Gokul, Global Head of Technology Hardware Research at JPMorgan Chase, estimates that building 1 GW of AI infrastructure, based on Nvidia's next-generation AI computing system VR200, would cost approximately $40-45 billion. Meanwhile, leading AI model companies are currently generating about $30 billion in revenue per GW of computing power (increasing from $10 billion to $30 billion in one year). If a cloud computing company leases 1 GW of computing power to an AI model developer, JPMorgan Chase estimates that the cloud provider could earn approximately $17 billion per GW.
