AI: The new AI 'Table Stakes'. RTZ #473

...the Ante goes up both in AI capex and LLM AI capabilities

The Bigger Picture, Sunday September 8, 2024

As we go into the Fall of 2024, just two months shy of the second anniversary of OpenAI’s ‘ChatGPT moment’ of November 2022, the ‘Table Stakes’ in this AI Tech Wave have gone way up for the top LLM AI players. And contrary of consensus, it’s not just the tens of billions of AI infrastructure capex per company. It’s increasingly the collection of top features and capabilities of the top LLM AI platforms. That is the ‘Bigger Picture’ I’d like to cover this Sunday.

This ‘Table Stakes’ capex point was raised this week by Linkedin founder and Microsoft Board member Reid Hoffman, as the Information highlights in “OpenAI Investor Hoffman Says Musk’s New AI Chip Cluster Is ‘Table Stakes’”:

“Elon Musk’s recent claim that his xAI startup had assembled a cluster of 100,000 Nvidia chips in four months represents “table stakes” for the kinds of artificial intelligence Musk is pursuing, venture capitalist Reid Hoffman said Thursday at The Information’s AI Summit.”

“I don't think it's necessarily [that] he's in the lead,” said Hoffman, a partner at venture firm Greylock Partners and a director at Microsoft. He said such a cluster allows xAI to catch up to others that have been investing heavily in AI for years.”

Reid was of course referring to Elon Musk’s proud declaration of one of the larger clusters of AI super computers assembled:

“The biggest artificial intelligence developers are abuzz over Elon Musk’s claim on Monday that he completed a cluster of 100,000 Nvidia H100 chips in four months, to be used to train the next models of his startup, xAI.”

“If true, the cluster would be the most powerful cluster of graphics processing units in the world, but its rapid buildout would fly in the face of conventional wisdom about how long it takes to build such a system. Normally, completing such a cluster could take a year, according to two data center developers who spoke to me.”

“Such a feat would be very on-brand for Musk, who’s had plenty of success achieving nearly impossible goals. But Musk also has a penchant for exaggerations, and his claim about the cluster, dubbed Colossus, seems to be another such example.”

“Musk may be overstating how many of the GPUs are actually operating in a single cluster. No other company has been able to successfully string together 100,000 GPUs, due to the limitations of networking technology that connects chips to each other so they can act like a single computer. Still, a networking solution may be on the way: several such supercomputing chip clusters are in the works, according to The Information’s AI Data Center Database.”

The context of the higher ante and ‘Table Stakes’ of course has been echoed by top tech CEOs from Meta’s Mark Zuckerberg, to Microsoft’s Satya Nadella, to OpenAI’s Sam Altman and beyond. Again, as Reid Hoffman clarifies, despite Wall Street doubts as I’ve discussed:

“Analysts and investors have begun to question how much companies such as Microsoft and Meta Platforms are spending on AI. But Hoffman said he doesn’t think the spending is excessive, given the opportunity presented by AI.”

“Tens of billions of dollars is actually in fact very cheap” for tech companies valued in trillions of dollars to spend on AI “to be central to the next most important general purpose technology wave that's coming at software,” Hoffman said.”

The additional ‘Table Stakes’ I’d like to highlight that are a reality beyond the prodigious amounts of AI infrastructure capex for AI chips and data centers, are the increasing sets of capabilities these companies need to field as AI Scales.

This includes world-class AI Voice and multimodal capabilities, increasing AI Reasoning and Agentic roadmaps towards AGI (Artificial General Intelligence), Enterprise grade memory capabilities, and all done of course with the highest guard rails for Safety, Privacy and Trust as the AI Scales.

All higher ramping execution bars for all players in this AI Tech Wave. We’ve not seen this number of ‘Table Stake’ increases, to these kinds of levels in prior tech waves.

It’s already taking a toll on the best-financed AI companies beyond the ‘Mag 7’, as evidenced by the rising number of ‘AI Acquihires’ we’ve seen to date. And that trend is likely to be extended into 2025. As Bloomberg highlights in “AI Startups Struggle to Keep Up With Big Tech’s Spending Spree”:

“AI Startups Struggle to Keep Up With Big Tech’s Spending Spree”.

“The divide between the haves and have-nots in the AI industry is getting harder to ignore.”

“Adept, Character.AI, Aleph Alpha and other high-flying AI startups that had raised hundreds of millions in capital — a tremendous amount by most standards in tech — are now struggling to compete in a fast-moving industry against better-funded rivals. Some are being carved up in pseudo acquisitions and watching as their leaders get raptured by larger tech firms.”

“Meanwhile, a select few companies with deep pockets are trying to maintain their dominance by investing aggressively in the computing power, training data and talent needed to develop the next generation of AI systems. The list includes Meta, Alphabet and Microsoft, as well as a handful of startups like OpenAI, Anthropic and Elon Musk’s xAI, each of which have raised billions.”

It means, all the top companies have to be able to ‘chew gum and walk at the same time’. Even as new major LLM AI companies enter the fray.

They all have to execute on multiple sets of AI capabilities concurrently and globally. Serial execution is less and less of a luxury.

Especially when the race is also on to secure DISTRIBUTION to billions of mainstream users for the increasing range of expensive AI services. Like What OpenAI executed in its partnership with Apple. As it executes on its own ‘Apple Intelligence’ AI Strategy. And as Google is doing with its AI Gemini and Gemini Live strategy.

And that is likely to thin out some of the participants sooner than later as the AI Ante continues to climb into 2025.

That is the ‘Bigger Picture’ to keep in mind on AI ‘Table Stakes’ going forward. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)