AI Is About Cost-Effective Computing Power at Scale
Will Sam Altman and Jonathan Ross be Microsoft's and Google's next CEOs respectively?
Customers at the end of the day have a problem they need to solve. The customer may be a builder of LLMs looking to reduce query time and to increase response accuracy by improving inference computations. The customer may be a cloud computing company that houses LLMs and other AI models that then makes those models available for commercial use. The customer may be a software application company that leverages various permutations of AI in its front-end and/or back-end processes. In each of these examples, processing speed matters, cost matters and effectiveness matters.
Speed matters. Speed matters, so it is critical to deploy the fastest chips in your process, correct? Not necessarily. Perhaps chip 1 is not quite as fast as chip 2, but when deployed across an array of racks, chip 2 generates faster, more cost-efficient outcomes than chip 1. Keep an eye on chip designer Groq to this end. Groq was founded by Jonathan Ross who was an early engineer on Google’s TPU effort. Groq is working on a capital raise as we speak. Perhaps Google or another cloud provider is looking to acquire Groq rather than make a discrete investment?
Cost matters. Cost matters, but it’s not as simple as comparing the cost of one chip to another. The equation to be solved is more about identifying which hardware and software configurations with which component parts are most cost-effective for achieving the desired outcome. I believe that Google Cloud Platform has an advantage over Microsoft Azure and AWS in part because it has been designing its own purpose-built TPU chips since 2016. Google is vertically integrated from a hardware and software standpoint - it knows how to build “systems” where the hardware and software are optimized for each other to achieve the desired result whether it be for a product such as Google Photos, the Google Pixel phone, or the Gemini LLM housed inside of Google Cloud Platform. AWS is taking a similar approach to GenAI, Apple will do the same, Nvidia has a cloud business and Groq wishes to build a cloud business. Google has a head start to this end and it is well-capitalized. Google does not have to play the capital raise game the way that OpenAI, Groq and others do.
Effectiveness matters. Obviously. Google, AWS and Azure have leased their remote compute power for better than a decade. LLMs and GenAI is the next leg of the cloud server business currently dominated by those three giants. I believe that Microsoft could be left behind if they don’t acquire OpenAI or an OpenAI competitor. Satya Nadella will never fully imbue Microsoft with the best AI minds if he opts for AI partnerships with companies such as OpenAI versus bringing them in-house.
My guess is that the next cohort of CEOs at Google, Amazon, Microsoft and even Apple will be professionals with deep chip / hardware / software systems knowledge. Perhaps Sam Altman will be Microsoft’s next CEO. Perhaps Jonathan Ross will be Google’s next CEO.