Author | Lin Wanwan
Editor | Sleepy.txt
Initially, no one expected that the real bottleneck for AI would not be capital or large models, but electricity.
With long-term full-load training and AI inference running 24/7, a problem arises: 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, with new large-load grid connections taking 2-4 years, making "readily available electricity" a scarce resource across the industry.
Generative AI has brought a raw and harsh reality to the forefront: what is lacking is not the models, but the electricity.
The story has thus taken a turn, with crypto mining companies—those who were the first to treat electricity as "production materials"—beginning to move from the margins to the center stage of capital.
Iris Energy (IREN) is a case in point on this route. This year, IREN's stock price surged nearly 600% at one point, with a 52-week range from $5.12 to $75.73. As Bitcoin's price remained attractive, it decisively redirected its electricity to transform its self-built AI data center.
When giants like Microsoft made long-term orders worth a total of $9.7 billion, the market first intuitively understood the realistic path of "from mining to AI": first comes electricity and land, then GPUs and customers.
However, not all mining companies, like IREN, choose to bet everything on AI. In this power-driven migration of computing power, there is another steady force worth our attention—Bitdeer.
Bitdeer Technologies Group (NASDAQ: BTDR), founded by crypto legend Wu Jihan and headquartered in Singapore, holds nearly 3GW of global electricity resources and has avoided the shallow trap of relying on others for "power" from the start. As the wave of AI arrives, Bitdeer has not chosen the aggressive "All-in" approach like IREN, but has retained its profitable Bitcoin mining as a "foundation" while steadily upgrading some of its mining sites to AI data centers.
This "offensive and defensive" strategy makes it the best sample to observe how global players think and layout in this computing power competition.
To this end, we interviewed Wang Wenguang, Vice President of Global Data Center Business at Bitdeer, hoping to gain insights into the current global electricity shortage for AI, and how they view the transition of mining companies to AI data centers—whether it is capital speculation or a genuine demand for AI. We engaged in an in-depth dialogue on this series of questions.
Why is the electricity shortage in the U.S. so severe?
Interviewer: Let's start with a fundamental question. Do you think electricity prices will continue to rise in the future?
Bitdeer: I think they will, because this is a very important supply-demand relationship for the future.
Interviewer: Regarding the electricity shortage in the U.S., there is a saying in the market that it is difficult to obtain "electricity permits" in the U.S.?
Bitdeer: It's not that the so-called "electricity permits" cannot be approved, but that the physical speed of expanding the power grid cannot keep up. For many years after heavy industry moved abroad, the U.S. power grid construction did not expand systematically. After mining companies moved to the U.S. in 2021, much of the "already connected and signed PPA" electricity was locked in by mining companies. With the influence of ChatGPT, pure AI players came in and found a large amount of immediately usable electricity in the mining sites.
This explains why large companies are willing to collaborate with mining companies; rather than waiting 2-4 years to bring 500MW online from scratch, it is better to transform existing facilities in 12 months.
Interviewer: When did the industry truly realize that "inference also consumes a lot of electricity"?
Bitdeer: Probably after the widespread adoption of GPT-4. As companies embedded models into customer service, office work, search, risk control, etc., the demand for inference became long-term and scenario-based, and electricity consumption did not decrease as initially envisioned.
This has led to two types of changes.
One is engineering upgrades: from stronger air cooling to liquid cooling/hybrid cooling, with cabinet power, distribution paths, fire protection, and monitoring all elevated to the level of AI data centers.
The other is resource strategy: electricity has become the real number one bottleneck. People are no longer just talking about "buying cards," but are focusing on securing electricity and grid connections, long-term PPA agreements, grid connection schedules, and cross-regional capacity backups. When necessary, they are looking to upstream sources for electricity (self-generation/direct procurement) like mining companies.
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 easily expanded. We have done natural gas self-generation in Canada to ensure power supply for mining sites, which follows this logic. Today's AI is almost identical.
Interviewer: How does the electricity consumption scale of AI data centers differ from traditional internet data centers?
Bitdeer: It's not just a quantitative change, but a change in magnitude. In the past, 20-30 MW was already considerable for traditional internet data centers, but now AI data centers often demand 500MW or even 1GW. AI has transformed data centers from a "rack business" into "power engineering," requiring everything to be recalibrated: lines, substations, cooling, fire protection, redundancy, PUE… The experience from traditional internet data centers is still useful, but it is no longer sufficient.
Interviewer: Why has "electricity" become the most scarce upstream element?
Bitdeer: Chips can be expanded because they come from silicon and capacity management; electricity is hard to expand because it comes from power generation and grid upgrades. The mining industry has already tried "seeking energy upstream," including self-generation projects in Canada; the path for AI is similar—whoever secures electricity first will have the deployment time advantage.
The New Battlefield of AI: From "Competing for GPUs" to "Competing for Power Grids"
Interviewer: What specific changes do mining companies need to make to transition to AI data centers? Previously, people said "Bitcoin computing power can be used for AI," but mining chips (ASICs) are not compatible with the GPUs needed for AI. So why can mining companies now "provide AI computing power"?
Bitdeer: The global mining industry was once divided; Bitcoin relied on mining chips (ASICs), which are efficient but have a single use; Ethereum relied on NVIDIA GPUs, which are versatile but have exited the mining stage after transitioning to PoS.
So, when people talk about "mining sites transitioning to AI," they are almost always referring to Bitcoin mining sites undergoing transformation. The key point is that mining sites are no longer "calculating hashes," but are upgrading themselves into AI data centers.
This is an infrastructure upgrade, removing ASIC racks and replacing them with GPU servers; upgrading the "just enough" power system to a professional-grade supply and distribution system with N+1/2N redundancy; upgrading traditional air cooling to a cooling system capable of handling high-density GPUs; and standardizing and auditing the facilities for sealing, dust-proofing, and fire protection.
Completing these four steps transforms a crypto mining site from a "mining workshop" into an "AI data room."
Why can mining companies transition faster than AI companies building their own? Electricity.
AI is a business of "electricity and heat," and the construction cycle for AI data rooms is 3-4 years, making time cost the biggest barrier. Mining companies happen to hold these "hard assets," thus starting their transformation from a more advanced position.
Interviewer: Recently, Microsoft and Amazon signed long-term AI contracts with crypto mining companies. Iris Energy (IREN) signed a contract with Microsoft worth $9.7 billion over 5 years; another company, Cipher, signed with Amazon Web Services for $5.5 billion over 15 years. These are seen as the first cases of collaboration between mining sites and large companies. What are your thoughts?
Bitdeer: Iris Energy is a forward-looking Australian company that has been mining in the U.S. for a long time.
Iris Energy's choice to pivot to AI serves as a signal; at a time when Bitcoin prices were high and peers were still expanding mining, it redirected some of its electricity to invest in self-built AI data centers. Consequently, AI companies began to approach them proactively.
The real trigger point comes from the substantial commitments of hyperscalers—such as Microsoft's $9.7 billion promise—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 "the exchange of electricity and time."
The heat of AI has amplified the demand for infrastructure, opening up collaboration opportunities.
Interviewer: Why are leading mining companies more likely to be chosen by U.S. AI giants at this stage?
Bitdeer: Because of "available electricity + engineering delivery speed." The site selection and grid connection from the previous cycle for mining companies have now become the upfront capital for AI data centers. Time is the biggest discount factor; it directly determines who can go live within the window period, acquire customers, and generate rolling cash flow.
Interviewer: Are there significant challenges in land selection for AI data centers?
Bitdeer: Overall, not really. In the U.S. and most countries, what is truly scarce is electricity, not land.
The reason is simple: places that can access 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 costs.
For example, Bitdeer's large data centers in Norway and Bhutan are located far from population centers, where electricity resources are concentrated and land costs are low. The same applies in the U.S.; such parks are not located in urban core areas but in more peripheral locations, making land easier to find and cheaper. The "first principle" of site selection is electricity and grid connection; land usually follows electricity and is not the main bottleneck.
Interviewer: AI is now being referred to as an upstream business of "steel, electricity, and land," even likened to another form of real estate. What are your thoughts?
Bitdeer: After the emergence of large models, the electricity consumption intensity of AI far exceeds most people's expectations.
Initially, people thought "training consumes electricity, but inference would be light," but the opposite is true; inference also has long-term high electricity consumption as it becomes more mainstream. As ChatGPT and DeepSeek enter daily use, 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, accelerator cards are running at nearly 100% load, naturally consuming high power;
- Data room side: Heat density is far higher than traditional servers, PUE is significantly elevated, and cooling itself also consumes a lot of electricity;
- Scale side: The electricity demand of AI data centers has jumped from the traditional internet data center's 20-30MW to 500MW or even 1GW, which was almost unimaginable in the era of traditional internet data centers.
So, comparing it to "real estate" is only half right; it indeed requires land, buildings, and long cycles (construction cycles often take 3-4 years), but what determines life and death is electricity and heat—whether large-capacity grid connections can be secured on time, and whether N+1/2N redundancy and efficient cooling can be achieved. In this regard, it bears a strong resemblance to steel, electricity, and land.
What are the characteristics of AI data centers?
Interviewer: What are the characteristics of the data center construction model in the U.S.?
Bitdeer: Due to electricity constraints and historical paths in the U.S., hyperscalers often need to personally engage and collaborate with mining companies to obtain usable electricity.
Interviewer: Is it possible for foreign companies to establish AI data centers in the U.S.?
Bitdeer: Simply put, AI data centers are a highly regional business. The truly large-scale deployments, often in the hundreds of megawatts or even gigawatts, are still predominantly led by local giants in the U.S. We are only discussing AI data centers, not traditional internet data centers.
Interviewer: Will AI data centers evolve into tools of geopolitical influence? Will this affect your decisions?
Bitdeer: I agree with that assessment.
The foundation of AI is data, which is inherently subject to sovereignty and security constraints. To prevent data leakage and security risks, various regions are tightening related policies: even if the U.S. allows foreign investment in data centers, as AI accumulates more data, countries will likely move towards "local deployment, local compliance, and data not leaving the country."
In simple terms, AI in the U.S. stays in the U.S., the Middle East in the Middle East, and Europe in Europe; regionalization will be a long-term trend.
Industry Landscape and Potential
Interviewer: Besides IREN and Bitdeer, which mining companies have the potential to transition to AI data centers?
Bitdeer: It depends on who has the resources. First, look for those with substantial electricity, and then see if they can quickly convert their mining sites into GPU data rooms. Those with grid connections, land, substations, and the ability to implement N+1/2N redundancy and liquid cooling/high density are the most likely to secure AI contracts.
On the other hand, companies that are purely hosting or asset-light, without control over electricity and facilities, will be at a disadvantage in transitioning to AI data centers.
In the U.S., companies like Riot, CleanSpark, Core Scientific, TeraWulf, and Cipher, which have resources under their control and reliable expansion capabilities, are more likely to attract attention from large firms.
So the conclusion is straightforward: electricity is the ticket, and the ability to transform is the speed; only when both are in place can you be at the forefront.
Overall, it is crucial to see who controls "high-quality, sustainable large-load available electricity." Companies with more self-owned grid resources have greater potential; those primarily relying on hosting, lacking self-owned energy and facilities, will not have an advantage in this structural transformation.
What is Bitdeer Thinking?
Interviewer: What is Bitdeer's strategy and path in transitioning from mining to AI?
Bitdeer: Wu Jihan's vision has always been to create a full industry chain. Bitdeer controls about 3GW of electricity and facility resources, which is our greatest foundational advantage.
Initially, when we entered AI, we did not anticipate that "electricity" would become a core bottleneck, so we started with self-built and self-operated models: we established a partnership with NVIDIA, becoming an NVIDIA PCSP, and deployed a small-scale H100 cluster in Singapore, launching our own AI Cloud and taking on external training business; this project has been successfully implemented.
Subsequently, we also set up a second data center in Malaysia. As hyperscalers entered this field and began collaborating with mining companies, we simultaneously advanced the upgrade of large-load facilities to AI data centers: we have announced the complete transformation of a site in Norway with approximately 180MW into an AI data center, and we are also converting a site in Washington State, U.S., with about 13MW.
Ultimately, the essence of AI is very similar to crypto mining—both are businesses of "electricity + infrastructure"; we possess the full chain capability from electricity, facilities to computing power operations, making the transition to AI relatively smooth.
Interviewer: What are the core differences between Bitdeer and other mining companies like IREN?
Bitdeer: Three points. First, we will not fully transition into an AI company; based on calculations, the current profits from crypto mining still exceed those from AI data centers, and mining provides stable cash flow and good returns.
Our second advantage is our international engineering organizational capability. Bitdeer's team has unparalleled engineering organization and execution capabilities worldwide. The same AI data center that typically takes two years to build in the U.S. can often be completed in a year and a half by us. This is achieved through parallel advancement and supply chain collaboration, synchronizing key aspects such as civil engineering, electromechanical, power distribution, and cooling, compressing the usual 24-month cycle to about 18 months, thus forming usable capacity more quickly.
The third point is that our company strategy remains prudent: the AI industry is very young, even younger than crypto, and we are not going "all-in," but rather pursuing a longer-term development pace.
Interviewer: Where is Bitdeer's electricity infrastructure primarily distributed?
Bitdeer: Bitdeer is currently laying out about 3GW of electricity and related infrastructure globally, covering five countries: the U.S., Canada, Norway, Ethiopia, and Bhutan, to support the construction and operation of mining and AI data centers.
Cost and Financing
Interviewer: I saw a Goldman Sachs report mentioning that an AI data center could cost $12 billion. Is it really that expensive?
Bitdeer: It is indeed large; the scale is "dozens of times" higher. To give you a more intuitive comparison: for a Bitcoin mining site (in the U.S.), building 1 MW costs about $350,000 to $400,000. However, building 1 MW for an AI data center costs about $11 million. This is because the investment in AI data centers is a complex of "heavy electromechanical + heavy standards," plus the queue for grid connection, environmental assessments, energy assessments, and regional compliance, which typically takes 18-36 months.
You will find that the essence of an AI data center is not just "buying a few more cards," but rather connecting a piece of land to become a "city of electricity" capable of handling 500MW to 1GW, ensuring proper electricity connections, managing heat dissipation, providing sufficient redundancy, and navigating compliance—all of which are very costly.
Interviewer: Where does the money come from? Is financing needed?
Bitdeer: To be honest, financing is necessary.
Let me share some common financing strategies in the industry:
Project financing/infrastructure loans: Using the facility and equipment as collateral, relying on long-term leases or computing power offtake (customers committing to buy your computing power for many years) to reassure banks.
Equipment leasing/leaseback: Leasing GPUs and some electromechanical equipment to spread out the long-term costs, avoiding the need to pay a large sum upfront.
Long-term PPA: Locking in electricity prices and available capacity first, which makes lenders willing to offer lower interest rates.
Binding with large firms: Large customers or firms providing minimum consumption guarantees, prepayments, guarantees, or even joint ventures (JVs) to secure cheaper funding.
These terms can be seen in the collaborations between IREN, CoreWeave, and Google/Microsoft.
Interviewer: Will Bitdeer also need financing? Will there be announcements about collaborations with large firms soon?
Bitdeer: We cannot disclose much about that at this time.
Conclusion
Not long after the interview ended, Bitdeer provided its next answer in the capital market.
On November 13, Bitdeer announced that it would raise $400 million through the issuance of convertible preferred notes, granting initial purchasers the option to purchase up to $60 million in additional notes within 13 days, with the total fundraising potentially reaching $460 million. The new funds will be used for data center expansion, ASIC mining machine research and development, AI and HPC cloud business expansion, and 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 money is more like writing the judgments from the interview into the balance sheet: one end connects to the cash flow foundation of mining, while the other connects to the long-term business line of AI data centers. It may not immediately reflect in the next quarterly report's revenue and profit, but over the next few years, it will gradually rewrite the power structure of the computing power business—who has the right to sit at the negotiation table, and who can only wait in line for electricity on the grid connection list.
Looking back from the results, this round of AI infrastructure stories is not complicated: electricity has become the true upstream, time has become the new currency, and the facilities and grid connection indicators in the hands of mining companies have turned into "old assets" that others cannot buy even with money.
As the noise around models and applications gradually recedes, the market will likely need to review the accounts again: whose narrative is no longer important, but rather which companies can connect every megawatt of electricity in a world of power shortages and run it steadily, will have the qualification to remain at the table in the next phase.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
