AI's Power Bottleneck: How Crypto Miners Pivot to Data Centers

The Energy Crunch Stifling AI's Boom

Initially, no one anticipated that power, not capital or massive AI models, would pose the biggest challenge to AI's growth. Constant AI model training and round-the-clock inference demand significant energy, straining the electricity infrastructure. In the US, in particular, connecting new electrical loads to the grid can take two to four years due to aging infrastructure, making readily available power a precious commodity.

The Pivot to Crypto Miners

This situation has led to an unexpected shift: crypto mining firms, previously seen as peripheral to the industry, are now becoming key players in the AI space. They've recognized the potential of AI data centers, converting their mining operations to cater to AI's computing needs. Iris Energy (IREN) is a prime example, experiencing a significant stock surge this year after shifting some of its mining operations into AI data centers. The rising demand for these centers, coupled with long-term contracts with companies like Microsoft, has highlighted the value of existing power and land infrastructure in this space.

Bitdeer's Cautious Approach

In contrast to IREN, Bitdeer Technologies Group, led by crypto legend Wu Jihan, has taken a more cautious approach. With nearly 3 GW of power resources globally, Bitdeer is maintaining its profitable Bitcoin mining operations while strategically upgrading some of its facilities to AI data centers. This strategy allows them to balance risks and opportunities, making them interesting observers in the race to meet AI's growing power needs.

Interview with Bitdeer's VP

To gain further insights into this evolving dynamic, we interviewed Wang Wenguang, VP of Global Data Center Business at Bitdeer. We discussed the challenges of AI's power scarcity, the transition from crypto mining to AI data centers, and whether the current hype is driven by speculation or genuine demand.

US Power Shortage

Wang Wenguang clarified that the power shortage in the US isn't due to difficulty in obtaining permits but rather to the aging grid infrastructure struggling to keep up with demand. While crypto miners had already secured a significant amount of available power, the emergence of AI has exacerbated the issue, forcing tech giants to collaborate with crypto miners for access to needed power.

Inference Consumes Significant Power

Wang Wenguang also stressed that inference, contrary to popular belief, also consumes significant power, especially as AI models are integrated into various applications. This has led to upgrades in infrastructure, such as advanced cooling systems, and a shift in resource allocation strategies, with power becoming a critical factor.

AI Data Centers vs. Traditional Internet Data Centers

Wang Wenguang explained that AI data centers demand significantly more power compared to traditional internet data centers. They are transforming data centers into intricate electrical projects requiring careful attention to infrastructure and redundancy.

The Race for Power

Wang Wenguang emphasized that power has become the most critical upstream element for AI infrastructure. Just as crypto miners sought out power sources, AI companies must now secure their power supplies to ensure timely deployment.

Transforming Crypto Mines into AI Data Centers

Transforming a crypto mine into an AI data center involves significant infrastructure upgrades, including replacing ASICs with GPUs, enhancing power and cooling systems, and implementing enhanced security measures. Crypto miners' existing power assets allow them to complete these conversions faster than building new data centers from scratch.

Partnerships with Tech Giants

The recent long-term contracts between crypto miners and companies like Microsoft and Amazon are a testament to the value they place on miners' power infrastructure. These partnerships allow tech giants to access needed power quickly while providing miners with stable revenue streams.

Location Considerations

Wang Wenguang explained that the location of AI data centers is primarily dictated by the availability of affordable power. These locations are often in energy-rich, sparsely populated areas, making land acquisition less challenging.

AI as a Resource-Intensive Industry

Wang Wenguang acknowledged that AI is a resource-intensive industry demanding significant amounts of power, chips, and space. The high-power nature of inference, coupled with the large power requirements of AI data centers, makes power a critical factor in determining success in this space.

US AI Data Centers

US AI data centers are characterized by their focus on collaborating with crypto miners for power and innovative, infrastructure-driven solutions. He emphasized that these projects are currently dominated by large domestic players.

Geopolitical Considerations

Wang Wenguang acknowledged that AI data centers can become geopolitical tools due to the sensitive nature of the data they handle. This leads to stricter policies on data storage and residency, impacting Bitdeer's strategies.

Potential AI Pivot Players

Besides IREN and Bitdeer, Wang Wenguang identified other mining companies with the potential to enter the AI space, provided they possess significant power and can quickly convert their operations. These companies include Riot, CleanSpark, Core Scientific, TeraWulf, and Cipher.

Bitdeer's AI Strategy

Bitdeer's AI strategy involves leveraging its existing power resources and infrastructure. They've already established a small-scale AI data center in Singapore and are expanding in Malaysia. They also plan to convert existing Bitcoin mining facilities into AI data centers.

Bitdeer's Competitive Advantages

Bitdeer differentiates itself from competitors by not fully converting all its operations to AI, maintaining profitable Bitcoin mining operations, and possessing strong international engineering capabilities. They aim for rapid execution of infrastructure upgrades and maintain a cautious approach in the rapidly evolving AI industry.

Power Infrastructure Distribution

Bitdeer currently has approximately 3 GW of power infrastructure globally, located in the US, Canada, Norway, Ethiopia, and Bhutan. This infrastructure supports both crypto mining and AI data center operations.

Funding Costs

Wang Wenguang acknowledged that building AI data centers is significantly more expensive, exceeding the costs of Bitcoin mining facilities by many times. To secure funding, Bitdeer is exploring various options, including project financing, long-term power purchase agreements, and collaborations with major tech companies.

Bitdeer's Future Path

Bitdeer plans to raise up to $460 million through the issuance of convertible notes to fund data center expansion, ASIC development, and expansion of its AI and HPC cloud business. This investment demonstrates their strategy to secure a leading position in AI computing by leveraging their power resources and infrastructure.

Conclusion

Power scarcity is proving to be the real bottleneck in the AI boom. Crypto mining firms, with their existing power assets, are well-positioned to meet the growing demand for AI data centers. Bitdeer's strategic approach, balancing crypto mining and the transition to AI, reflects the evolving dynamics in this landscape.