ThinkPatternGet the app
Story
TECHNOLOGY · SEP 17, 2026

HPE Leader Proposes $/FLOP Metric for AI Infrastructure

Vinod Bijlani of Hewlett Packard Enterprise proposes using dollars per floating-point operation to measure AI infrastructure productivity and efficiency.

Vinod Bijlani, an AI practice leader at Hewlett Packard Enterprise, proposes the adoption of dollars per floating-point operation ($/FLOP) as a critical metric for measuring AI infrastructure productivity. This approach differs from traditional tokenomics, which measures the cost of consuming AI, by evaluating how efficiently capital, energy, and physical capacity are converted into actual computation.

Bijlani argues this distinction is necessary as AI shifts toward multimodal workflows and persistent agents. He notes that low token prices can mask underlying inefficiencies, such as rising power costs and memory bottlenecks. Data from Epoch AI indicates that while performance per dollar of chip spending has improved by roughly 49% annually since 2023, the demand for training compute for frontier models has expanded fivefold per year since 2020.

To optimize these economics, Bijlani recommends increasing utilization, optimizing data movement, and matching specific models to specific tasks. He also emphasizes the need to integrate power and cooling into initial designs and to measure the full platform rather than isolated components.


Reported across 1 outlet
Actors
Hewlett Packard Enterprise

Keep reading in the app

The full story and every source, free in the app.