So, let's say $200k/year per job. $3.2 billion a year in payroll, and spending $185 billion, or 60 years' worth of payroll on compute. Big expectations there.
Attribution is the missing layer in the AI infrastructure stack. Companies can track GPU utilization and token volumes, but they cannot yet connect compute spend to business output. Until that link exists, pricing and allocation are approximations. The capex return problem is fundamentally an attribution problem.
rationalized still means you're spending. the controls aren't there to cut it - they're there so you know what you're cutting before the invoice tells you.
So, let's say $200k/year per job. $3.2 billion a year in payroll, and spending $185 billion, or 60 years' worth of payroll on compute. Big expectations there.
To say the least.
Attribution is the missing layer in the AI infrastructure stack. Companies can track GPU utilization and token volumes, but they cannot yet connect compute spend to business output. Until that link exists, pricing and allocation are approximations. The capex return problem is fundamentally an attribution problem.
rationalized still means you're spending. the controls aren't there to cut it - they're there so you know what you're cutting before the invoice tells you.