About the author
Evan Hill
TI Youth Observer
Defining "AI"
Intense venture capital interest in anything labeled "AI" has created a gold rush for expansion into the space. To paint a picture, approximately 154 billion USD was invested globally into "AI" in 2023,1 with Goldman Sachs forecasting this number to expand to 200 billion USD by 2025.2
To understand the economics of "AI," we must define it. "AI," as it is currently understood post-ChatGPT, refers to a program able to sort and output results in one or more natural language mediums, and for shorthand, this article will adhere to this definition. "AI" consists of two base software components: a large-scale generative algorithm, usually based on the transformer architecture proposed by Google in 2017,3 and a massive, curated dataset used to train the program. It is worth noting that the term "AI" is misleading, as these products are not actual artificial intelligence. These programs are better described as high-end word matching programs with outstanding natural language processing (NLP) systems, or large language models (LLMs).
Bad Business
Before delving into the grim economics of "AI," we should acknowledge that some systems labeled "AI" show great promise. Fields including transportation, programming, and food and beverage are being revolutionized by niche, specialized "AI" products which provide tangible value by improving output and/or streamlining labor requirements. Unfortunately, this is not the case for the overwhelming majority of products.
From a purely technical perspective, many "AI" products end up looking like solutions without problems. At best, they represent novel technical advancements with no immediate monetization model or practical use case, and at worst, they're cynical bundles of redundant tech in an over-saturated market or dressed-up vaporware seeking to cash in on a speculative hype bubble. An example of the extreme saturation and lack of technical moats (an innovation or advantage that is prohibitively difficult for competitors to recreate) in the space is illustrated by the approximate 1.8 million "AI" projects currently on GitHub, many of which are LLMs that function comparably to ChatGPT.4
A poster example of an "AI" company without a clear monetization model is OpenAI, creator of the immensely popular ChatGPT. With estimated daily operating costs of approximately 700,000 USD,5 and a projected 5 billion USD deficit in 2024,6 OpenAI exemplifies a series of interesting projects with no clear path to profitability. This could evolve assuming an innovative new approach or technical moat is developed, but at present, this seems unlikely.
Present evidence seems to suggest the development of "AI" moats is unlikely, at least in the LLM and generative art spaces. An ominous leaked 2023 memo from Google bemoans this exact problem, explicitly stating in its title that "We (Google) have no moat, and neither does OpenAI,"7 whilst drawing attention to several semi-open source projects that perform almost identically to Google and OpenAI flagship LLM products,8 at a fraction of the cost. The lack of a technical moat eliminates OpenAI and Google's lowest hanging approach to profitability, a variant of the Microsoft model - distribute moated equipment with excellent support at a loss until it is ubiquitous, then eventually profit by flipping and selling ancillary software as a service. This model does not work when everyone can rig up an LLM.
From a business standpoint, this is a nightmare. That anyone can download an open source LLM project comparable to ChatGPT, monetization models are not figured out, and operating costs are high, indicate LLMs are bad business. OpenAI and Google hold sway over a large user base from brand recognition alone, but that isn't enough. There is existential value in the user data collected from running an LLM at scale, but Google already has similar data, and it is difficult to imagine an application that offsets the daily 700,000 USD expenditure OpenAI faces emerging quickly. If LLMs had a strong technical moat, there might be a "uniqueness" justification, but this simply doesn't exist. In addition, OpenAI has implemented aggressive user monetization protocols on their products, but these paywalls cannibalize their only advantage - name recognition in a wide field.
The point here is not that the recent developments in "AI" are all bad. It is that many of the companies developing them, when subjected to casual scrutiny, display low potential for creating sustainable long-term value.
Incestuous Investment
The underlying base principle of technology investment is simple - early adopters and innovators create products which have natural technical moats and become ubiquitous, which is profitable to both the company (in terms of service fees or sales) and investors (in terms of the inevitable valuation spike). The natural progression of this schema encourages investors to adopt a "unicorn chaser" mentality, where they invest early in a herd of companies which sound promising, in hopes that a few reach a vaunted multi-billion-dollar valuation. This is where issues begin to arise. The race to get in early with the mere presence of prestigious institutional investors is often enough to immediately drive up the valuation of an early-stage tech company, despite the fundraising company frequently having no proven concept, profit model, or technical moat.
It must be stressed that the explosive early valuation potential of tech stocks is the primary catalyst for the situation we now see. Excitement and fear of missing out led to the tech sector becoming a degenerate, stagnant place where non-innovative, non-profitable companies without clear monetization plans or proof of concept can receive multi-million (sometimes multi-billion) USD valuations. Any such environment will inevitably become a haven for bad actors and speculative financial practices. We've seen this before with the dotcom bubble and crypto boom.
Bad actors within this system can easily exploit the situation. Cynical startups quickly realized that slick marketing and promising the world are quick ways to raise capital and receive a sky-high valuation. Cynical investors realized that the ability to almost magically raise a company's valuation could be quite profitable. Both cynical startups and investors are incentivized to parasitically exploit, then exit this scenario by offloading shares or selling the enterprise when valuations are peaking, leaving earnest investors and the new buyers with nothing. In many ways, this alleged investing schema strongly resembles a pump-and-dump scam.
While this practice is certainly slimy, what truly matters is that these high financial activities are at the direct expense of others and the real economy. Technology is valued because it has the potential to be almost unimaginably profitable through innovation - by divorcing actual innovation and technical moats from the process, we end up with a bizarre game of hot potato, where the potato is useless "AI" vaporware, the winners are cynical investors and startups, and the losers are everyone else. When a valuation skyrockets, money does not magically appear in the company bank account - the value of the shares held by investors and company employees expands. To extract capital from these shares, someone needs to pay for them, and assuming the technology made by the company has no ability to create money, cynical investors and startups are mutually incentivized to dump their trash projects on the public or another company.
It is worth noting that this phenomenon is not new, and not exclusive to the "AI" industry. "AI" simply represents the current path of least resistance and a slightly revised approach. A variant example is WeWork, a company at best tangentially in the technology and innovation space, that was able to "hack" capital sources and its own valuation by presenting itself as "cutting edge" and IPO, only to predictably crash in the most spectacular fashion.
Not Everything Is Doom and Gloom
Though the dismal cycle of usual bad actors will undoubtedly continue and move on after the hype around "AI" dies, it is worth noting that certain niche technologies in the broad categorization do show tangible value propositions with solid monetization schemas and technical moats.
Though the technology involved differs significantly from LLMs, an excellent example lumped into the "AI" grouping is self-driving cars, which have been deployed by 19 companies in 16 cities for limited test runs in China.9 The question of mass implementation for this technology is a "when," not an "if." The implications, new jobs, and opportunities this transformation will bring are almost unfathomable. Transportation, shipping logistics, city planning, traffic control, and countless other intertwined industries will experience a wave of change and opportunity as this warps the way we move and experience life.
Food and beverage, at least at the fast-casual level, has always been about providing consistency and speed. An enormous section of Hamburger University and the Speedee System as implemented by McDonald's is concerned with mitigating human error to the nth degree possible, ensuring consistency and optimized delivery times.10 "AI" kitchens, at least in the context of fast-casual, show tremendous promise, requiring less space than a traditional kitchen, operating extremely fast, having less wastage, requiring minimal personnel, and having no capacity for human error.11 These kitchens are still experimental and require human supervision, but the ability to almost immediately set up a food dispensary that operates 24/7 and prepares a wide variety of fresh dishes palatable to the market in question will be transformative.
On a somewhat tangential note, companies like Nvidia and AMD have taken a calculating long-term approach by selling metaphorical pickaxes in the form of GPUs to "AI" gold miners. Though not technically "AI" producers, Nvidia and AMD serve as foils for the "AI" industry, having sky-high valuations based on strong technical moats, profitable business models, and well-scripted plans for future innovation. Even then, Nvidia's valuation is vastly overinflated, and only makes sense if you believe the firm will spearhead a new industry, a proposition which may well be true, but has conveniently overlooked risk factors.
What all these industries and organizations have in common are connections to the real economy, in this case, represented by a moat, a business model built to be profitable, and real-world applications that generate value, as opposed to a convoluted plan to self-enrich utilizing smoke, mirrors, and high finance. "AI" is in a bubble - no industry can have so many obviously unprofitable and untenable enterprises and not be primed for a burst - but within this bubble of bad actors are companies and ideas that will endure through the inevitable collapse, because they contribute value in a tangible, measurable form.
Conclusion
The point of this is not to condemn "AI." It is to advocate for a real economy-based approach when evaluating the "AI" industry, and by extension the greater technology space. This is perhaps Sisyphean, as history teaches us that the cycle of financial upsides creating complacency, and eventually a speculative hype bubble, is something we tend to repeat. A trend throughout history is that those who understand an industry from a technical level and invest in products with solid monetization plans and real economy applications tend to see success. "AI" is no different. Much like blockchain, "AI" represents an advancement with niche technical applications which became over-hyped, and then overrun with bad actors seeking to profit at the expense of the public.
Technology should be appraised in terms of underlying ability to monetize, technical moat, and tangible value created. Too many venture capital funds focus on "what's hot" and appoint a class of software-illiterates to play a glorified game of following the leader in a desperate chase for a unicorn that is frequently a wild goose. The existence of this opened doors for unethical startup owners who are willing to stretch truths for valuation. One thing is clear however - those who evaluate products based on technical specifications and tangible value to the real economy will succeed long-term and weather the coming bubble burst.
Please note: The above contents only represent the views of the author, and do not necessarily represent the views or positions of Taihe Institute.
This article is from the September issue of TI Observer (TIO), which explores the current turbulence in the global economy, analyzing the effects of geopolitical tensions, supply chain fragmentation, development of financialization, hollowing out of the real economy, in order to shed light on future economic transformations. If you are interested in knowing more about the August issue, please click here:
http://en.taiheinstitute.org/UpLoadFile/files/2024/9/30/1810401645ddb395e-0.pdf
——————————————
ON TIMES WE FOCUS.
Should you have any questions, please contact us at public@taiheglobal.org
© 2013-2021 Taihe Institute (TI) . All rights reserved. Taihe Institute (TI) .京ICP备16023694号-1
Privacy Policy丨Terms of Service
Contact Us: public@taiheglobal.org
