LLM End Game

In her book “Technological Revolutions and Financial Capital,” economist Carlota Perez explains the life cycle of great surges of economic development, such as the Industrial Revolution; the age of Mass Production; or the Information Age (internet). Each evolves in a cycle beginning with some defining event that pushes technology onto center stage, like Henry Ford’s assembly line. The next stage is the frenzy, when expectations tend to get out of control to varying degrees. That is followed by a turning point that evolves into synergy as the new technology is integrated into society, and of course, followed by maturity, leading us into the next great cycle.

We envision AI traversing a similar path. The introduction of ChatGPT was the big bang moment that catalyzed the dawn of the AI era. We now find ourselves in the frenzy. As long-term investors our desire is to ‘see through’ this frenzy and stay committed to ideas that will survive into the synergy phase. That isn’t easy, but Marc Weiss, CIO of Open Field Capital, and co-manager of the Palumbo Emerging Growth Portfolio, offers us his view of the frenzy, where it is heading, and how we should respond.

All participants can’t win. They never do. But Weiss does not subscribe to a zero-sum game scenario which emphasizes the losers, rather, he envisions the growing the pie and win-win situations. That pathway is filled with bumps in the road, some of them can be rather significant bumps. Those bumps are typically the catalyst for investors to make for the exits. Weiss believes those that see the cycle as long-term and are open to mid-course corrections are in position to maximize the opportunities these huge trends generate.

Our thesis is that the universal bullishness surrounding the AI theme creates the risk that some/many investors may lose patience leading to AI stock volatility and downside risk. Most investors are on board with the AI theme and are heavily invested in the ‘big names’ that dominate the major stock indices. In his report, Weiss advises that revenue expectations and valuations are becoming dangerously inflated and, if we may put words in his mouth, we take Wayne Gretzky’s advice and skate to where the puck is going, not to where it is now. Here is our quick summary of Marc’s report. If you would like a copy of the full report, please contact us. We are happy to send it along.

The AI Revolution is Real

AI is writing software; AI passed the Medical Licensing exam with a 100% score; AI is diagnosing radiology images; AI is helping to discover new materials and drug molecules; AI robots are showing rapid improvement in capabilities, etc. The enormous capabilities of AI are becoming more evident each day.

Whether AI can produce sufficient financial returns to justify its enormous cost in any reasonable time frame is more uncertain. For example, NVIDIA’s CEO, Jensen Huang, envisions a world where AI spending will be 10% of a knowledge workers’ salary. In another breath, he suggests that AI can capture $10 trillion of revenue on a $50 trillion of human intelligence driven GDP base. That’s 20%. That is a very big difference. Which is correct?

Weiss believes that such grand predictions may not come to pass. Vibe coding (AI writing software code) is already producing companies at the fastest pace ever. But vibe coding monthly subscription prices are mostly in the range of $10-$50 per month. How this market evolves is very uncertain, however, what appears to be certain is that expectations for LLMs are already reaching the most optimistic scenarios.

Technical Challenges: “Bumps in the Road”

AI development, while incredibly impressive to date, is not without its challenges. The overall trend in AI hallucination rates has shown consistent improvement across many traditional large language models, however, rate of accuracy improvement is slowing. Compute is growing faster than data so incremental data provides smaller accuracy improvements. (Hallucinations occur when an LLM generates a plausible sounding but incorrect, misleading, or fabricated response.)

To counter this growing problem the industry has rapidly adopted reasoning models. (A reasoning model is a type of AI system that solves complex problems by breaking them down into a series of logical steps before producing a final answer.) Ironically, this has produced more hallucinations, not less. New techniques, such as reinforcement learning, are now being developed to reduce hallucination rates. The question is whether this will be enough. And if it is enough, will the reduction happen as quickly as hoped? It is not yet clear that the industry will be able to solve the hallucination problem enough to enable mass enterprise deployments of high value/complex workflows and justify spending 10% of a knowledge workers salary to get it.

Over the long term, Weiss suggests that Huang’s assumptions may indeed be correct, however, we could see a multi-year period where enterprise AI spending growth fails to meet current expectations.

Consolidation Ahead

If future AI revenues are smaller than expected and all the frontier LLM models have similar performance, as suggested by Dev Ittycheria, MongoDB CEO, then there may not be enough revenue to support all the current players in the market. Google, OpenAi, X.ai, Anthropic, Meta, Mistral, etc. are engaging in a continuous and massive LLM training spend. It is well within the realm of possibility that one or more of these large LLM model builders may potentially drop out of the game in the future.

We have seen such consolidation before in the semiconductor industry. In 2007, there are twelve semiconductor companies that could produce products at the then current cutting edge of technology, which was 45 nanometers. Today, there are only two companies that can produce at the current technological edge of 3 nanometers.

What’s Next? Is this a Bubble?

If LLM revenues are less than expected, then we expect that the pressure to use more cost-effective chips will increase. This could cause a share shift towards NVDA challengers such as privately held Cerebras and Groq.

There is a lot of talk about an AI bubble. Weiss prefers to think of this as an evolution of understanding. In 2000, the internet was an amazing new world. This led to Lucent, and Cisco being among the most valuable companies at the peak in March 2000. Investors were correct to be excited, but it was so early in the cycle that investors did not fully understand how to invest in this new internet trend. Many of the largest companies today developed from that time. Google and Salesforce.com IPOs were in 2004, and Amazon Web Services was introduced in 2002. The iPhone was introduced in 2007. Weiss concludes that if there is an LLM driven correction it is important to note that the AI investment opportunity is much broader than LLMs alone.

Have a great week!

What We’re Reading

 

Palumbo Wealth Management (PWM) is a registered investment advisor. Advisory services are only offered to clients or prospective clients where PWM and its representatives are properly licensed or exempt from licensure. For additional information, please visit our website at www.palumbowm.com.

The information provided is for educational and informational purposes only and does not constitute investment advice and it should not be relied on as such. It should not be considered a solicitation to buy or an offer to sell a security. It does not take into account any investor’s particular investment objectives, strategies, tax status or investment horizon. You should consult your attorney or tax advisor.

The views expressed in this commentary are subject to change based on market and other conditions. These documents may contain certain statements that may be deemed forward‐looking statements. Please note that any such statements are not guarantees of any future performance and actual results or developments may differ materially from those projected. Any projections, market outlooks, or estimates are based upon certain assumptions and should not be construed as indicative of actual events that will occur.

All information has been obtained from sources believed to be reliable, but its accuracy is not guaranteed. There is no representation or warranty as to the current accuracy, reliability or completeness of, nor liability for, decisions based on such information and it should not be relied on as such.

All information has been obtained from sources believed to be reliable, but its accuracy is not guaranteed. There is no representation or warranty as to the current accuracy, reliability, or completeness of, nor liability for, decisions based on such information, and it should not be relied on as such.

The views expressed in this commentary are subject to change based on the market and other conditions. These documents may contain certain statements that may be deemed forward‐looking statements. Please note that any such statements are not guarantees of any future performance, and actual results or developments may differ materially from those projected. Any projections, market outlooks, or estimates are based upon certain assumptions and should not be construed as indicative of actual events that will occur.

Past performance is no guarantee of future returns.

, , , , , , , , ,

General News

By: Adam