AI Investing: Make The Right Choices!
Throwing Money Into A Dark Hole - A Dollar Auction Paradox
The AI Gold Rush
The world is experiencing an AI frenzy reminiscent of the dot-com boom at the turn of the century. AI businesses are springing up daily, with venture capital flooding into anything related to artificial intelligence. Meanwhile, speculators are driving up the valuations of companies like NVIDIA to extraordinary levels, much like they did with Cisco and Microsoft 25 years ago.
The dot-com boom ended in bust, with most early entrants either going bankrupt or else achieving little or nothing - remember Pets.com, AltaVista, Napster and MySpace? Billions of invested dollars evaporated. Even the companies that eventually succeeded were so overpriced in 2000 that it took 15 years for their intrinsic value to catch up with their inflated share prices, leaving investors with a miserable experience.
AI investors should learn from history and not fall into the same trap as dot-com investors.
So, as savvy investors, how should we interpret this surge of activity, and how should we respond?
Behavioral economics offers us valuable insights. For instance, the "Dollar Auction" paradox in economic game theory highlights the potential for irrational and self-destructive actions, which are almost certainly at play in the AI sector.
What Is The Dollar Auction Paradox?
A one-dollar bill is up for auction, and the highest bidder wins the dollar. On the face of it that sounds like fun and a straightforward opportunity for profit - just don’t bid more than a dollar, and you stand to gain.
But there’s a twist. This is an open auction, where everyone can see the bids placed by others. Bids cannot be retracted, only increased, and everyone who places a bid must pay the value of their final bid, regardless of whether they win the auction.
You start by bidding one cent—a no-brainer! But then someone else bids two cents. Now, you must decide: raise your bid to three cents, or lose the one cent you've already committed. You choose to raise your bid.
This continues for a while and, before long, you find yourself increasing your bid to 99 cents. No problem you think, a small profit is better than nothing. But now the person who bid 98 cents faces a dilemma. If she stops bidding, she loses 98 cents. If she bids one dollar, she’ll break even, which is far better than losing 98 cents, so she takes that option.
Now, the pressure is back on you. You’re on the verge of losing 99 cents. The only option that makes sense is to bid $1.01 - if you win, you’ll only lose a net one cent, which is better than losing 99 cents.
This cycle of madness continues, with all players hemorrhaging money. But with so much already sunk into the game, the only choice seems to be to keep playing. The incremental cost of losing another cent always seems more palatable than losing the much larger amount already bid, so raising the stakes becomes the default choice.
This is the "Dollar Auction" paradox—a situation where competitive bidding spirals out of control, leading everyone to lose money. It’s a powerful example of how the attempt to act rationally and maximize personal gain can result in irrational and self-destructive behavior.
How Does This Relate To The Corporate World?
The aviation industry is infamous for destroying more capital than it has ever generated. Fierce market competition drove companies to overspend in their pursuit of market share, only to realize that the rewards were far less valuable than the costs they incurred. In this asset-heavy industry, businesses often felt compelled to spend more to protect their sunk costs, but this frequently led to a cash burn that spiraled into insolvency. This cycle explains why so many airlines eventually land in the corporate graveyard (pardon the pun).
A similar dynamic is playing out in the artificial intelligence (AI) sector today. Intense competition exists among tech companies, startups, and even nations to develop the most advanced AI capabilities. This “AI arms race” incentivizes ever-greater spending, even when the potential returns may not justify the escalating costs.
The speculative frenzy surrounding AI results in people buying in to the narrative about its future transformative value and entirely ignoring fundamental unit economics. Investors seem willing to place enormous and often irrational bets in this sector, so there’s currently no shortage of cash to burn through.
Once companies have invested significant time and money into AI projects, they become reluctant to give it up, and so costs mount and the problem compounds - is this not the Dollar Auction Paradox at play?
The AI Industry
The large language model (LLM) industry has witnessed unprecedented growth and investment over the past decade, with tech giants and startups alike pouring billions into research and development. This investment trend has accelerated in recent years, particularly following the success of OpenAI’s ChatGPT model.
Major players including Google, Microsoft and Meta have been at the forefront, investing tens of billions of dollars into developing new and improved AI models. Microsoft’s partnership with OpenAI, Google's work through DeepMind and Google Brain, and Meta's development of models like LLaMA highlight the scale of investment in this space.
OpenAI is a relatively recent 2015 start-up, and its remarkable success in competing with the tech titans, evocative of the biblical David and Goliath story, has resulted in other new entrants seeking to disrupt the market - companies including Cohere, AI21 Labs and Hugging Face have all entered the playing field.
One of the most impressive newcomers is Perplexity AI, which, despite being launched as recently as August 2022, is already challenging Google's dominance in search. Google typically presents a list of websites in response to a query, often cluttered with unhelpful sponsored links, that require users to sift through and conduct further time consuming research. In contrast, Perplexity AI offers a more streamlined experience. With a chatbot-like interface, users can ask questions conversationally, much like they would with another person. Perplexity AI then provides a comprehensive answer in clear, conversational language, drawing from a variety of relevant sources and including citations for easy reference.
The capabilities and utility of AI are evident to everyone, but this alone does not guarantee commercial success. The image at the top of this post was AI generated using DALL-E in a matter of seconds. Did I pay for it? No. Similarly, for the past year I have been reaping the benefits of a variety of LLMs for a multitude of purposes, yet I have not paid for any of those either.
Earlier, we examined the challenges faced by the fiercely competitive airline industry, which also represented a quantum leap for humanity. Before air travel, reaching distant destinations took days or even weeks by land or sea, but suddenly, anywhere in the world could be reached in a matter of hours. Yet, despite this transformative impact, airline after airline has fallen into insolvency, and those that survive continue to struggle with profitability.
Despite the massive investment flowing into AI, profitability in the sector remains elusive. The focus, at least until now, has been on growth and market positioning rather than immediate profitability and so many companies have dug themselves into deep financial holes with no strategic plan for how they will eventually climb back out.
The Google and Facebook model may provide an answer. Both companies offer a valuable service, but neither has successfully captured direct value from their user base - most users have never paid a cent. In reality, the users aren't the customers; they're the commodity being sold. Advertisers are the true customers funding these businesses. However, there's only so much advertising spend to go around, and it's uncertain whether this model would be effective for companies operating LLMs in any event.
Various strategies are emerging: these include offering paid API access, integrating AI capabilities into cloud services, enhancing existing products with AI features, and developing tailored enterprise solutions. Some companies, like OpenAI with ChatGPT, have also introduced consumer-facing subscription services.
As the technology matures and becomes more efficient, and as more use cases are developed, the path to profitability may become clearer. The shift from training-focused investments to inference-focused spending may also help improve the economics of LLM deployment.
There may be a huge addressable market, but the question is ‘which companies will capture that value?’
Creating Value Does Not Equal Capturing Value
Consider this: the advent of computer memory was crucial to the evolution of computing. Without the ability to store and retrieve data, there could be no computing. However, the companies responsible for pioneering memory technologies, which arguably made modern day computing possible, failed to capture much of the value they created.
'There are all kinds of wonderful new inventions that give you nothing as owners except the opportunity to spend a lot more money in a business that’s still going to be lousy. The money still won’t come to you. All of the advantages from great improvements are going to flow through to [others].’
Charlie Munger
This happened for a variety of reasons all of which seem to also apply in the field of AI.
Memory devices quickly became standardized and commoditized which led to intense price competition, reducing profit margins. In the AI market there are a multitude of models including GPT, Gemini, Llama, Mistral, Claude, Jamba, Perplexity, etc. Some of these are open source or else currently free to use. If all of your competitors are providing free or cheap access to similar models, taking a decision to charge for a service could be catastrophic commercially. Does anyone in the AI space have any pricing power?
Memory technologies evolved rapidly, making it difficult for companies to maintain a competitive edge for long due to rapid technological obsolescence. Since AI is evolving equally rapidly, this is a key risk for those involved.
Developing and manufacturing memory devices requires significant upfront investment, a capital intensity that also exists in the continual research and development of AI.
Many memory technologies were based on similar underlying principles, making it challenging to protect innovations through patents. Is the same not true in the world of AI?
While the memory sector was unable to extract the value that it had created, those offering a more differentiated product or service thrived. Software companies like Microsoft were able to create products with higher barriers to entry, with the added benefit of network effects and lock-in, allowing them to maintain higher profit margins. Semiconductor companies like Intel and NVIDIA focused on developing more complex, higher-value components (CPUs and GPUs) that were harder to commoditize and offered greater performance differentiation.
It could be that those developing the valuable AI models befall the same fate as those in the memory industry. They have laid the foundations upon which others in tangential industries will prosper.
Will any of the current LLM players survive?
Will the industry consolidate down to a few winners as is usually the case in the tech space?
Will we see vertical integration with those offering differentiated services and able to profit from the evolution of AI, integrating the AI technologists into their own tech stack? We have already seen an element of this with Microsoft making a sizeable investment in OpenAI.
Will LLM developers pivot into more commercially viable activities as Intel did when it pivoted out of the low margin memory business and into the very profitable production of semi-conductors?
Only time will tell how the AI sector plays out. However, one thing is certain, with so much sunk cost there is certainly an element of the Dollar Auction Paradox at play in this nascent industry.
Microsoft’s Dollar Auction Paradox
Microsoft’s efforts in AI, over a span of more than a decade, involved a dedicated workforce of over 1,500 researchers focused exclusively on internal AI projects and cost the company tens of billions of dollars. Yet, it found itself lagging behind OpenAI - a relative newcomer established in 2015.
This situation was met with fury by Microsoft’s CEO, Satya Nadella. When ChatGPT was launched, and was evidently leaps and bounds ahead of anything Microsoft had developed, he sent an internal memo to his AI team, asking, "Why do we have Microsoft Research at all?"
This scenario highlights that even the largest companies with the deepest pockets aren't guaranteed to come out on top in the AI wars.
Recognizing the need for a strategic shift after so much sunk cost and with its reputation at stake, Microsoft made a bold move by investing $13 billion in a profit-sharing partnership with OpenAI. This collaboration aimed to harness OpenAI's innovative approach and cutting-edge AI models, including GPT (Generative Pre-trained Transformer) technology.
The partnership with OpenAI has enabled Microsoft to quickly regain ground in the AI sector. Thanks to this collaboration, Microsoft’s Bing now has enhanced search capabilities, Microsoft 365 now has Copilot - AI-powered assistance across all Office applications, and Copilot is also available on GitHub as a programming tool for developers.
However, Microsoft isn’t putting all its eggs in one basket. Despite its close partnership with OpenAI, the company continues to diversify its AI investments.
Despite the partnership with OpenAI, Microsoft and Meta jointly announced the release of Llama 2, Meta's open-source LLM. Microsoft also recently announced a $1.5 billion investment in G42’s Arabic LLM - this is an AI company based in the United Arab Emirates.
Beyond these collaborations, Microsoft continues to invest in its own AI research and development and is pursuing its own initiatives. Projects like Project Silica for sustainable data storage, and Project AIM for solving optimization problems are part of this broader strategy.
In 2021, Microsoft strengthened its healthcare AI capabilities with the $19.7 billion acquisition of Nuance Communications. Through its venture capital arm, M12, the company has made smaller investments in various AI startups. Microsoft has also forged partnerships with academic institutions and industry leaders to advance AI research and applications.
In many ways, it feels as though Microsoft is trying to win at the race track by betting on every horse in the race. It doesn’t appear to be a very coherent strategy. Rather than using a sniper rifle to focus on a specific target, it seems to be using a scatter gun approach and simply hoping for the best. This raises the question: has Microsoft trapped itself in a Dollar Auction Paradox, where it risks losing money regardless of the outcome?
Alternatively, this strategy might be more calculated—possibly aimed at attracting as many AI developers as possible to its Azure cloud platform, with these significant investments serving that purpose. If so, it’s a high-stakes gamble with potentially long-term payoffs.
Amazon’s AI Strategy
Amazon’s strategy in the AI market centers on leveraging its market-leading cloud infrastructure, AWS (Amazon Web Services), to become a key enabler for AI and machine learning (ML) applications across industries.
Amazon has introduced its own large language model, Titan, which offers capabilities similar to other popular LLMs like GPT, with one key distinction: unlike other LLM developers, it is not looking to promote its own AI models, allowing it to sidestep the Dollar Auction Paradox entirely. What sets Titan apart is its seamless integration with various AWS services, making it easier for businesses to incorporate these advanced AI capabilities into their existing workflows.
Ultimately, Amazon aims to make AWS the go-to platform for AI and ML (machine learning) by offering a wide range of tools and services. These include pre-trained models, managed AI services, and infrastructure that supports custom model training. True to Amazon’s customer-centric approach, the company provides a broad range of options to meet diverse business needs.
Services like Amazon SageMaker allow developers with varying levels of expertise to build, train, and deploy AI models with minimal overhead, so AWS's scalable infrastructure is arguably best suited for the job.
For example, Amazon SageMaker enables developers with varying levels of expertise to build, train, and deploy AI models with minimal overhead, leveraging AWS's scalable infrastructure, which is arguably the best suited for the task. Amazon has also actively partnered with various companies and research organizations to integrate popular AI models into its ecosystem. Although Amazon doesn’t own stakes in these models, it has made OpenAI's GPT, Anthropic’s Claude, and Cohere’s LLM available through the AWS Marketplace, allowing users to access cutting-edge AI technologies without requiring Amazon to develop them in-house. By supporting third-party AI models, Amazon aims to democratize AI, making it more accessible to a broader audience.
In late 2023, Amazon announced Amazon Q, a generative AI-powered assistant designed to help businesses with various work-related tasks. Amazon Q integrates with a company’s existing systems and data and is customizable to meet each organization’s specific needs. It can be seamlessly connected with various AWS services and third-party business applications.
In other words, Amazon is entirely agnostic in relation to who wins or who loses in the AI wars. As the AI infrastructure provider, with a differentiated product that benefits from scale economies, network effects, a customer centric focus, a competitive pricing structure, the advantages of being a market leader and high switching costs, it is difficult to see how it could fail to thrive as the AI landscape evolves.
The Amazon approach is certainly far more coherent than that of Microsoft and far less risky.
Conclusion
During California's great gold rush of 1849, the real winners weren't the gold prospectors but those who sold picks and shovels. While a few prospectors struck it rich, most lost everything in their quest for fortune. If we could travel back to the mid-1840s, the smart move would be to invest in those selling the tools, rather than gambling on which prospector might find gold.
History may not repeat itself, but it certainly rhymes. In the AI equivalent of the gold rush, Amazon appears to be the one selling the best picks and shovels. As between Microsoft and Amazon, I know where my money is being invested.
Discussion
How are you investing in AI? Post your ideas and reasoning in the comments below.
Please share this post with others who can contribute meaningfully to the discussion.
Let’s get this investment community working!
I think that most business will benefit from AI not just tech giants. Traditional companies like those in Europe will improve their margins and increase profits, I think that the biggest opportunities are in these companies where nobody expects growth.