The world is in the middle of a compute arms race. Data center construction is now rivalling office development in total spending, fuelled by the AI boom and a flood of capital chasing high-performance computing capacity. Bitcoin miners, once defined by hashrate and halvings, are quietly becoming key players in this new industrial build-out. Yet behind the data centers lies a more strategic question, how do these facilities actually make money? In this article, we break down the core business models shaping the AI/HPC economy and how miners are adapting to capture their share of this next wave of digital infrastructure.
The Great Compute Build-Out
The Three Dominant AI/HPC Business Models
In-House Infrastructure
Co-Location Services
Cloud and Compute-as-a-Service
👉 What separates a $1 billion data center portfolio from a $10 billion one isn’t just scale, it’s revenue composition. In this week’s Premium section, we break down how investors actually value compute infrastructure and why business models now drive valuation more than megawatts.
Why Revenue Models Matter to Investors
How Business Models Shape Competitive Advantage
How Investors Value Compute Infrastructure
The Great Compute Build-Out
According to the U.S. Census Bureau, data center construction is on pace to overtake office construction in the coming months, a milestone that would’ve sounded absurd just a few years ago. For perspective: in 2022, U.S. office construction outspent data centers by a factor of seven. Today, the two lines are nearly converged, and data centers are about to pull ahead.
This shift tells a bigger story. The world’s next industrial boom isn’t made of steel or oil, it’s made of compute. The capital flowing into AI and high-performance computing (HPC) infrastructure rivals past waves of industrialization. And Bitcoin miners, are positioning themselves right in the middle of it.
In a recent JPMorgan report, analysts noted that publicly listed mining firms have decoupled from Bitcoin’s price. Since July, their market caps have risen sharply while Bitcoin has traded sideways. The reason: many are pivoting to AI infrastructure, which offers steadier returns and higher margins than mining alone.
The market has started to re-rate these companies based on their AI potential rather than their Bitcoin exposure. The correlation between mining stocks and Bitcoin has “clearly broken down.”
This decoupling reflects a deeper economic reality: the post 2024 halving era has made pure Bitcoin mining a tougher business. Miners are seeking ways to repurpose power and capital into more predictable compute revenue and AI/HPC is where demand is exploding and margins are shifting.
The Three Dominant AI/HPC Business Models
Data center might look the same from the outside but under the hood, the revenue mechanics differ dramatically. Understanding these models is essential, because they define how capital turns into cash flow, how risk is priced, and how scalability is achieved.
1. In-House Infrastructure
Tech giants like Amazon and Tesla run their own AI/HPC data centers to power internal operations and cloud services. This model is similar to a self-mining business whereby a company is mining Bitcoin for their own account by owning and operating the ASIC mining hardware. This model is vertically integrated: full control, full CapEx, full benefit. But it’s very expensive. Building out AI/HPC facilities can be 10–20× more capital-intensive than a Bitcoin mining site, and far less modular. For this reason, only deep-pocketed enterprises can justify owning and operating such infrastructure at scale.
2. Co-Location Services
The co-location model — the closest cousin to Bitcoin hosting — leases space, power, and cooling to clients who install their own AI or HPC hardware. Operators provide the physical environment and network connectivity; customers bring their GPUs. The operator earns a steady, utility-like income stream from long-term contracts. Margins depend on scale, density, and service layers. Some offer “bare-metal” racks only, others provide remote hands, compliance management, and on-site monitoring.
Co-location is becoming the strategic bridge between the mining world and the AI/HPC economy. Like traditional hosting, it monetizes infrastructure without taking direct exposure to workload risk.
There are three main co-location tiers:
Retail Co-Location – These are small tenants renting racks. They provide a high yield per kilowatt but demands more operational involvement. In general contracts are relatively short. It is comparable to high-yield, high-touch retail ASIC hardware hosting.
Wholesale Co-Location – These are large tenants leasing multi-MW suites. Contract durations are generally between 5 and 15-year and the churn is lower.
Hyperscale / Build-to-Suit – Custom-built campuses for giants like Microsoft or Google. Tenures can stretch 10–20 years or longer and include performance clauses. The margins are thin but the scale enormous. Hyperscale deals often take two forms. One form a lease where the hyperscaler rents a portion of a campus or completed facility. The other type are so-called build-to-suit partnerships where the developer finances and constructs a site to the hyperscaler’s precise specifications. Once deployed these clients rarely churn.
Each layer serves a different purpose in capital structure. Retail drives yield and margin, wholesale provides lower margins but stable, predictable cash flow and hyperscale unlocks financing leverage.
For miners entering AI/HPC, co-location offers a natural transition path. They already understand modular build-outs, power procurement, uptime management, and customer hosting. The difference lies in the density, complexity, and contractual depth of HPC customers. Some operators also apply hybrid models, dedicating a part of a site to retail or wholesale co-location, another to GPU-as-a-Service, selling compute cycles rather than physical space.
3. Cloud and Compute-as-a-Service
This model removes the need to own hardware altogether. Instead of renting racks or rooms, customers pay directly for compute capacity. For Bitcoin miners, this concept isn’t entirely new. Cloud mining has been around for more than five years, offering short- to mid-term contracts where investors rent hashrate rather than buy ASICs. It’s a way to gain exposure to Bitcoin mining and its potential rewards, without investing directly in infrastructure or equipment. The same logic applies to AI/HPC, where customers lease GPU power instead of deploying their own clusters.
Renting out computing power usually is in the form of Pay-Per-Use. This can be in a lease or rental model, customers lease HPC hardware, infrastructure, or software licenses for a fixed period, typically on a monthly or yearly basis. Alternatively customers pay for AI/HPC resources based on their usage (similar to how consumers pay for utilities like electricity). These are the three most common cloud computing services offered in AI/HPC:
IaaS (Infrastructure-as-a-Service): Pay-as-you-go hardware access (AWS, Azure, Google Cloud).
PaaS (Platform-as-a-Service): Development environments with integrated AI tools.
AIaaS (AI-as-a-Service): Pre-trained models or APIs available on demand.
This pay-as-you-go model has drawn Bitcoin miners who see an opportunity to monetize their power and data-center expertise in new ways. Hut 8, for instance, has launched a GPU-as-a-Service business under its Highrise AI subsidiary, deploying around 1,000 NVIDIA H100 GPUs in a Chicago data center to serve enterprise AI clients. A pivot aimed at generating steadier cash flow next to its more volatile mining rewards.
In 2024 Core Scientific entered into a large-scale hosting agreement with CoreWeave to convert up to 100 MW of its infrastructure and deliver approximately 70 MW for GPU/HPC workloads under a 12-year contract. Later, a ~$1.2 billion expansion further boosted the contract, raising total HPC capacity with CoreWeave to about 590 MW across six sites. In July 2025, CoreWeave announced an all-stock acquisition of Core Scientific valued at about $9 billion (subject to close in Q4 2025). This effectively transforms the combined business into one of North America’s largest AI compute providers.
On the upper end of the spectrum, IREN’s $9.7 billion deal with Microsoft underscores just how valuable these transitions can become. Under the agreement, Microsoft will access NVIDIA’s GB300 GPUs housed in IREN’s Texas campus, with phased deployments through 2026 and a 20 percent prepayment structure. For IREN, originally known as a large-scale Bitcoin miner, the contract represents an estimated $1.9 billion in annualized revenue and a clear validation that former mining infrastructure can serve as a foundation for hyperscale AI demand.
Together, these examples show how “compute-as-a-service” models are dissolving the old boundary between physical infrastructure and digital workloads. What began as mining companies operating ASICs are now evolving into full-stack compute campuses, capable of serving both decentralized finance and centralized AI, depending on which market offers the better yield.
👉 What separates a $1 billion data center portfolio from a $10 billion one isn’t just scale, it’s revenue composition. In this week’s Premium section, we break down how investors actually value compute infrastructure and why business models now drive valuation more than megawatts.
Why Revenue Models Matter to Investors
How Business Models Shape Competitive Advantage
How Investors Value Compute Infrastructure
