Are AI Stocks Overvalued?
How AI Hype Could Be Hiding a Pending Market Correction
Introduction
In November of 2022, OpenAI released a text-based user interface to their AI Model “ChatGPT”. Little did they know, this was an inflection point in human history. This easily accessible software hit 100 million users in a record two months (for context, the previous poster-child of 21st century technophobia, TikTok, took nine), leading to a grandiose entrance into the ethos of society seen scarcely throughout history. Along with this rise of chatbot adoption, unsurprisingly, was the rise of companies trying to capitalize on the promise of an automated workforce… labor costs being the highest expense in most companies; executives were drooling. Investors, commercial and retail alike, did not shy away from the AI-mania that ensued in the following two years (and continues to this day). This, of course, is epitomized no better than the historic rise of Nvidia - currently the most valuable company on earth.
If you’re someone who has used ChatGPT, or any of it’s leaderboard-stretching1 counterparts, you’ve certainly found what they can do to be uncannily impressive. From conversations that sound all too human2, to solving our most complicated questions (except for the comically obvious blunders it still produces), to creating near-photorealistic images and videos on the horizon… it’s safe to say that these “AI” are incredible. Knowing all this, how could I doubt that AI stocks are worth every penny?3 Allow me to unpack that claim below.
1 - Failures of implementation
The most pressing indication that AI stocks are overvalued are the widespread failures of implementation - either by a lack of implementing systems, or more often, implementation that isn’t adding value as quickly as people expect it to.4
The obvious use-case to look at is programming. People are often concerned that programmers are now programming themselves out of work, as AI begins to do the programming on it’s own. To be clear, this is a rational fear as the technology advances, but I’m aiming to address the current state of affairs. According to a recent study, experienced developers are ~20% less productive while using AI, while simultaneously feeling ~20% more productive.5 There is a disconnect.
Again, the use of AI in development leading to a reduced output is not necessarily indicative of the future; AI will keep getting better. But it tells a story of developers taking as much time implementing automation for a task as they would need to simply do said task themselves. Junior developers, unsurprisingly, report more productivity while using AI, but they also presumably have less requisite knowledge of high-quality coding to be able to assess the quality of the code that AI produces... i.e. they may also actually be less productive too.
Sure the incessant rise of processing power will counter much of these coding inefficiencies, but will it also mask underlying issues waiting dormant inside poorly-written code? If AI can't evolve fast enough, I suspect yes.
This does not mean that AI won’t be implemented successfully, but that where it has been implemented, the gains have yet to be fully realized, and I suspect they won’t be for the next few years.
"But companies are reporting good numbers."
Let me ask you a couple questions: When you call a service line, is the automated system significantly better than it was two years ago? At your job, do you really hand off your work to an agent to surf the web, find the answer, document it for you, and begin building out a solution without the need for heavy oversight & debugging? Are your AI generated answers on Google truly adding large amounts of value to your life verses clicking the first (unsponsored) webpage on traditional search? In my experience, the answer to all the questions is largely no.
That’s not to say there’s no improvements in quality of life: Chatbots allow for the inexpensive learning of language without the fear of public embarrassment, music-generation allows for those less musically inclined to make pleasant sounds, image-generation gives us a shortcut to creating hilarious memes, and workplace LLMs can be utilized to summarize disparate documentation sources to point us in the right direction. The question at hand isn’t “does AI add value?”, rather it’s “does AI advancement justify the expensive stock prices we see today?”
We’ve all heard the expression: “Lies, damn lies, and statistics.” I worry we are seeing this play out today - people follow incentives. After investing record amounts of capital into AI, you need shareholders to feel like it's paying off. This is functionally no different than any other large investment a company makes. They tell their (nearsighted) story through quarterly returns. Call it an intuition, but I'm suspicious of the effectiveness AI implementation is having on enterprise returns thus far. My gut tells me they're riding the AI publicity train, making money selling AI solutions to each-other, all the while praying that AI accelerates their future productivity fast enough to account for any financials they had to manufacture in the previous quarters otherwise unsustainably. (Any Charles Ponzi fans are no-doubt smiling ever-so-slightly).
2 - Automating away developers
We’ve already seen companies begin replacing junior developers with AI coding agents(ish). As the AI gets better, the companies will seek to replace the intermediate programmers with better agents, and even better AI… well, it’ll replace advanced programmers. While the decades-long timescale required to replace the computer science workforce seems all-too-generous given the fast pace of AI advancement, it’s still a risk. If for some reason we plateau in AI advancement soon, companies may have automated themselves to death. And worse yet, who will fix it? In this impending world, there will be few junior developers in the pipeline to become the next generation of top-talent to fix the systems when the AI get things wrong. But hey, maybe the handful of experts who pursue the path anyway will be all we need; they’ll be paid generously.6
3 - Staunch open source competition
Anyone who has stayed abreast to the worlds of AI &/or finance certainly remember DeepSeek’s “R1 Moment” and the reality check it gave the closed-model development companies of the west. But do you remember Kimi’s K2 moment? What about the recent GLM-4.5 model by Z.ai: a nearly state-of-the-art open-source model, that makes DeepSeek’s R1 look expensive. Why don’t these matter anymore?
Here’s a boring answer: I truly don’t know. While I admittedly don’t have insider knowledge of the models in the works (GPT-5 expected to release soon), when I look at the advancement trends (and the requisite capital to fund them), I don’t know why the more recent equivalents to “R1 moments” didn’t send shivers down the spines of investors. Has the market become complacent with open-source AI competition? Perhaps. Has the west grown distrustful of the numbers coming out of China, resembling those concerns following the uncovering of a now-famous virus SARS-CoV-2? Perhaps. Are there truly signs that closed US models are at a breakthrough that open-source competition will simply not be able to match? Perhaps.
I can’t confidently tell you the answer to any of these questions. I can, however, say it at least appears that we’ve failed to fully factor in our competition around the world, and it just might come back to bite us.
4 - The US chip moat isn't as it seems
Over the last couple years, the United States has held leverage over China. Namely, computer chips. While initially it was clear that US export restrictions on top-tier (Nvidia) chips hampered Chinese AI development (while also inadvertently forcing them to be more creative), this leverage is weakening.
For starters, China is researching next generation computer chips even faster than we are. Sure computer chip development takes time, but to fail to recognize that China is now putting at least as much effort into chip development as the US, is to be running with blinders. Not only is China sustainably accelerating it’s research, it has the added bonus of allegedly smuggling in Nvidia chips in the order of ~$1billion. While smuggling chips to surpass export restrictions is small in the grand scheme of the ‘next industrial revolution’, it can certainly have an impact on the tightening race for AI supremacy.
5 - China’s AI game-plan is a crowd pleaser
While America’s AI Action Plan makes it clear how we aim to win the AI race - deregulation, western-isolation, and accelerated investment - China has outwardly taken a different approach… Cooperation. While we should recognize the United States’ similar aim to eventually expand exports of her capabilities, it’s worth noticing how we’re doing it. Sure the average American’s (particularly Gen Z’s) views of China have begun to warm in the last couple years, we still largely view them unfavorably. Unfortunately for the US President, this sentiment is not universally shared. In fact, according to the 2025 Democracy Perception Index, the world views China more favorably than the United States (seemingly correlated with Donald Trump’s return to office). China has capitalized on this by promoting a brand of collaboration and thoughtful AI governance.
Sure we don’t know China’s intentions, and it’s easy to be cynical, but perception is often reality. And reality suits the victor.
Maybe AI Stocks aren't overvalued
Having laid out my case for why AI stocks may very-well not be worth the price the market has pushed them to, it may seem lacking courage to hedge here. That said, we simple don’t know the future, and it would be imprudent to espouse confident predictions about where we’re headed. So, in the name of either prudence, humility, or uncertainty - here are a few quick reasons that AI stocks may in fact not be overvalued.
My intuitions could be wrong about how incentives are manifesting in board rooms & financial statements. I’m by no means a forensic auditor, and to err is human. My intuitions, while based in what I read and see, are only as extensive as my own limited understanding.
China could be lying about the low cost of models, and secretly using smuggled chips that should be factored into higher-than-reported cost estimates.
To this point, while US AI-stock prices may be justified if Chinese open-source competition isn't being truthful, if they're finding ways to purchase top-tier Nvidia chips, then Nvidia stock (unlike other AI-adjacent stock) would be less affected by this discovery.
Stocks are forward looking, so growth in the future could surprise us. A paradigm shift combined with immense capital investment could certainly do this.
AI Singularity: AI reaches above-human super-intelligence, and continually improves itself, leading to exponential 'snow balling' of capability, i.e., an intelligence boom... In this scenario, even the high P/E stocks of today could be undervalued.
All is to say, I, and everyone else, don’t know for certain what will come of these sky-high stock prices and seemingly limitless AI potential. But, what is clear, is that we’re entering a new age and can’t look back.
Note: This newsletter is for informational and entertainment purposes only and does not constitute financial advice. See Full Disclaimer.
Examples: ARC-AGI, Simple Bench, LMArena - among others.
Remember the Turing test? Me neither.
In this case, “AI Stocks" refers to technology companies creating AI models, companies heavily implementing AI, or companies with AI-enabling products like chip manufacturers (Think: Nvidia).
SoftBank aims to have 1billion agents by 2026 (6 months from now). Do we need to debate if this maps to reality?
This study has a very small sample size, so take it with a grain of salt.
Albeit for different reasons, we are seeing how being in the top nth percent of AI programmers can be rewarding (see: Meta’s super-intelligence lab). This may mark the rise of the ‘Individual Contributor Billionaire’.