‘Some people get rich studying artificial intelligence. I make money studying natural stupidity.’ — Ziad K. Abdelnour
On Friday news broke of Sam Altman being removed as CEO of OpenAI by the board for undisclosed reasons but a memo stated that no malfeasance was behind the decision. Through a flurry of news over the weekend and protest by employees within the company the board was forced to spend time reconsidering the termination but ultimately Sam will not return to his post and Microsoft has hired the former CEO in what some view as an attempt to prevent him from creating a new company. Nearly 500 employees of OpenAI have signed a letter saying they may quit and join Sam Altman at Microsoft unless the startup's board resigns and reappoints the ousted CEO. The response by employees and the resignation of Greg Brockman co-founder of OpenAI along with the hype that the company has received through ChatGPT is reminiscent of many over hyped companies of the past where the CEO was given a bit too much credit for his capabilities or the companies ultimate ability to deliver. Companies such as FTX and Tesla are the first to come to mind in relation to the view of the CEO as a god by many of their fans.
Altman has an interesting curriculum vitae including becoming a partner in Y combinator in 2011, the Reddit CEO for eight days, and even a $43.3M exit in a failing location-based social networking company that couldn’t gain users. It is always fascinating to watch how good money chasing bad helps propel some of the venture capitalist(VC) backed companies regardless if they are ever successful except on paper or through more paper. But energy is what we are here to discuss on this substack and in this sector there are plenty of those type of overly inflated promises chasing growth that turn to dust just like the VC world along with plenty of fake tech related ideas with no substance and plenty of cash burning such as Bitcoin mining off excess natural gas. However ChatGPT and many of the artificial intelligence(AI) platforms are incredibility intriguing from an energy input standpoint because these platforms have an insatiable appetite to burn resources that aren’t just dollar bills.
Tuesday, Nvidia(NVDA) will announce its quarterly earnings and be highly focused on as the current darling of the market receiving much attention for the underlying engine that is driving the AI revolution with its chips. Nvidia is the undisputed leader of AI chips, with an estimated 80% to 95% market share, and in its last reporting of a 101% increase year-on-year(YoY) in revenue at $13.51B.
The focus of growth according to Nvidia is on the Hopper-based HGX platform that has increased the data center revenue by 171% and compute growth by 195% which is mostly used for large language models(LLM) and generative AI or basically platforms like ChatGPT that spit out human readable interpretations of data that read more like a story. An estimated 6.3B of the total revenue for the first quarter came from the sale of HGX components that go into the DGX servers translating into 36,600 machines and around 293,600 GPUs. Nvidia is still in the early stages of ramping these units in early 2023 and leading into this quarter should show even stronger rate of sales but it’s hard to fathom how much expansion is needed when platforms like ChatGPT have already lost much of their luster from the initial release with global user decline of 3% as the platforms show little actual use case in the real world.
Nvidia shows specs of 10.2KW as the max consumption of the DGX H100 which is about 1.6x higher than the DGX A100. This is on account of the higher thermal envelope for the H100, which draws up to 700 watts compared to the A100’s 400 watts. Though power consumption is not linear at scale when calculating the throughput of these platforms it is also safe to assume that a large portion of these cards will not be sitting idle in the data center.
According to Bloomberg New Energy Outlook (NEO) the energy transition scenario requires 46,000 terawatt-hours of power generation in 2050, nearly double today’s amount. The Net Zero Scenario, however, requires more than 80,000 terawatt-hours of generation, more than triple today’s amount.
As the growth of these large language models continues and the compute power of the brain to write and develop is transitioned into AI these numbers may be well underestimated for the electrical grid demand projections. Competing with the human mind on a per watt basis is an unpractical comparison since these machines can operate 24/7 but much like the biological competition these models tend to make a lot of mistakes and need a ton of fact checking at their current level of function. Yet the leverage of these platforms has shown that Hyperscale data centers have doubled their energy demand from 2015-2021 with expected growth to continue into the following years.
Economic development will always remain king when put against any obstacle including the goals set forth to combat climate change. Imagine the priorities of a person in such extreme poverty that their daily budget is less than $10 per day. That's the reality of nearly half the world's population of 7.88 billion where 47% of the world lives on less than $6.85 per day. As a reference in the United States a person is counted as being in poverty if they live on less than roughly $24.55 per day. According to the World Bank Poverty and Inequality Platform India has 136.81M people living in extreme poverty on $2.15 per day.
There is high demand on energy to bring a large portion of the world out of poverty to meet some western standards of home heating and transportation. Increase there is also the added aspect of unlocking the wealth of knowledge for the masses that has been expressed by some as infinite in these AI driven platforms. Today a small portion of Earth’s +8.1 billion people have access to the technologies at scale and demand, like most everything requiring more energy, is set to increase. It will be interesting to watch over the next decade to see what scale AI can achieve and the total energy it will require in addition to projections that may have been shortsighted at true scale. Overall we at Energy Crisis are betting on the fact that demand is understated for a world very hungry for energy and all input types, including fossil fuels, will need to grow to meet the needs.