Following Nvidia’s remarkable quarterly results – revenue of $22.1B, 265% yoy growth, net income of $12.3B, 769% yoy growth – the company has now achieved a market cap of $2.19T. To provide perspective, this market cap surpasses the combined market caps of all companies listed on LatAm’s stock exchanges.
Given this context of surging demand for AI chips, VC capital is naturally flowing into AI chips startups.
AI chips address two critical phases of the AI lifecycle: (1) the training of new models based on existing data and (2) the inference of those models on live data in production.
Training-focused startups establish benchmarks based on the time required and the efficiency to train benchmark models. Cerebras (https://www.cerebras.net/), a Fabrica Ventures Fund I portfolio company, is now the most prominent training-focused startup and it has been training benchmark models more than 10x faster than Nvidia’s DGX A100 GPU clusters.
Inference-focused startups, in comparison, stand out for their speed and energy efficiency of inference on already trained models. Inference-focused startups usually focus on specific use cases, including speech detection, object detection, and facial recognition.
Groq, an inference AI chip startup, gained widespread attention after this interview last month: (https://www.youtube.com/watch?v=pRUddK6sxDg). It has raised $363M in equity and was valued at $1.0B in 2021 — D-1, Tiger and Social Capital are the main investors.
After the interview everybody is talking about Groq. And developers are flushing into the platform; there are already more than 16K apps developers utilizing Groq ecosystem.
“The best way to think about LPU [Language Process Unit] is all of these other architectures you’re hearing about are really good at parallel compute. Some of them are really good at sequential compute. The thing is, you can’t produce the hundredth word until you’ve produced the 99th. Very much like a game of go or chess, language is the same thing. It’s just a larger space. So, we are very good at sequential problems.” — Jonathan Ross, founder and CEO of Groq.
Conclusion
When we invested in Cerebras in 2021, Statista forecasted that the AI chip market would increase $70B from 2021 to 2027 (from $13B to $83B) and now we see that it completely missed the AI chip market exponential growth.
Groq is on our Fund II short-list.