In the past, cloud computing behemoths were the dominant force in negotiations concerning the leasing of AI data centers. Yet, the landscape is evolving. As cloud service providers such as Microsoft are in urgent demand for NVIDIA server racks to be deployed, data center operators like CoreWeave are witnessing a surge in their bargaining power. The stipulations within service level agreements have become more lenient, and payment terms are now skewed in favor of these operators. To facilitate the seamless construction of their chip-centric data centers, chipmakers including NVIDIA and AMD have taken the initiative to offer credit guarantees, thereby bolstering the operators' clout in negotiations even further.

Electricity expenses in data centers exhibit substantial volatility, with price disparities soaring up to 400% within a single month. The strain stemming from electricity shortages is intensifying, exacerbated by soaring construction costs that have doubled, rendering cost projections a formidable challenge. Moreover, in the face of a constrained supply of large-scale data centers, NVIDIA and other industry players have put forth strategies for compute fragmentation and decentralized AI inference, hinting at the potential rise of distributed networks as the prevailing trend.