Every company says they need GPUs. How many actually know what they’re doing with them?
Every company says they need GPUs. How many actually know what they’re doing with them?
A few years ago, the topic of conversation in the tech world was “We need to migrate to the cloud.”
This year, “We need GPUs!” is a similar statement in energy but contains a different buzzword.
The process for cloud adoption occurred the same way.
Hence the current activity at ground level is as follows.
Many tech companies are currently being pressured from above to find GPU capacity. They go out and find it. They provision resources from AWS, Azure, or GCP. They find NVIDIA A100s or H100s. They then pass it on to their engineering group. The engineering group find the answer for, “What are we trying to build?”
GPUs simply form the framework and base layers for AI computing. Consequently, acquiring GPUs is entirely fruitless without knowing how they will fit with the types of algorithms being executed, the architecture, the data flow, the training processes, and possibly having personnel to operate them.
The majority of companies have one or more of the above-listed components. Very few possess all of them.
Three elements demonstrate that companies who are successfully creating value with invested GPU capital share:
- A business challenge to solve versus a technology challenge to attain.
- The data necessary to complete the efforts was acquired prior to, or concurrent with the acquisition of the GPUs.
- Outputs generated were assessed from the first week rather than the last quarter.
Investing in computing capacity for AI is smart. The critical distinction between being GPU-rich and being able to use them for AI exists. Many companies are yet to progress to the AI-ready stage from the GPU-ready stage.