Introduction | Research | Industry | Politics | Safety | Predictions #stateofai | 70 GPU demand sees NVIDIA print blowout earnings as it enters the $1T market cap club Q2 23 data center revenue was a record $10.32B, up 141% from Q1 23 and up 171% from a year ago. The stock was bearish for 2022 even though annual revenue came in at $27B, a 61.4% increase from 2021. NVIDIA now commands a $1.1T market capitalisation, up from $8.5B (130x) 10 years ago. stateof.ai 2023

Introduction | Research | Industry | Politics | Safety | Predictions #stateofai | 71 Selling faster than Coachella: GPUs snapped up from upstart infra providers CoreWeave, Lambda, and Crusoe Cloud, three selected NVIDIA partners that build and run GPU datacenters, together have tens of thousands of GPUs in their eet. Lambda made 9-gures $ worth of H100s available in its on-demand cloud and sold out in just over an hour. CoreWeave is one of the largest GPU operators in the market with a scale similar to several hyperscalers. The company is fully booked through the end of the year with their build schedule and are signing contracts in Q1 2024. stateof.ai 2023 Introduction | Research | Industry | Politics | Safety | Predictions #stateofai | 72 Private companies are shoring up NVIDIA GPUs and wielding them as a competitive edge stateof.ai 2023

Introduction | Research | Industry | Politics | Safety | Predictions #stateofai | 73 Footballers Compute is the new oil in Gulf States? Saudi Arabias King Abdullah University of Science and Technology (Kaust) has allegedly purchased >3,000 H100s to build a supercomputer, Shaheen III, that should be operational by end of 2023. Its LLM-focused researchers are primarily Chinese nationals that cannot access the US because their universities are restricted. Meanwhile, the United Arab Emirates Technology Innovation Institute in Masdar City, which developed the Falcon LLM, is also said to be procuring compute resources from NVIDIA. Finally, Abu Dhabi-based G42 entered into a deal with US-based Cerebras to procure up to $900M worth of the companys Wafer-scale compute systems and build 9 interconnected AI supercomputers. There is likely much spend more to come stateof.ai 2023 Introduction | Research | Industry | Politics | Safety | Predictions

Compute Index: NVIDIA A100 clusters The number of large-scale NVIDIA A100 GPU clusters has grown since last year, particularly at Tesla and Stability, as well as new clusters at Hugging Face. #stateofai | 74 stateof.ai 2023 Introduction | Research | Industry | Politics | Safety | Predictions #stateofai | 75 Compute Index: NVIDIA H100 clusters Its early days, but private and public companies are announcing new H100 infrastructure for large-scale model training. As of writing, Google and Inection are not yet at full scale and we understand others including OpenAI, Anthropic, Meta, Character.ai, Adept, Imbue, and more have signicant capacity. We expect more to come online soon. stateof.ai 2023