By Titik Puspa 
In a move that underscores the escalating arms race for artificial intelligence dominance, leading AI research lab Anthropic has finalized a staggering $45 billion agreement to lease approximately $45 billion worth of AI compute power from Nscale, a burgeoning British AI infrastructure company. This colossal deal, revealed by a source close to the transaction to TechCrunch, positions Anthropic to significantly amplify its computational capabilities, a critical factor in developing and deploying advanced AI models.
Nscale, a relatively new entrant to the AI infrastructure scene having been founded just in 2024, has rapidly demonstrated its ambition and capacity to forge strategic alliances with major technology players. The company has already announced significant partnerships, including a notable deal with Microsoft, signaling its intent to become a key provider of the essential hardware backbone for the AI revolution. This latest agreement with Anthropic is set to leverage Nvidia’s cutting-edge Vera Rubin chip architecture, a system described as the pinnacle of current chip design. The Vera Rubin system is a sophisticated integration of six distinct chips working in concert, representing a substantial leap forward in processing power and efficiency. This state-of-the-art compute capacity is anticipated to begin powering Anthropic’s services in late 2027, providing a robust foundation for its future AI endeavors.
The multi-year agreement, first brought to light by Bloomberg, is a clear indicator of Anthropic’s aggressive strategy to scale its operations and enhance its competitive edge against rivals like OpenAI. The compute power will be primarily sourced from Nscale’s flagship data center, strategically located in West Virginia. This facility is designed to meet the immense demands of large-scale AI training and inference, offering the high-bandwidth, low-latency connectivity crucial for such operations. The six-year duration of the deal suggests a long-term commitment from both parties, highlighting the sustained growth expected in the AI compute market.
This substantial investment by Anthropic is not an isolated event but rather the latest in a series of calculated moves over the past eight months to aggressively expand its compute capacity. The company has been relentlessly pursuing partnerships to ensure it has the necessary infrastructure to train more complex models, conduct extensive research, and deploy its AI solutions at scale. This proactive approach is vital in a landscape where compute availability is becoming a significant bottleneck, and access to powerful hardware is directly correlated with the ability to innovate and compete.
Earlier this month, Anthropic inked a $10 billion deal with Volta, a freshly established AI cloud startup founded only in January of this year. This agreement secures a six-year supply of cloud computing power from a data center situated in Norway. The choice of Norway is likely driven by factors such as access to renewable energy sources for powering energy-intensive data centers and potentially favorable regulatory environments. Following this, in July, Anthropic announced a $5 billion compute-related deal with AMD, another major semiconductor manufacturer, further diversifying its hardware suppliers and ensuring a broad base of computational resources.
The strategic partnerships extend even further back. In May, Anthropic revealed a significant computing deal with SpaceX, the aerospace company led by Elon Musk. This alliance, notably, places Anthropic in an unusual position of potential strategic alignment with Musk, whose public rivalry with OpenAI CEO Sam Altman has been a prominent feature of the AI landscape. The SpaceX deal draws computing capacity from two different SpaceX data centers and is reportedly providing Anthropic with an impressive $1.25 billion worth of capacity each month. This arrangement suggests a flexible and potentially rapid deployment of resources, crucial for meeting evolving project needs.
Anthropic’s expansion efforts also include deepening existing relationships. In April, the company announced a significant expansion of its partnership with Amazon, securing access to an additional 5 gigawatts of compute. This massive infusion of power from Amazon Web Services (AWS) is indicative of Anthropic’s reliance on hyperscale cloud providers for its core infrastructure needs. In the same month, Anthropic bolstered its computational arsenal by expanding its relationships with both Google and Broadcom, further diversifying its access to powerful computing resources, including Google’s Tensor Processing Units (TPUs).
The pursuit of AI horsepower is a pervasive trend across the entire industry, and Anthropic is far from alone in its ambitious drive. The race to secure as much compute capacity as possible is a defining characteristic of the current AI era. Major players such as Google, OpenAI, and Meta are all engaged in similar aggressive strategies to acquire or lease vast amounts of computing power. This collective surge in demand is not only driving innovation but also reshaping the landscape of the semiconductor and data center industries.
The demand for AI compute is intrinsically linked to the increasing complexity and scale of AI models. Large language models (LLMs), like those developed by Anthropic, require immense computational resources for training. This training process involves feeding massive datasets to the models and iteratively adjusting billions, or even trillions, of parameters. The more parameters a model has and the larger the dataset it is trained on, the more sophisticated its capabilities can become, but also the greater the compute demands. For instance, training a state-of-the-art LLM can require hundreds of thousands of GPU-years, a metric representing the computational power of thousands of GPUs running for a year. The cost of such training can run into tens or even hundreds of millions of dollars, making access to affordable and scalable compute a strategic imperative.
Furthermore, the deployment of AI models, known as inference, also consumes significant computational resources, especially as these models are integrated into consumer-facing products and services. As AI applications become more pervasive, from personalized recommendations to complex scientific simulations, the demand for inference compute will continue to rise exponentially. This creates a virtuous cycle where advancements in AI drive demand for better hardware, and improved hardware enables the development of more powerful AI.
Nvidia, the dominant player in the AI chip market, is a key beneficiary of this compute race. The company’s GPUs have become the de facto standard for AI workloads due to their parallel processing capabilities. The introduction of new architectures like Vera Rubin signifies Nvidia’s continuous innovation to meet the ever-growing demands of the AI industry. The Vera Rubin architecture, with its integrated design and advanced features, is expected to offer significant performance improvements over previous generations, making it highly attractive to companies like Anthropic. The ability to integrate multiple specialized chips within a single system can lead to greater efficiency, reduced power consumption, and faster data transfer, all critical factors for large-scale AI operations.
The strategic importance of compute cannot be overstated. It is the fundamental resource that fuels AI research and development. Companies that can secure more compute have a distinct advantage in training larger, more capable models, experimenting with new AI architectures, and deploying their products faster to market. This has led to a situation where compute has become a scarce and highly valuable commodity, akin to oil or rare earth minerals in previous economic eras. The intense competition for compute also has broader economic implications, influencing the growth of the semiconductor industry, the development of data center infrastructure, and the geopolitical landscape of AI leadership.
The deals Anthropic has struck are not merely about acquiring hardware; they represent a strategic vision for the future of AI. By diversifying its compute sources across hyperscalers like Amazon and Google, specialized AI infrastructure providers like Nscale and Volta, and even leveraging partnerships with companies like SpaceX, Anthropic is building a resilient and scalable compute foundation. This multi-pronged approach mitigates risks associated with relying on a single supplier and allows for flexibility in adapting to evolving technological needs and market conditions.
The significant capital outlay for compute power also reflects the immense potential that companies like Anthropic see in the AI market. The ability to develop and deploy advanced AI models has the potential to disrupt numerous industries, from healthcare and finance to transportation and entertainment. The substantial investments being made are a testament to the belief that AI will be a transformative force, and companies that can effectively harness its power will be at the forefront of future economic and technological advancements. The $45 billion deal with Nscale, in particular, signals a long-term commitment to pushing the boundaries of what is possible with AI, powered by the most advanced hardware available.
