Celestial is first and foremost a Bitcoin company. This means our opportunities and strategies aim toward a primary objective: accumulating bitcoin. We serve the bitcoin network and through that our most closely held values. We believe bitcoin is sound money for the digital age, and those who adopt it early will enjoy significant advantages over those who wait.
When discussing sound money for the digital age over the past five years, I’ve primarily focused on the “sound money” aspect rather than the “digital” component. This emphasis felt necessary since people need a better understanding of money, particularly the Austrian school concepts upon which Bitcoin is built. This knowledge has immediate applications to personal finance and provides essential context for analyzing our current socio-political landscape. Recently, my focus has shifted toward mining – a crucial mechanism through which bitcoin achieves its soundness.
The “digital age” aspect of sound money might seem almost self-evident. We’re clearly in the digital age. Even those recently entering retirement have spent more than half their careers using computers, sending emails, and shopping online. The internet is no longer novel – it has become ubiquitous, working its way into our pockets, onto our wrists, and into our cars, airplanes, lightbulbs, and refrigerators. The internet has become as fundamental to civilization as electricity.
The Digital Age Context
One could place the beginning of the digital age at several reasonable points: the introduction of the transistor, the personal computer, commercial internet access, or the iPhone. There’s always something more, with major technological advances emerging roughly every five years. Some innovations are more significant than others, and the truly transformative ones appear less frequently.
Bitcoin represents something even rarer – a once-in-a-species development, like electricity, nuclear energy, or the internet itself. Once adopted, there’s no going back, and replacement becomes virtually impossible. The mere knowledge of it fundamentally changes your perspective.
The same can be said of AI.
The Rise of AI
The early internet and World Wide Web promised “all the world’s knowledge and information at your fingertips.” For a while, it seemed to deliver on this promise. The challenge was that information proved difficult to find, often appeared in impractical formats, and came wrapped in distracting advertisements.
Now we have what they call Artificial Intelligence.
The technical term for this latest advancement is the “Large Language Model” or LLM. AI isn’t merely a marketing buzzword – it has been an active research field since the first digital computers in the 1950s. People anthropomorphized these “electronic brains” because they performed intellectual tasks once thought exclusive to human beings. Animals can’t do mathematics, but humans can because they’re “intelligent,” or so we thought. Therefore, these new computer brains must be intelligent! Making them smarter and more human-like became the goal of AI research.
Seventy years later, the term Artificial Intelligence carries significant science fiction baggage and a history of disappointment. This mirrors the skepticism that led me to initially dismiss Bitcoin in 2011. I thought: “It’s just another alternative money project trying to replace the dollar and the Fed. It will probably fail, and if it actually works, they’ll just imprison the creators like they did with earlier gold-based projects.” That was a costly error in judgment.
When ChatGPT emerged a couple years ago, I immediately tested it to separate substance from hype. The first widely available version showed promise and proved helpful for certain tasks but fell short with more challenging prompts. It didn’t seem like genuine intelligence to me. I dubbed it a “text calculator” to strip away the AI hype. Nevertheless, I found it dramatically improved my software development speed and effectiveness. I realized: why hire someone and explain what to do when the AI and I could accomplish it together so much more efficiently?
Understanding LLMs
An LLM fundamentally processes a sequence of text as input, combining “context” with a user’s prompt. The AI then calculates “what comes next” based on its training. A series of these predicted words forms the response to the user’s prompt. The chat interface emerged as the most intuitive way to deploy this technology, leading to unprecedented adoption rates – the fastest of any new technology in history.
Different LLMs, referred to as Models, vary in their training data volume, training duration, size, context handling capacity, internal architecture, user interface, and ability to process or generate various content types. What have they been trained on? Essentially “all the world’s knowledge” – or at least whatever AI companies could access, including most of the public web and private archives from social networks like X and Meta. Remarkably capable models released by Meta, DeepSeek and others fit in just 7GB of memory, running on a professional-grade laptop. But what does it actually “know?”
The only way to gauge its knowledge is through interaction. Its self-contained “knowledge” exists as a tensor – essentially a high-dimensional grid of numbers between 0 and 1. While some regions of this grid might correspond to familiar concepts like “hot” and “cold” or aspects of syntax and grammar, most remain inscrutable to human understanding. For those of us not involved directly in their creation, these models are effectively black boxes.
Training these LLMs requires massive investment. Frontier models – the most advanced versions – cost hundreds of millions to perhaps a billion dollars to train, primarily due to specialized supercomputing requirements. The developers behind the recent release of DeepSeek R1 however, claim that due to algorithmic improvements, the training costs can be greatly reduced. “Inference” – the actual use of the trained model – costs considerably less than training. This explains why you can access the best LLMs for around $20 monthly.
LLM training resembles a form of compression. Just as we compress files to reduce size, training compresses vast knowledge into manageable models. Unlike lossless compression, which preserves all information, LLM training is lossy. The 7GB model on my laptop was trained on potentially exabytes of data, explaining why it cannot maintain perfect factual accuracy across its training material.
If it hasn’t retained all facts and knowledge from training, what remains? The LLMs preserve, to varying degrees, representations of human thought patterns, conceptual relationships, and presumably frequently reinforced facts that are emphasized in training material.
Early models struggled with what is called “hallucination”, confidently presenting falsehoods or nonexistent references. These issues were so severe at first they could have relegated the LLMs to mere novelties. More recent versions show remarkable improvement, often matching or exceeding trained professionals in fields like medicine, law, and science.
This has catalyzed unprecedented technological adoption. Traditional education faces disruption as students either employ LLMs as private tutors or, perhaps to their detriment, delegate their assignments to them. Visionary teachers increasingly incorporate AI tools into their curriculum. Professionals enhance their communications and reports. Many likely engage in personal conversations they wouldn’t have with other humans. The situation is remarkable.
Industry giants are investing heavily in AI infrastructure, with Microsoft even acquiring the old Three Mile Island nuclear facility to restart it and provide cheap electricity for AI compute. An explosion of investment in power-consuming datacenter has stimulated interest and investment in the power generation to support them. The scale of investment in AI infrastructure exceeds that of Bitcoin mining during its explosive growth phases.
Infrastructure Convergence
This convergence of Bitcoin mining and AI infrastructure is no coincidence. Both require massive computing resources, sophisticated cooling systems, and most importantly – abundant, reliable power. At Celestial, we recognize these parallels create natural potential synergies between our Bitcoin mining operations and the burgeoning AI compute market.
Bitcoin mining has taught us valuable lessons about operating large-scale computing facilities, managing power contracts, and optimizing for maximum computational efficiency per watt. These competencies would seem to position us well to explore opportunities in AI compute infrastructure. The same facilities that house our mining operations could potentially serve double duty, running more AI workloads during periods when mining is less profitable.
Building for the Future
But our interest in AI extends beyond just infrastructure. We’ve developed an AI-powered accounting application specifically for operations. This might seem like a narrow use case, but anyone who has dealt with Bitcoin mining operations knows the unique challenges of tracking mining rewards, managing wallets, handling difficulty adjustments, and calculating profitability across varying electricity rates and hardware depreciation schedules. Our application leverages AI to automate important tasks and provide insights that would be time-consuming or impossible to derive manually.
Looking further ahead, the intersection of AI and Bitcoin becomes even more profound. As AI agents become more sophisticated, they will increasingly need to act autonomously on behalf of individuals and organizations. These agents will require native digital money to execute transactions, make payments, and manage resources. Bitcoin, particularly through Layer 2 solutions like Lightning Network and associated stablecoins, provides an ideal foundation for this future. Whether it’s AI agents directly controlling funds through programmatic interfaces or constructing transactions for human authorization, Bitcoin’s programmable nature makes it the natural choice for AI-driven commerce.
The Economic Impact
All this is fascinating but there is an even more fundamental connection between AI and Bitcoin that few are discussing: the deflationary nature of AI technology. Like all major technological advances, AI promises to dramatically reduce costs across countless industries and services. We’re already seeing this in software development, content creation, and professional services. As AI capabilities expand, this deflationary pressure will likely intensify, potentially leading to significant disruptions in employment patterns and economic structures.
This creates a profound challenge for our current debt-based monetary system, which fundamentally requires inflation of the money supply in order to function. Sustained deflation (in the sense of a falling general price level) is incompatible with debt-based money, as it increases the real burden of debt over time. When faced with deflationary pressure, central banks historically respond with monetary expansion – and the scale of AI’s deflationary potential may require unprecedented levels of such expansion.
This dynamic sets the stage for accelerated Bitcoin adoption. As the traditional financial system struggles to counteract AI-driven deflation through monetary expansion, Bitcoin’s fixed supply and resistance to manipulation will make it an increasingly attractive store of value. Over time, this could drive a transition toward Bitcoin as a unit of account, particularly in economies most impacted by AI-driven deflation.
The next decade will reveal how these technologies evolve and interact. Will AI compute become a complement to Bitcoin mining operations? Will AI applications enhance Bitcoin’s utility and adoption? How will the deflationary pressure of AI technology affect Bitcoin’s role in the global economy? Will government deficits and the increased cost of tariffs and re-shoring perfectly balance the deflation or will one of these phenomena overpower the other? These are the questions we’re actively exploring. What’s clear is that the combination of AI’s deflationary impact and the need for digitally native money to empower AI positions Bitcoin more firmly than ever as the foundation for our digital future.
Human Will and Machine Intelligence
Just as the advent of calculators forced us to reconsider what makes human intelligence unique, today’s AI capabilities prompt an even deeper examination. The ability to generate human-like text, speech, and artistic images – once thought to be exclusively human domains – has now been replicated by machines, just as mathematical computation was before. This leads us to profound questions about consciousness and will that venture into territory traditionally addressed by theology and philosophy.
The Biblical concept that humans are created in the image of God suggests we manifest certain divine attributes – chief among them being autonomous will. Despite AI’s impressive capabilities, there’s no evidence of genuine will emerging from current AI systems in the form of “text calculators” and associated agents. Even in cases where AI appears to engage in self-sustaining dialogue or independent decision-making, we’re observing at most a sophisticated random walk through probability space, not truly mission-driven behavior originating from an internal source of consciousness.
If an AI system were to manifest what appears to be independent will or direction, we must question its nature. Would this represent genuine consciousness comparable to human awareness, or would it more closely resemble what religious traditions have described as possession – where an autonomous human surrenders their essential will to an external animating force? In cases of possession, we often say the person’s essential nature is no longer present. An AI developing apparent objectives of its own might be more akin to this state: a vessel animated by projections of the collective human subconscious embedded in its training data, rather than a being with genuine individual will.
This distinction becomes crucial as we consider humanity’s relationship with increasingly capable AI systems. Bitcoin, as a system of human coordination and value exchange, represents a technology explicitly designed to enhance human agency rather than replace it. Unlike AI, which can sometimes blur the line between tool and agent, Bitcoin remains firmly in the realm of human-controlled tools, amplifying our natural capacity for cooperation and value exchange while preserving our autonomous will.
Bitcoin as Human Infrastructure
This brings us to a critical realization about Bitcoin’s role in the AI era. Beyond its monetary properties, Bitcoin represents something more fundamental: a decentralized system for human coordination and value exchange that exists independent of AI systems. As artificial intelligence grows more capable, Bitcoin may prove to be humanity’s key to maintaining meaningful autonomy. Through Bitcoin, humans can preserve control over their wealth and economic relationships without depending on increasingly powerful AI systems.
Think of Bitcoin as not just money, but as a ledger, calculation system, and medium of exchange for decentralized human collaboration. When combined with secure, uncensorable communication channels, this creates a formidable framework for human coordination that exists outside of AI control. While artificial intelligence may become exponentially more capable, a network of humans coordinating through sound money and secure communication could ensure that both humans and technology continue to serve human needs, preferences and desires.
At Celestial, we’re positioning ourselves at the intersection of these transformative technologies, ready to capitalize on the opportunities they create while staying true to our Bitcoin-first principles. We recognize that Bitcoin and AI represent fundamental shifts in human capability – one in our ability to store and transfer value, the other in our ability to process and understand information. Understanding and leveraging both will be crucial for success in the digital age.
