I think we can simplify the business model into three categories.
A . Compute resource [ CPU, Memory, Storage] provision over the cloud has lot of competition, however many MNC like HP, HPE, Dell will eventually join the game so, I would say this is a low margin proposition.
B. Compute resource, backup and services related to IP address, domain address registration etc can be a package to customer, this is an interesting space to be in because switching cost is higher if they have customers. But we don’t have data on number of such customers.
C. GPU as a Service.
This space is a very niche area and requires a lot of computing resources, having one GPU per server is of less use so, companies buy a ton of GPUs so, that customer queries get answered very quickly. E.g. Computers use GPU to simulate Weather patterns, Predict the failure of the supply chain based on orders, anomaly detection in turbines, DNA analysis, Pharma etc
So, there are two variables in the GPU equation, Training an AI model using large Data set and building an output from the model, aka Inference which is intelligent enough to predict or do the required goal in the shortest time.
With more data, the training of AI model will be good but it requires lot of trail and error as AI world is very specific to the use case. There is no standard way of doing things in AI world because data is unique.
The below graph from link shows how much GPU is required for training. Now if we map it to the e2e client, what can a customer do with 1 GPU and a small machine? That too all the machines run with virtual CPU/KVM.
What can the customer achieve then?
- They can do a prototype of an AI/ML model and train the model with a limited amount of clean data.
- They can measure the model outcome but for better inference lot of data and bandwidth is required.
So, Once we have data on the below points, we can understand the exact usage of the infrastructure and the possibilities of scaling this to new frontiers.
- Client Categorization
- GPU usage with respect to clients
- Number of customers onboarded
- Retention of customers on their platform.
- Number of customers using 1 GPU vs 4 or X GPU
Subscribe To Our Free Newsletter |