SAN FRANCISCO: A New Era of AI Resource Sharing
Leading artificial intelligence companies are investing substantial amounts—amounting to hundreds of billions of dollars—into building data centers crucial for AI development. Major technology giants like Amazon and Google, along with financially strong startups such as Anthropic and OpenAI, form an exclusive group with the necessary funds and connections to access this computing power. This situation effectively sidelines many other entities.
Anjney Midha, a tech entrepreneur with past ties to venture capital firm Andreessen Horowitz, aims to change this dynamic through his new startup, Amp. Based in Menlo Park, California, Amp is working to acquire excess computing capacity from data center operators across the U.S. and internationally, redistributing these resources to those who need them. Amp is creating a global pool of specialized microchips accessible to startups, educational institutions, and organizations that typically lack the computational resources to train advanced AI models.
Midha highlighted the challenge faced by many companies unable to secure the computing capabilities they need, as the wealthiest and most powerful companies monopolize the infrastructure for their exclusive use. Amp has successfully raised over $1.3 billion from a variety of investors, including Andreessen Horowitz, startup incubator Y Combinator, and several cloud computing firms.
Several promising startups have committed to utilizing and sharing Amp’s computing resources. These include Periodic Labs, focusing on scientific exploration, and Eleven Labs, specializing in AI systems for voice synthesis.
Amp’s initiative is part of a broader effort to consolidate AI infrastructure. Chipmaker Nvidia and French startup Mistral have recently announced a collaboration to pool computing resources for developing AI systems tailored to European companies, aiming to reduce dependence on U.S. tech giants.
Midha likened his startup to the creation of an electrical grid, where electricity can be distributed among homes and businesses. Similarly, a shared pool of computing capabilities can be accessed by startups and academic institutions. Investors provide capital that Amp uses to procure computing power from data center operators. In return, AI startups join the coalition to leverage this computing strength for model development, contributing additional funds or resources in exchange.
Some startups might offer crucial digital data needed for AI model training, while others could reciprocate by sharing their software code or collaborating on AI model training. The primary benefit of such collaboration is collective bargaining, noted Liam Fedus, CEO of Periodic Labs. Individually, his startup would struggle to secure the necessary computing power. However, if Amp negotiates with data center operators on behalf of multiple startups, it can exert greater influence.
“Pooling our demand allows us to engage in significantly more serious discussions about acquiring computing resources,” Fedus stated.
