BitDeer: From Bitcoin Miner to "AI Landlord"

Written by: Lin Wanwan

Everyone initially thought that the real bottleneck of AI was not capital, nor large models, but electricity.

Large-scale training is running at full capacity for a long time, and AI inference is operating 24/7, which brings about a problem: there is not enough electricity, and chips are forced to sit idle. The U.S. power grid infrastructure has lagged behind in the past decade, and adding new large loads to the grid can take 2 to 4 years, making “readily available electricity” a scarce resource across the entire industry.

Generative AI brings to the forefront a raw and brutal truth: what's lacking is not the model, but the electricity.

The story took a turn, as cryptocurrency mining companies, the group that first regarded electricity as a “means of production,” began to move from the margins to the center of the capital stage.

Iris Energy (IREN) is a sample of this route. This year, IREN's stock price once surged nearly 600% within the year, with a 52-week range from $5.12 to $75.73. It decisively withdrew power and transformed its self-built AI data center while the Bitcoin surge still held appeal.

When giants like Microsoft come forward with long-term orders worth a total of $9.7 billion, the market intuitively understands for the first time the realistic path of “from mining to AI”: first comes electricity and land, followed by GPUs and customers.

However, not all mining enterprises, like IREN, choose to bet their entire fortunes on AI. In this massive migration of computing power driven by electricity, there is also a steady force worth our attention - Bit Deer's.

Bitdeer Technologies Group (NASDAQ: BTDR), a company founded by crypto legend Wu Jihan and headquartered in Singapore, holds nearly 3GW of power resources spread across the globe, avoiding the shallow trap of relying on others for “power supply” from the very beginning. As the wave of AI arrives, Bitdeer did not choose the IREN-style radical “All-in,” but retained its profitable Bitcoin mining as a “fundamental base,” while steadily upgrading some of its mining farms to AI data centers.

This strategy of “advancing to attack and retreating to defend” makes it the best example of how global players think and plan in observing this competition of computing power.

To this end, we interviewed Wang Wenguang, Vice President of Global Data Center Business at the mining company Bit Deer, hoping to shed light on the current global AI power shortage, and their views on mining companies transitioning to AI data centers, whether they see it as capital speculation or a genuine demand for AI. We conducted an in-depth dialogue on this series of questions.

Why is the power shortage in the United States so severe?

Dongcha: First, let me ask a basic question in a broad direction, do you think electricity prices will continue to rise in the future?

Bit Deer: I think it will, because this is a very important supply and demand relationship for the future.

Dongcha: Regarding the power shortage in the United States, there is a saying in the market that it is difficult to obtain an “electricity license” in the U.S.?

Bit Deer: It's not that this so-called “power permit” cannot be approved, but rather that the physical speed of expanding the power grid cannot keep up. In the years following the relocation of heavy industry from the United States, the construction of the U.S. power grid has not systematically expanded. After mining companies moved to the U.S. in 2021, a lot of “already connected to the grid, already signed PPA” electricity was locked by the mining companies. With the influence of ChatGPT, pure AI players came in and discovered a large amount of electricity that could be immediately used in the mining sites.

This explains why large companies are willing to cooperate with mining enterprises. Instead of waiting 2-4 years to build 500MW from scratch, it's better to improve the existing park in 12 months.

Insight: When did the industry truly realize that “reasoning is also energy-intensive”?

Bit Deer: Probably after the widespread adoption of GPT-4. As companies integrate the model into customer service, office, search, risk control, and other areas, the demand for reasoning becomes long-term and scenario-based, and electricity consumption has not decreased as initially envisioned.

This brings about two types of changes.

One is the engineering upgrade: from stronger air cooling to liquid cooling / hybrid cooling, cabinet power, distribution paths, fire protection, and monitoring have all been elevated to the level of AI data centers.

The other is the resource strategy: electricity has become the real number one bottleneck. People are no longer just talking about “buying cards”, but instead focusing on securing electricity and grid connection, long-term contracts (PPA), grid connection scheduling, cross-regional capacity backup, and when necessary, obtaining electricity upstream like mining companies (self-generation/direct procurement).

In fact, we have already seen the same trend in the mining industry; chips can be infinitely expanded (silicon comes from sand), but electricity cannot be expanded. We have done natural gas self-generation to supply power to the mining site in Canada, and that is the logic behind it. Today's AI is almost identical.

How does the electricity consumption scale of AI data centers differ from that of traditional internet data centers?

Bit Deer: It's not about quantity, but about magnitude. In the past, 20-30 MW was already considerable for traditional internet data centers, but today, AI data centers can demand 500 MW or even 1 GW. AI has transformed data centers from a “rack business” into an “electrical engineering” endeavor, requiring a complete re-measurement: lines, substations, cooling, fire protection, redundancy, PUE… The experience from traditional internet data centers is still useful, but no longer sufficient.

Insight: Why has “electricity” become the most scarce element in the upstream?

Bit Deer: Chips can be scaled because they come from silicon and capacity management; electricity is hard to scale because it comes from power generation and grid upgrades. In the past, mining has tried “to look upstream for energy,” including self-generating projects in Canada; the path of AI is similar to this—whoever can secure electricity first will get the deployment time first.

AI's New Battlefield: From “Grabbing GPUs” to “Grabbing Power Grids”

Insight: Mining companies are transitioning to AI data centers, but what exactly needs to change? Previously, it was said that “Bitcoin mining power could be used for AI,” but mining chips (ASIC) are not compatible with the GPUs required for AI. So why can mining companies now “provide AI computing power”?

Bit Deer: Global mining was once divided into two parts. Bitcoin relies on mining chips ASIC, which are highly efficient but have a single purpose; Ethereum relied on NVIDIA GPUs, which are versatile but have exited the mining stage after transitioning to PoS.

So, the so-called “mining farms turning to AI” in today's market almost all refer to Bitcoin mining farms transitioning. The core point is that the mining farms are no longer “calculating hashes,” but are upgrading themselves into AI data centers.

This is an infrastructure upgrade, replacing ASIC racks with GPU servers; upgrading the power system from a “just enough” setup to a professional-grade power supply and distribution system with N+1/2N redundancy; upgrading traditional air cooling to a cooling system capable of supporting high-density GPUs; and standardizing and auditing the data center's sealing, dust-proofing, and fire protection facilities.

By completing these four steps, the encrypted mining farm will transform from a “mining workshop” into an “AI data center.”

Why can mining companies build faster than AI giants? Power.

AI is a business of “electricity and heat,” and the time frame for building AI data centers is 3-4 years, with time cost being the biggest barrier. Mining companies happen to hold these “hard assets,” so their starting line for transformation is closer.

Dongcha: In recent days, Microsoft and Amazon have successively signed long-term AI contracts with cryptocurrency mining companies. Iris Energy (IREN) signed a contract with Microsoft worth a total of 9.7 billion over 5 years; another company, Cipher, signed with Amazon Web Services for 5.5 billion over 15 years. This is seen as one of the first cases of collaboration between mining farms and major companies. What is your view on this?

Bit Deer: Iris Energy is a forward-looking Australian company that has been mining in the United States for a long time.

Iris Energy's shift towards AI is like a flare, as Bitcoin prices are high and its peers are still expanding mining, it diverts some of its power to invest in its own AI data center. Consequently, AI companies are coming to them proactively.

The real trigger point comes from the substantial investments of hyperscalers—such as Microsoft's commitment of about $9.7 billion—allowing the market to clearly see for the first time that the relationship between mining companies and hyperscalers is not just about “technical integration,” but rather about “the exchange of electricity and time.”

The popularity of AI has amplified the demand for infrastructure, thus opening up cooperation opportunities.

Dongcha: Why are leading mining companies more likely to be chosen by American AI giants at this stage?

Bit Deer: Because of “available electricity + engineering delivery speed.” The site selection and grid connection of mining enterprises in the previous cycle have now become the pre-capital for AI data centers. Time is the greatest discount factor, which directly determines who can go live during the window period, acquire customers, and generate rolling cash flow.

Dongcha: So, is it difficult to select land for AI data centers?

Bitzlato: Overall not large. In the United States and most countries, what is truly scarce is electricity, not land.

The reason is simple: places that can receive large amounts of electricity are mostly energy-rich areas (natural gas fields, coal mining regions, near hydropower stations, etc.), which are sparsely populated and have low land prices.

For example, Bitdeer's large data centers in Norway and Bhutan are located away from population centers, where power resources are concentrated and land costs are low. The same goes for the United States; such parks are not located in the urban core but rather in more peripheral areas, which are easier to find and cheaper in price. The “first principle” of site selection is electricity and grid connection, with land typically following electricity and not being the main bottleneck.

Dynamic Observation: AI is now being referred to as an upstream business like “steel, electricity, and land,” even resembling another form of real estate. What is your perspective?

Bit Deer: After the release of large models, the power consumption of AI far exceeds most people's expectations.

Initially, everyone thought that “training consumes a lot of energy, and inference would be light,” but the reality is the opposite. After inference becomes mainstream, it also continues to consume a lot of energy for a long time. As ChatGPT and DeepSeek become part of daily life, the number of terminal connections increases, and the baseline noise of inference continues to rise.

From an engineering perspective, AI is essentially a resource-consuming industry:

Chip side: During training, the acceleration card runs at basically 100% load, which naturally leads to high power consumption;

Data center side: The heat density is much higher than traditional servers, PUE is significantly elevated, and cooling itself also consumes a large amount of electricity;

Scale side: The electricity demand of AI data centers has jumped from the 20–30MW of traditional internet data centers to levels of 500MW, or even 1GW, which was almost unimaginable during the era of traditional internet data centers.

So comparing it to “real estate” is only half correct; it indeed requires land, factories, and long cycles (the construction cycle often takes 3–4 years), but what determines life and death is electricity and heat, whether large capacity can be connected to the grid on time, and making N+1/2N redundancy and efficient heat dissipation. In this regard, it is very similar to the strong dependence on steel, electricity, and land.

What are the characteristics of AI data centers?

Dongcha: What are the characteristics of the data center construction model in the United States?

Bit Deer: Due to power constraints and historical paths in the U.S., hyperscalers often need to engage directly and collaborate with mining companies to obtain available electricity.

Dongcha: Is it possible for foreign companies to operate AI data centers in the United States?

Bit Deer: Simply put, AI data centers are a strong regional business. The large-scale deployment, often exceeding hundreds of megawatts and thousands of kilowatts, is still dominated by major companies in the United States. We only discuss AI data centers and do not involve traditional internet data centers.

Insight: Will AI Data Centers evolve into tools of geopolitical influence? Will this affect your decision-making?

BitDeer: I agree with this judgment.

The foundation of AI is data, and data is inherently subject to sovereignty and security constraints. To prevent data leakage and security risks, various regions are tightening related policies: even if the United States allows foreign investment to establish data centers, as the amount of data controlled by AI increases, countries are likely to move towards “local deployment, local compliance, and data not leaving the country.”

In simple terms, AI in the United States is in the United States, the Middle East is in the Middle East, and Europe is in Europe; regionalization will be a long-term trend.

Industry Landscape and Potential

Dongcha: Besides IREN and Bitdeer, which mining companies have more potential to transition to AI data centers?

Bit Deer: To see who has a chance, first check if they have large electricity supply, and then see if they can quickly transform the mining site into a GPU data center. The type that has grid connection + land + substation, and can also achieve N+1/2N redundancy, liquid cooling / high density, is the most likely to get AI orders.

Another pure custody / light asset model, where you don't control the electricity and the park, and transitioning to an AI data center becomes passive.

In the United States, resources like Riot, CleanSpark, Core Scientific, TeraWulf, and Cipher that are in their own hands and have reliable expansion are more likely to be targeted by large companies.

So the conclusion is straightforward: electricity is the ticket, and transformation power is the speed; only when both are in place can you run ahead.

Overall, the key lies in who controls “high-quality, sustainable, large-load available electricity.” For example, companies with more self-owned grid resources have greater potential; those that primarily rely on hosting and lack their own energy and parks are at a disadvantage in this round of structural transformation.

What is Bitdeer thinking?

Dongcha: What are Bit Deer’s strategies and paths in “Mining Transition to AI”?

Bitdeer: Wu Jihan's approach has always been to create a complete industrial chain. Bitdeer holds approximately 3GW of power and park resources, which is our greatest underlying advantage.

When we first entered AI, we did not anticipate that “electricity” would become a core bottleneck, so initially we went for self-built and self-operated: we established a partnership with NVIDIA, became an NVIDIA PCSP, deployed a small-scale H100 cluster in Singapore, launched our own AI Cloud, and undertook training services externally. This project has been successfully implemented.

Subsequently, we also laid out a second data center in Malaysia. As Hyperscalers enter this field and begin to collaborate with mining enterprises, we are simultaneously advancing the upgrade of our high-load parks to AI data centers: we have announced the overall transformation of an approximately 180MW site in Norway into an AI DC, and the conversion of an approximately 13MW site in Washington State, USA.

Ultimately, the essence of AI is very similar to Crypto mining — both are businesses of “power + infrastructure”; we have the full chain capability from electricity, parks to computing power operations, so the transition to AI is relatively smooth.

Moving Insight: What are the core differences between Bitdeer and other mining companies like IREN?

Bit Deer: Three points. First, it will not be 100% converted into an AI enterprise; based on calculations, the current profits from Crypto Mining still outperform those from AI data centers, and mining has stable cash flow and better returns.

Our second advantage is the international engineering organization capability. The engineering organization and execution capability of the Bit Deer team is unparalleled in the world. For the same AI data center, the typical pace in the United States takes two years, but we can usually achieve it in a year and a half. This is achieved through parallel advancement and supply chain collaboration, synchronizing key aspects such as civil engineering, electromechanical, power distribution, and heat dissipation, compressing the conventional cycle of about 24 months to approximately 18 months, thus generating usable capacity faster.

The third company's strategy remains steady: the AI industry is very young, even younger than Crypto, and does not go “all-in”, pursuing a longer-term development pace.

Dongcha: Where is the current distribution of Bitcoin power infrastructure mainly located?

Bitdeer: Bitdeer is now mainly laying out approximately 3 GW of power and related infrastructure globally, covering five countries: the United States, Canada, Norway, Ethiopia, and Bhutan, to support the construction and operation of mining and AI data centers.

Cost and Financing

Dongcha: I saw in the Goldman Sachs report that an AI data center might cost 12 billion USD. Is it really that expensive?

Bit Deer: Indeed large, on the scale of “tens of times.” Here’s a more intuitive comparison in “human terms”: A Bitcoin mining farm (in the U.S.): Building 1 MW costs about 350,000 to 400,000 USD. However, building 1 MW for an AI data center costs about 11 million USD. This is because the investment in an AI data center is a composite of “heavy machinery + heavy standards”; coupled with grid connection queuing, environmental assessment/energy assessment, and regional compliance, the cycle usually takes 18 to 36 months.

You will find that the essence of an AI data center is not about “buying a few more cards,” but rather about connecting a piece of land into a “city of electricity” capable of consuming 500MW–1GW, properly connecting the electricity, dissipating the heat, ensuring sufficient redundancy, and overcoming compliance challenges, all of which are very costly.

Movements: Where does the money come from? Is financing needed?

Bit Deer's: To be honest, everyone needs financing.

Share a few common financing tricks in the industry:

1. Project financing / Infrastructure loan: Use the park + equipment as collateral, relying on long-term leases or power offtake (customers commit to buying your computing power for many years) to reassure the bank.

2. Equipment leasing / sale-leaseback: Lease GPUs and some electromechanical equipment to spread out the costs over a longer period, avoiding the need to pay such a large amount of cash all at once.

3. Long-term PPA: First lock in the electricity price and available capacity, and only then will the debt side be willing to offer low interest rates.

4. Binding with large enterprises: Major clients / large enterprises provide minimum consumption, prepayment, guarantees, or even joint ventures (JV), allowing you to access cheaper funding.

In the collaboration between IREN, CoreWeave, and Google/Microsoft, these terms can be seen.

Dongcha: Will Bit Deer also seek financing? Will it announce its partnership with major companies soon?

Bit Deer: This cannot be publicly discussed in detail at the moment.

Conclusion

Not long after the interview ended, BitDeer presented its next answer on the capital market.

On November 13, Bitdeer announced that it would raise $400 million by issuing convertible preferred notes, granting initial purchasers the option to add up to $60 million in notes within 13 days, making the total fundraising amount potentially reach $460 million. The new funds will be used for the expansion of data centers, R&D of ASIC miners, development of AI and HPC cloud services, as well as general corporate purposes.

As electricity has become the most scarce upstream resource in the AI industry, where this $460 million will ultimately be invested and how many megawatts of new load will be connected will largely determine Bitdeer’s position in the next round of computing power competition.

For Bitdeer, this amount of money is more like writing the judgment from the interview into the balance sheet: one end connected to the cash flow base of the mining sector, and the other end linked to the business line of AI data centers, which has a long slope and thick snow. It may not immediately reflect in the next quarter's financial report in terms of revenue and profit, but will gradually rewrite the power structure of the computing power business in the coming years—who qualifies to sit at the negotiating table and who can only wait in line on the grid connection list.

Looking back from the results, the story of this round of AI infrastructure is not complicated: electricity has become the real upstream, time has become the new currency, and the parks and grid connection indicators in the hands of mining companies have turned into “old assets” that others cannot buy even with money.

As the hype surrounding models and applications gradually subsides, the market is likely to revisit the ledger: no longer is it important whose narrative is the loudest; only those companies that can connect every megawatt of power and operate steadily in a world of electricity shortages will qualify to remain at the table in the next stage.