The AI Bottleneck: Tantalum and Cobalt

As the world marvels at the astonishing capabilities of generative AI, a competition about its physical foundation has already quietly begun. Beyond code and algorithms, the surging computing power of AI heavily relies on two essential rare metals: tantalum and cobalt. Although they rarely appear in the spotlight, they together form the cornerstone of AI infrastructure from hardware stability to data access.

High purity cobalt target material

High-purity cobalt tube

High purity cobalt plate

In AI server clusters, tens of thousands of GPUs are performing parallel computing at ultra-high frequencies. The current fluctuation in this computing state is extremely large, and the instantaneous power consumption is astonishing. Any small voltage fluctuation can lead to calculation errors or system crashes.
This is the battlefield where tantalum capacitors are irreplaceable:
High reliability core: Tantalum capacitors have unparalleled volume efficiency (CV value) and high stability. In a limited PCB space, it can provide charge storage capabilities far beyond other types of capacitors, ensuring that the core voltage of AI chips remains rock stable even under severe load fluctuations.
Guardian of data integrity: Every AI training involves instantaneous throughput of massive amounts of data. Tantalum capacitors provide a "pure" electrical environment for precise data transmission inside and outside the chip by quickly charging and discharging and filtering out noise in the circuit, which is the key to ensuring accurate calculation results.
Supply chain reality: Global tantalum resource supply is highly concentrated and lacks elasticity. Its main sources, such as the Democratic Republic of Congo, Rwanda, and other Central African regions, have long faced challenges in geopolitical and ethical procurement (such as "conflict minerals"). For AI hardware manufacturers, the stable supply of tantalum capacitors is directly equivalent to the certainty of AI computing power expansion.
Without the ultimate power integrity brought by tantalum capacitors, the high-frequency high performance of top AI acceleration cards such as NVIDIA H100 and AMD MI300 would be impossible to talk about. It is the "ultimate stabilizer" for AI computing power to move from theory to reality.

The role of cobalt in AI hardware is more diverse, running through two key links: data storage and power management.
The engine of next-generation storage technology:
Traditional AI data centers are facing the challenge of a "memory wall" - the speed of data transfer between processors and storage severely restricts overall efficiency.
Cobalt is a key material for manufacturing magnetic random access memory (MRAM). MRAM has disruptive features such as high speed (such as SRAM), high density (such as DRAM), and non-volatile power off (such as Flash). The MRAM embedded with cobalt can serve as a physical carrier for integrated storage and computing architecture, greatly reducing data transfer, and is specifically designed for AI computing modes.
Adhesive for 3D NAND Flash Memory:
The 3D NAND flash memory used for storing massive training data has exceeded 300 stacking layers. Cobalt, due to its excellent conductivity and filling ability, is used as a high-precision coating material between storage units at the atomic scale, ensuring the integrity and stability of signals between layers. It is the "loyal guardian" of data warehouses.
The energy cornerstone of edge AI:
Edge computing scenarios such as autonomous vehicle and AIoT devices are inseparable from lithium batteries with high energy density. Cobalt plays a core role in stabilizing the structure, improving energy density, and enhancing rate performance in lithium-ion battery cathode materials such as NCA/NMC. It is the 'blood' of mobile AI devices.
 

The importance of tantalum and cobalt has sounded the alarm for global AI industry participants:
For investors： they should pay attention to electronic component companies that have a stable layout in the tantalum supply chain and actively practice responsible procurement standards, as well as semiconductor companies that are committed to researching and developing cutting-edge storage technologies such as cobalt based MRAM.
For both buyers and manufacturers：a dual track strategy must be established. One is to strengthen control over upstream resources of tantalum and cobalt through long-term agreements, direct investment, and other means; The second is to invest heavily in recycling technology to efficiently recover these two metals from electronic waste and build "urban mines" to hedge the supply risk of primary minerals.
For technology developers：exploring alternative solutions to tantalum capacitors (such as polymer based high-performance capacitors) and low cobalt/cobalt free battery technology and storage materials is a key technological backup path for the long-term development of the AI industry.

Underneath the surface of the AI arms race lies a deep competition for fundamental elements such as tantalum and cobalt. The competition for computing power begins with code and ends with atoms. Understanding and controlling the supply chain of these 'invisible pillars' not only concerns costs, but also directly determines the speed and endurance in the era of intelligence.