each new model and dataset triggers infrastructure investments, energy, cooling, hardware, real estate, which in turn opens new capacity for even more AI projects.
This positive feedback loop suggests the AI wave we’re witnessing is only in its infancy.
Climate and Corporate Responsibility: The Path Forward
The irony here is that many AI innovations could help address climate challenges, like real-time optimization for energy grids, predictive analytics for resource management, and advanced climate modeling. Yet, the expansion of AI itself can be carbon & resource-intensive if deployed haphazardly.
Resilient Infrastructure: As corporations break ground on these smaller data centers, we need to plan for resilient energy and water systems, ideally harnessing low-carbon power.
Sustainable Innovation: Tools like AI can help us innovate in carbon capture, supply-chain optimization, and more. But that innovation must be matched by a commitment to net-zero strategies in the data centers themselves.
Collaborative Governance: Regulators, energy providers, and tech companies must work in tandem to set sustainability standards. AI’s potential to transform our world is vast, but it must be wielded responsibly.
Fueling, Not Slowing, an AI Revolution
DeepSeek isn’t halting AI’s momentum; it’s amplifying it. We’re standing at a pivotal moment, one in which the scale, complexity, and ubiquity of AI will broaden faster than many people realize. The biggest shift is not in whether large-scale models get cheaper, but rather in how quickly thousands of new models will come online.
The real takeaway? This is only the beginning. We’ll see an explosion of corporate AI instances, domain-specific AI agents, micro data centers, and robust national AI strategies. Far from an AI plateau, we’re entering an era of AI colonization, one that will transform every corner of the global economy (and this is before even considering the Robotics boom that will be further enabled by the proliferation of these foundational models).
The challenge and opportunity before us is to navigate this revolution with foresight. If we fail to address the energy, water, and carbon implications head-on, we risk deepening existing crises. But if we approach this proliferation with principles-based and systems thinking, we can harness the incredible power of AI to advance sustainability and global resilience: one localized data center, one micro-grid, one specialized model at a time.