In January 2026, SpaceX filed an application with the FCC to put up to one million satellites in low Earth orbit. Not for internet. For AI compute. Orbital data centers, powered by near-constant solar energy, designed to process artificial intelligence workloads at a scale that would dwarf every terrestrial data center combined.
In the filing, SpaceX described this as "a first step towards becoming a Kardashev II-level civilization — one that can harness the sun's full power."
That sentence stopped me. Not because of the ambition — Musk has never been short on that — but because of the framing. SpaceX is building intelligence infrastructure and calling it an energy play. I think they have it backwards. And I think getting the framing right matters more than most people realise.
The Kardashev Scale, Briefly
In 1964, Soviet astrophysicist Nikolai Kardashev proposed a scale for classifying civilisations based on energy consumption:
- Type I — a civilisation that can harness all the energy available on its home planet. Solar, wind, geothermal, nuclear, tidal — everything. Full planetary energy control.
- Type II — a civilisation that can capture the entire energy output of its star. Think Dyson spheres.
- Type III — a civilisation that commands the energy of an entire galaxy.
Carl Sagan later extended the scale into a continuous formula, placing humanity at approximately 0.73 — not even Type I yet.
A 2023 study published in Scientific Reports used machine learning to forecast our trajectory. The conclusion: at our current rate, humanity won't reach Type I for millennia. By 2060, we'd barely move the needle — from 0.7276 to 0.7449.
The scale is logarithmic. Getting from 0.73 to 1.0 requires roughly a tenfold increase in total energy consumption. And the maths is unforgiving: fossil fuels are finite, wind power is insufficient by orders of magnitude even at maximum theoretical extraction, and geothermal and tidal are negligible at Type I scale. Only two energy sources have the theoretical capacity to get us there: nuclear fusion and solar.
So if you take the Kardashev Scale at face value, the path to Type I is an energy problem — and it's one we're losing.
The Assumption Worth Questioning
But here's the question I keep coming back to: is energy actually the bottleneck?
Kardashev wrote his paper in 1964. At that time, intelligence was essentially fixed. It was human brains. You couldn't scale them. You couldn't network them beyond the speed of human communication. You couldn't replicate them. The only variable you could meaningfully scale was energy — so it made sense to use energy as the proxy for civilisational advancement.
That assumption no longer holds.
We now live in a world where intelligence itself is becoming a scalable resource. Large language models, autonomous systems, reinforcement learning agents — these are not just tools. They represent a fundamental shift in what can be scaled. For the first time in human history, the ability to process, distribute, and act on information is decoupling from the number of human brains available to do the thinking.
This changes the Kardashev equation in a way that I think most people are underestimating.
Intelligence, Not Energy
Consider what a Type I civilisation actually does. It doesn't just consume energy — it coordinates at planetary scale. It controls weather systems, manages global resource allocation, optimises infrastructure across continents, and responds to planetary-level threats in real time.
That's not an energy problem. That's a coordination problem. And coordination is a function of intelligence — the ability to sense, process, decide, and act across billions of nodes simultaneously.
A civilisation that uses ten times less energy but coordinates a thousand times more effectively is arguably more advanced than one that brute-forces its way to Type I through raw energy consumption. The machine learning forecast predicting millennia to reach Type I assumes the old paradigm: linear energy scaling, human-speed coordination, thermodynamic brute force. But if AI enables exponential efficiency gains — optimising energy grids, managing fusion reactors, coordinating autonomous infrastructure — then you can achieve Type I outcomes at a fraction of the energy Kardashev assumed necessary.
The real ladder isn't energy capture. It's intelligence infrastructure — the capacity to deploy, govern, and scale artificial intelligence systems that make planetary coordination possible.
Put differently:
- The old framing: Type I = capture 10× more energy.
- The reframing: Type I = deploy enough intelligence to coordinate a planet.
If that's true, then the civilisational bottleneck isn't building more power plants or launching more solar-collecting satellites. It's building the intelligence layer — and critically, building the governance layer that ensures that intelligence is deployed safely, effectively, and in a way that actually serves civilisational progress rather than undermining it.
What This Means for What I'm Building
I'm building an AI governance platform called Stoneset — starting with EU AI Act compliance. But Stoneset is the first layer of something larger.
If the civilisational bottleneck is intelligence infrastructure, then the companies that matter most over the next century will be those that build the compute, power the infrastructure, govern deployment, execute intelligence in the physical world, and advance the frontier of what intelligence can do. Prospy is designed to operate across all of those functions — structured as operating subsidiaries that form a single integrated stack.
Stoneset (AI Governance SaaS) — The rules layer. Who gets to deploy AI, under what constraints, with what accountability. Starting with EU AI Act compliance because it's the most immediate, concrete, and commercially viable entry point.
Compliance Services — The advisory and implementation layer. Hands-on work that builds domain expertise and generates cash flow while the SaaS scales. Every engagement is a masterclass in how AI governance works in practice, not theory.
Prospy Compute — GPU cluster leasing for AI training and inference. The substrate of the intelligence economy. Every AI company needs compute access, demand is growing 30–40% annually, and supply remains constrained.
Prospy DC — AI-optimised datacentre facilities. Purpose-built for high-density GPU workloads with advanced cooling and high power density per rack. AI datacentres require 5–10× the power of traditional cloud facilities — and need different architecture entirely.
Prospy Energy — Power generation and procurement for the datacentre operations. Energy is 40–60% of datacentre operating cost. Owning the power supply is the single highest-leverage cost reduction in the infrastructure stack. Every major hyperscaler is investing billions in energy for exactly this reason.
Prospy Robotics — Industrial robotics and autonomous systems. Intelligence without physical agency is just analysis. The transition to Type I requires autonomous systems that can build, maintain, repair, and optimise physical infrastructure at scales humans cannot.
Prospy Capital — The investment arm. Angel investments in AI startups, strategic equity positions in adjacent companies. Every portfolio company is a potential Stoneset customer, Prospy Compute tenant, or acquisition target.
Prospy Labs — Frontier research. Novel model architectures, energy-compute interface systems, fundamental research that pushes the boundary. The longest-horizon play and the one most likely to produce the step-change moments.
When I first sketched this structure, I thought of it as a portfolio — diversified bets across adjacent sectors. But the Kardashev reframing revealed something I hadn't fully articulated: these aren't separate businesses. They're layers of a single stack.
Governance sits at the top because nothing underneath works without it. You cannot deploy intelligence infrastructure at civilisational scale without rules governing what gets deployed, how it's monitored, and who's accountable when it fails. You cannot build autonomous physical systems without a governance framework that society trusts. You cannot run a frontier research lab without a compliance infrastructure ensuring those capabilities don't cause catastrophic harm.
And critically, the subsidiaries form a closed-loop flywheel: energy powers the datacentres, which host the compute, which is rented by AI companies, who need Stoneset for governance, whose revenue funds more infrastructure. Each cycle increases the group's cost advantage and deepens the moat. No other company in the market can offer a single customer AI governance, compliant compute, purpose-built facilities, and clean energy. That vertical integration is the long-term defensibility.
The Sequencing Logic
Vision without sequencing is fantasy. Here's the order and the reasoning:
Phase 1: Software & Services (now) Stoneset and Compliance Services are the starting point because they're commercially viable today, require minimal capital, and generate the revenue, expertise, and customer relationships that fund everything downstream. The EU AI Act is live. Companies need help now. This isn't a bet on the future — it's a response to present demand. Every Stoneset customer is an AI company. Every onboarding conversation reveals how AI systems are actually built, deployed, and where they fail. That intelligence feeds directly into the infrastructure plays downstream.
Phase 2: Compute Entry (near-term) Prospy Compute starts as a GPU brokerage — arbitraging cloud capacity for Stoneset's existing customer base. The structural edge: compliance customers are compute customers. You already know their workloads, their deployment timelines, and their infrastructure needs from the governance work. No other compute provider has that data.
Phase 3: Infrastructure (medium-term) Owned GPU clusters, the first datacentre facility, and the first energy procurement agreements. With governance credibility, regulatory relationships, and real-world deployment knowledge, the infrastructure play becomes defensible. Not "we built a data centre" — but "we built infrastructure informed by hundreds of governance assessments, designed from the ground up for compliant, auditable, trustworthy AI deployment." That's a moat no other operator can replicate.
Phase 4: Vertical Integration (longer-term) Energy transitions from procurement to owned generation. Robotics begins automating datacentre operations internally before expanding externally. The full flywheel becomes operational: energy, compute, customers, governance, capital, reinvest. The competitive moat becomes almost impossible to replicate because you'd need to build governance, compute, datacentres, and energy simultaneously.
Phase 5: Frontier Research (long-horizon) The lab that pushes the frontier. Funded by the cash flows of Phases 1–4, staffed by the best people I can find, and governed by the framework we've spent years building. Novel model architectures, energy-compute interface systems, and fundamental research that makes the Kardashev transition feasible.
The Honest Disclaimer
I want to be transparent about where this plan is versus where it goes.
Stoneset is early-stage. The full stack is exactly that — a vision. Phase 1 is the only thing that's real right now, and even that has a long way to go.
I'm writing this not because the plan is complete, but because I think the thinking matters — and because the best time to articulate a long-term plan is before you're locked into short-term incentives that distort it.
If Stoneset fails, the thesis doesn't. Someone will build the governance layer for AI. Someone will build the intelligence infrastructure that enables planetary-scale coordination. Someone will bridge the gap between where humanity is on the Kardashev Scale and where it needs to be. I'd like it to be me. But the important thing is that someone does it.
The Bet
The Kardashev Scale tells us that civilisational progress is about energy. I think that was true in 1964 and is becoming less true every year. The real measure of a civilisation's advancement is its capacity to deploy intelligence — to sense, reason, decide, and act at whatever scale its challenges demand.
We are at the very beginning of the intelligence era. The decisions being made right now — about how AI is governed, where it's deployed, what infrastructure it runs on, and who controls it — will determine whether humanity reaches Type I in centuries or millennia. Or whether we reach it at all.
I'm building towards being part of that. One layer at a time, starting with governance, compounding towards infrastructure, and always oriented by the conviction that intelligence — not energy — is the resource that will define the next chapter of human civilisation.
That's the thesis. Now back to work.
Bora Acikan — March 2026 London, UK