Why $META is our #1 AI Pick
$NVDA, $SNOW, $DDOG and Instacart mentions
ChatGPT was a fad! It captured the minds and hearts of the world by spitting out detailed answers to questions using any information it held in its vast and growing data base. As for accuracy, well, not so much. Some even call the output from these chatbots, hallucinations. The stock of Internet search leader Google with its 97% market share was massively derailed when investors suddenly believed Microsoft Bing with its ChatGPT functionality would steal market share. Google didn’t help itself with a botched introduction of Bard. Fast forward to today and ChatGPT usage has peaked with monthly visits dropping from 1.8 billion in May 2023 to below 1.5 billion in July. Lo and behold, Google still retains its 97% market share.
Yet, the AI revolution and accelerated computing are still in the first inning! Nope, there is no terrific business model for ChatGPT offering its large language model (LLM) to the masses while buying tons of processing power without generating revenues commensurate with usage. In fact, a shady press article out of Asia claimed OpenAI was going bankrupt given their lack of revenues and massive costs to running their data centers, also an exaggeration. On the most basic level, AI is based on accelerated computing. It allows you to take massive amounts of data, learn and identify patterns and deliver real time information to the user, namely corporations. LLM without accelerated computing is the equivalent of a traffic jam on the highway. The cars can only move as fast as the one in front of them and unequal driving skills cause even more chaotic results (how fast you can get to your destination). Running AI on CPU’s would be satisfactory, getting better and better with each passing generation of processor, but NVIDIA has invested, captured and owns the market for accelerated computing. An executive at NVIDIA was quoted as saying one AI server can process the same amount of data as 1,000 general purpose CPU servers. The $40,000 H100 price tag is perhaps 10-20x more expensive, but we are still then talking about a 50-100 times performance advantage on per dollar spent basis!
Hock Tan, the CEO of Broadcom was asked on his last conference call about his view of AI being additive or cannibalistic to the growth of his computing end markets. He candidly said it was too early to tell. Both Oracle and META said they plan to buy a lot of CPUs to sit alongside GPUs in their data’s centers. Marvell Technology reported mediocre revenue growth this past week with AI “labeled” business growing dramatically offset by traditional legacy products which they argued are seeing an inventory correction which will be cured in the quarters ahead. In our minds, this remains to be seen. Nvidia’s dynamic man of the hour, Jensen Huang, proclaimed on his conference call the age of accelerated computing and generative AI were upon us. With over $1 trillion in data center equipment installed globally and $250 billion in annual spend shifting from general purpose computing to accelerated computing, the capital spending cycle is just beginning. Ok, deep breathe.