Summary of Key Points
This news article highlights a novel development: the growing demand for AI computing power has outstripped the capacity of land-based data centers. A company called Panthalassa intends to establish AI data centers in the open ocean, utilizing ocean waves for power generation, seawater for cooling, and the vast marine environment as a physical infrastructure. As a result of this strategy, the company’s valuation has doubled in just three months (from 1 billion to 2 billion dollars). The article also compares Panthalassa’s approach with previous experiments by Microsoft, China’s efforts in this field, and the challenges faced by the expansion of AI computing power on land. It mentions SpaceX’s competition in sending computing resources into space and notes that GPU giants like NVIDIA are actively seeking new locations for their data centers.
Detailed Analysis
1. Land is Running Out of Space for AI Data Centers: The Ocean as a New Solution
AI models, such as GPT-4 and Wenxin Yiyuan, require thousands of GPUs for training, which collectively form data centers. However, land-based data centers face several significant challenges:
- Limited Space: Land around cities is becoming increasingly expensive, making it difficult to acquire large areas for data center construction.
- High Cooling Costs: GPUs generate a lot of heat during operation, and traditional data centers spend a substantial portion of their energy on cooling.
- Power Shortages: AI data centers consume enormous amounts of electricity; a large center’s annual power consumption is equivalent to that of hundreds of thousands of households.
The ocean offers a solution to these issues:
- Abundant Space: Approximately 70% of the Earth’s surface is covered by water, providing ample space for data centers.
- Natural Cooling: The cold seawater can be used to cool servers without additional energy costs for air conditioning.
- Renewable Energy: Ocean waves and offshore wind power are readily available, reducing reliance on land-based power grids.
This is why capital has suddenly turned its attention to ocean-based data centers as a breakthrough solution for the expansion of AI computing power.
2. Panthalassa: More Than Just a Start-up
Panthalassa is not a newcomer in this field; it has been working on wave power generation for nearly a decade and possesses unique expertise:
- Durable Equipment: The company knows how to design equipment that can withstand extreme marine conditions (corrosion resistance, impact resistance from waves).
- Power Conversion: It has developed technologies to convert the energy of ocean waves into electricity for data center use.
- Maintenance: Panthalassa understands how to deploy and remotely manage equipment in remote oceans, eliminating the need for frequent on-site repairs.
In 2024, Panthalassa tested a prototype near Washington State, demonstrating that wave power generation, computing, and satellite communication can work together effectively. The company aims to integrate these technologies to create floating data centers that use ocean waves for power and seawater for cooling, performing AI tasks (such as image recognition and speech translation) before transmitting results back to land via satellites.
Investors are optimistic about Panthalassa because of its solid foundation in marine engineering.
3. Previous Attempts: Microsoft’s海底 Data Centers and China’s Industrialization Efforts
The idea of deploying data centers in the ocean is not new:
- Microsoft’s Natick Project: Launched in 2015, this project involved placing servers underwater. The results showed that the failure rate was one-eighth of that of land-based data centers, and energy consumption for cooling was significantly reduced. However, Microsoft discontinued the project in 2024, possibly due to high maintenance costs or insufficient commercial returns.
- China’s Offshore Data Centers: China launched offshore data centers in 2026, utilizing offshore wind power and seawater for cooling. The goal is to achieve a PUE (Power Usage Efficiency) of less than 1.15 (traditional data centers typically have PUEs above 1.5), with 95% of energy coming from renewable sources. This approach addresses the lack of land in eastern coastal areas and the issue of transmitting electricity over long distances.
Panthalassa differs from these previous efforts by focusing on marine energy technologies, potentially providing more effective solutions due to its deeper understanding of oceanic conditions.
4. Is the Limit for Land-Based Data Centers Really Reached? NVIDIA’s Quest for New Locations
The expansion of AI data centers is encountering practical barriers:
- Regulatory Restrictions in the United States: As of August 2024, over 500 new data center projects have been suspended or restricted due to concerns about power consumption and environmental impacts. In July alone, another 150 projects were added to this list.
- Scarcity of Land and Power: NVIDIA recently invested $3 billion in Lancium, a company that owns land suitable for large-scale data centers and existing grid connections. This investment reflects NVIDIA’s concern about future locations for its GPU production.
These developments indicate that the availability of land for data centers is declining, forcing the search for new alternatives.
5. The Future of AI Computing Power: Space or the Ocean?
There are two main approaches being explored:
- Space Route: SpaceX and NVIDIA are collaborating to place GPUs on satellites (the Starmind AI1 satellite), utilizing solar energy for power and performing computations in space. The advantage is unlimited space and free solar energy, but the challenges include slow data transmission and difficult maintenance.
- Ocean Route: Panthalassa’s approach has been experimentally verified by Microsoft and China; it offers shorter distances to land and relatively easier maintenance. However, the ocean environment (storms, corrosion) poses additional risks and higher costs.
Regardless of the chosen path, NVIDIA is involved in these efforts because its GPUs need a place to be deployed. In the future, AI computing power may spread both into space and the ocean, moving beyond its traditional land-based limitations.
Conclusion
The explosive growth of AI computing power has outpaced the capacity of land-based data centers, leading to new competition in the oceans and space. Panthalassa’s approach represents a significant player in this race for new infrastructure solutions, reflecting the urgent need for innovation in AI infrastructure driven by capital demands.