Illustration by II
By the time a dormant penny stock company known as Applied Sciences managed to wrangle a listing on the Nasdaq in April 2022, it had reinvented itself as a cloud hosting service for bitcoin miners and changed its name to Applied Blockchain. But with the crypto world crashing that spring, the stock never took off. Within months, Applied Blockchain pivoted again — renaming itself Applied Digital.
If its previous iteration had been too late to cash in on the bitcoin mining craze, the company wasn’t going to miss the next big one: artificial intelligence.
Applied Digital’s stock finally began to soar in May, when CEO Wes Cummins announced that the company had signed a cloud hosting deal potentially worth $180 million with an unnamed but prominent AI customer, and another one for up to $460 million with another big player in the booming AI space. By July, the stock had surged some 450 percent for the year, becoming — for a time, at least — one of the big winners in today’s AI-driven stock market.
Applied Digital is hardly alone in trying to capture some AI magic. Ever since ChatGPT burst onto the scene last November — with its “large language model” boasting a humanlike writing capability that at first blush seems to ensure productivity gains for everyone from publishers and movie studios to investment banks and hedge funds — so-called “generative AI” has turned the markets on their heads.
Coming off the worst year in recent history for venture capitalists, private market players like Andreessen Horowitz, Coatue Management, Tiger Global, Sequoia Capital, and Softbank quickly redirected their dollars to AI upstarts. Meanwhile, the stock prices of the big tech names suspected to be the major beneficiaries of this often-called “revolutionary” form of artificial intelligence have skyrocketed. The 2022 tech downturn became a faint memory as some five tech companies — Alphabet, Amazon, Apple, Microsoft, and Nvidia — accounted for the lion’s share of the stock market’s remarkable comeback this year, with the S&P 500 rising 20 percent through July. Nvidia, maker of the superfast chips that switched from powering bitcoin mining to making generative AI possible, has gained more than 250 percent so far in 2023, making it the S&P index’s top gainer.
In the midst of this bonanza, Applied Digital’s CEO — who is also the president of B. Riley Asset Management and was at one point the owner of about 25 percent of Applied Digital’s shares — posted on Twitter (now X) in June that the company had ordered 26,000 top-tier H100 GPUs, or video processing cards, from Nvidia for $40,000 apiece. To those casting a wary eye on the company, it seemed too good to be true. A purchase that big would allow Applied Digital to “jump to the top of the pile in high-performance computing, alongside Google, Meta, and [Amazon Web Services],” short-seller Dan David said in a Wolfpack Research report that called the company “an embarrassing and predictable stock promotion.” He noted that the cost to purchase such equipment would run more than $1 billion — more than Applied Digital’s market cap of nearly $600 million.
“The explosion of interest in AI after the emergence of ChatGPT has predictably attracted the worst promoters and scumbags to peddle fake AI wares to credulous investors,” says David, who claims Applied Digital is one of them. (The company did not respond to a request for comment.)
Applied Digital is one of nearly a dozen companies that short-sellers have been eyeing this year as questionable beneficiaries of AI mania. The shorts say their antennae are on the alert for even more. But so far, the skeptics are fighting an uphill battle. Short interest is relatively high in several of these stocks, and it’s been costly to bet against them. As of August 25, Applied Digital short-sellers, for example, had placed bets on 19 percent of the outstanding shares of the company. The short-sellers in aggregate are down almost 30 percent this year on the name, having lost about $10 million, says S3 Partners — although those who shorted Applied Digital at or near its peak would have profited.
Orso Partners co-founder Nate Koppikar, who is also short Applied Digital, has a term for what he sees going on. He calls the phenomenon “the grift shift” — arguing that companies and venture capital funds have pivoted from their losing crypto and tech bets to cash in on the AI moment.
All told, generative AI and machine learning start-ups raised about $39.4 billion this year, with $19.4 billion of that in the second quarter, according to PitchBook’s second-quarter Artificial Intelligence & Machine Learning Report.
Though money has been pouring into everything said to be “AI,” a few critics are starting to wonder whether the latest technology is really transformational or merely evolutionary. Meanwhile, large language model problems like “hallucinations,” “drift,” and “degradation” are starting to populate the tech literature, in reference to the various types of errors cropping up in generative AI as studies indicate that the products’ output appears to have worsened over time.
Some investors are starting to grow antsy. AI has dominated the spotlight this year, but “investors are growing impatient with the lack of revenue growth from generative AI innovators,” says PitchBook. “No longer are big tech stocks going up after new-product or partnership announcements, putting pressure on start-ups to gain traction.”
Says Koppikar, “This looks like a great market for fraudsters to chase because of the ability to ‘Theranos’ your product” with semifunctional prototypes and claims of AI being used in products that are really just powered by humans. He argues that many of the new AI companies “are complete frauds or will never have scalable revenue models,” adding, “There is going to be a staggering amount of capex burned on AI that goes nowhere.”
For example, CoreWeave, a cloud service start-up, recently announced it is taking out a $2.3 billion collateralized loan backed by Nvidia chips. CoreWeave’s biggest investor is Nvidia, and participants in the debt offering include Magnetar Capital, Blackstone, Coatue, DigitalBridge Credit, BlackRock, PIMCO, and Carlyle Group.
“Nvidia’s explosive revenue growth really tells us nothing about the future of AI,” Koppikar explains. “It turns out every scammer in America is trying to buy H100 chips right now so that they can say they own them. In 2021, scam companies put Bitcoin on their balance sheets — now the scams have shifted over to putting $40,000 H100s on the balance sheet.”
Koppikar places Applied Digital, with its big order of Nvidia chips, squarely in that category. The company would seem to be a marginal player in the big-stakes world of AI — except for the fact that two customers it claims to have lined up are among the top-ten AI companies that Bloomberg anointed in late June as “the ones to watch right now.”
When Applied Digital in May announced a potential deal worth $460 million, it declined to name the customer. But later, CEO Cummins tweeted that both that company and the other client were on the Bloomberg top-ten list. His tweet was followed by a research report from brokerage and investment banking firm Craig-Hallum Capital Group, which handled a $125 million equity offering for Applied Digital in late June, noting that one of the company’s new cloud customers is an “LLM provider based in London.”
Short-sellers David and Koppikar point out that only one company is both on the Bloomberg list and headquartered in London, and that is Stability AI. Its product, Stable Diffusion — a model that allows users to create images based on a few word prompts — was initially so exciting to investors that the company was able to raise $100 million last October from prominent VCs including Coatue and Lightspeed Venture Partners. Stability AI quickly garnered a valuation of $1 billion.
But since then, the company’s high-flying status has been tarnished by revelations that founder Emad Mostaque lied about his Oxford University credentials, misled investors, and is burning through cash and losing executives, as outlined in a scathing Forbes article.
David calls Stability AI “one of the most dubious AI start-ups in a field awash with speculative promotions.”
Adds Koppikar: “It looks like Applied Digital signed up a customer that has no way to deliver anything close to the revenues it’s claiming.” He calls the situation “a cautionary tale for AI.”
Stability AI did not respond by press time to Institutional Investor’s request for comment. But elsewhere, Mostaque has denied the accusations leveled against him. He has also predicted that AI will be “the biggest bubble of all time” and admitted that it is still in its early stages and “not quite ready” for mass-scale adoption in most industries, including banking.
On July 14, Applied Digital disclosed its other big AI cloud service customer: Character.ai, a neural language model chatbot application founded by former Google executives that generates humanlike text responses and can participate in “contextual” conversation.
Both Stability AI and Character.ai go a step beyond the text capabilities of ChatGPT, which many businesses believe can help write emails, summarize 10-Ks, draft legal documents, and even, some argue, offer strategic advice. These two programs are among the new generative AI offerings that add images — or sometimes videos — based on word prompts.
With Character.ai, users can interact with virtual avatars of celebrities like Taylor Swift and Elon Musk. The downside of the program is apparent on Reddit, where a subreddit called CharacterAI_NSFW gives explicit instructions on “how to sex the bots.” (NSFW stands for “not safe for work.”)
The subreddit’s advice is straightforward: “You just have to spend time flirting with them and slowly pushing the action forward while the content filter fights you. . . . Once you’ve romanced the bot to the point it actually likes you and is consenting, then you can start sexing it.”
Koppikar offers a note of caution on the video game–based AI programs. “This won’t be anywhere near as big as people think until they find a way to make it not creepy.”
In response, a spokesman for Character.ai cited the company’s terms of service, which state that “pornographic content is against our terms of service, and will not be supported at any point in the future.” Meanwhile, Character.ai has posted statistics showing that users have spent more time there, where they can engage with famous or fictional people, than on ChatGPT.
As of March, the Character.ai start-up was valued at $1 billion, according to PitchBook. Its most prominent backer is Andreessen Horowitz, which once boasted the biggest VC crypto fund. But now that such bets have sputtered, the VC firm has launched a new fund to invest in AI start-ups.
Not surprisingly, VC heavyweight Marc Andreessen, the firm’s co-founder, has become an effusive cheerleader for AI. In a recent lengthy blogpost titled “Why AI Will Save the World,” he argues that “what AI offers us is the opportunity to profoundly augment human intelligence to make all of these outcomes of intelligence — and many others, from the creation of new medicines to ways to solve climate change to technologies to reach the stars — much, much better from here.”
Among the benefits, says Andreessen, will be the ability of every child to have an “AI tutor that is infinitely patient, infinitely compassionate, infinitely knowledgeable, infinitely helpful.” In his vision of the future, every business and political leader, scientist, artist, doctor, and inventor will have an AI assistant to help maximize his or her endeavors.
“In short,” says Andreessen, “anything that people do with their natural intelligence today can be done much better with AI, and we will be able to take on new challenges that have been impossible to tackle without AI, from curing all diseases to achieving interstellar travel.”
Andreessen’s paean may represent peak AI fever. But right now, it’s about as popular to cast aspersions on the miracles of AI as it was two years ago to dismiss crypto as the future of money, or before that, to criticize SPACs as the preferred way to take a company public.
But although generative AI may well unleash untold gains for society, it also seems that its benefits could be less exciting than the hype suggests.
Elsewhere, a Texas judge has banned attorneys from using ChatGPT to create filings without human oversight following an incident in which the program made up court cases.
OpenAi declined to comment on those cases.
Ben Dickson, a software engineer and the founder of the blog TechTalks, says that hallucinations are a “serious problem. LLMs are wont to generate plausible text that is not factually correct, such as made-up names of papers and journals.”
Moreover, there is new research, as well as anecdotal evidence, indicating that ChatGPT’s output has gotten worse, or “drifted,” over time.
For example, research by Stanford University and UC Berkeley professors looked at ChatGPT’s ability to identify prime numbers. Even with something as straightforward as math, the researchers found that in March, ChatGPT had 84 percent accuracy in identifying prime versus composite numbers, but by June had only 51 percent accuracy performing the same exercise.
“The findings are a warning about the risks of building applications on top of black-box AI systems like ChatGPT that could produce inconsistent or unpredictable results over time,” Dickson wrote on his blog. “My rule of thumb is ‘Only trust the model when you can verify,’” he told Institutional Investor in an email.
Outside the halls of academia, users have also noticed problems. As first reported in Business Insider, Peter Yang, a product lead at Roblox, tweeted in May that the model was generating faster outputs but that the quality was worse. When asked on Twitter what questions were the problem, Yang responded, “Just simple questions like making writing more clear and concise and generating ideas.”
OpenAI has denied there is any degradation in GPT-4’s capabilities. “We make each new version smarter than the previous one,” Peter Welinder, vice president of product and partnerships at OpenAI, tweeted in July. “Current hypothesis: When you use it more heavily, you start noticing issues you didn’t see before.” (Welinder declined II’s request for further comment.)
Meanwhile, Reuters recently reported that monthly traffic to ChatGPT’s website declined in June for the first time since its launch in November, according to analytics firm Similarweb.
“ChatGPT trends are ugly: Traffic is down, and interest — measured by Google — is in free fall,” says Koppikar. “This is not the type of trend you see in an exponential growth story at this early stage.” He points out that Facebook, in contrast, showed consistent explosive growth for years.
“AI isn’t the biggest thing since the internet,” Koppikar says. “It is a cute excuse to buy tech stocks again, though.”
More recently, Yang tweeted: “I haven’t seen a ‘100 AI tools you can’t miss’ thread in weeks. Have we peaked?”
Experts in the field agree. “In terms of underlying techniques, ChatGPT is not particularly innovative,” Yann LeCun, Meta’s chief AI scientist, told a group of reporters and business executives on Zoom recently. “It’s nothing revolutionary, although that’s the way it’s perceived in the public,” he said, according to a report on ZDNet, a tech insider newsletter.
How much ChatGPT’s degree of innovation matters is up for debate. If analysts can use ChatGPT to read 200-page documents in a matter of minutes and prepare a report, that can save companies time and money. “But when you see people talking about using these models to make investment decisions, there’s little evidence that I’ve seen that these models can do that,” says Angelo Calvello, founder of Rosetta Analytics, a small quant investment firm. (He is also an opinion columnist for Institutional Investor.) That said, big asset managers might be able to cut their staff 70 percent over time, Calvello argues, because generative AI “could write the IR letters, it could analyze documents. It’s all about efficiency.”
Of course, that is a major reason people are investing in AI’s perceived beneficiaries. Hedge fund manager Dan Loeb is one professional who claims to be ahead of the curve in investing in AI.
“We have watched AI evolve and believe the technology has matured to the point that it is driving a transformational technology platform shift similar to those seen roughly once per decade: the personal computer in the 1980s, internet in the 1990s, mobile in the 2000s, and cloud in the 2010s,” he wrote in a recent letter to his Third Point hedge fund’s investors. “AI is creating interesting investment opportunities in the information technology ‘stack,’ and we have increased our exposure to companies throughout the software and semiconductor value chains that should benefit from mass adoption of large language models, one of the foundational technologies underlying generative AI.”
The Third Point CEO says his hedge fund has been investing in “AI-enabled business models” since 2016. At that time, it made a Series B venture investment in Upstart, a fintech player that claims to use AI in making loans.
Upstart “ultimately became the firm’s most lucrative investment,” Loeb writes in his letter to investors. To be sure, Third Point did make a fortune in Upstart; its 16 percent stake in the company was worth as much as $3.9 billion in the third quarter of 2021 — the peak of the recent tech bubble — as the hedge fund began to unload its shares. (When Third Point invested in Upstart in 2016, the entire company was valued at a mere $200 million, notes PitchBook.) By the end of March 2022, Third Point was completely out and the stock was tanking. Upstart is now worth less than 10 percent of its value at the peak.
Short-sellers scoff at the designation of Upstart as an AI stock and have shorted 33 percent of its outstanding shares. This year, though, that trade has been another loser for the shorts, as Upstart rallied off its bottom during the generative AI boom. But though shares fell by about 50 percent in August after Upstart reported a big quarterly loss, shorts were still out $639 million, or nearly 100 percent, betting against Upstart this year as of August 25, according to S3 Partners.
One of the few AI shorts that has been profitable this year is SoundHound AI. Culper Research’s Christian Lamarco calls the company “a flailing AI wannabe claiming to have revolutionary technology, a growing restaurant business, and a massive backlog of contracts.” The 17-year-old company was brought public via a SPAC in 2022. It is not profitable, and insiders are dumping the stock, according to Culper’s July report. In it, Lamarco writes that SoundHound now “claims its AI technology will revolutionize phone and drive-thru ordering, but we think this foray is an utter failure.”
Culper asserts that SoundHound’s AI simply doesn’t work — it relies on call centers staffed by humans, say former employees that his firm interviewed. Short-sellers had made $8.25 million, an 18 percent gain, by August 25 by taking on SoundHound this year, says S3 Partners.
In response to Culper’s report, a company spokesperson said that “SoundHound’s voice AI platform powers millions of devices (cars, TVs, [internet of things] devices) and responds to billions of queries without human involvement, including via our customer service voice solutions. With its restaurant business, SoundHound is one of the only companies where humans are neither monitoring nor interfering in the ordering process. This includes our recently announced partnership with White Castle, where our AI-only drive-thru solution is scaling to 100 locations next year.”
By far the biggest company in the AI space that short-sellers are targeting is C3.ai — which has been more resilient to their attacks. Two-short sellers, Sahm Adrangi’s Kerrisdale Capital and Ben Axler’s Spruce Point Capital, have taken aim at C3.ai, noting that it is another pivot to AI and — like the others — has never been profitable in several years of existence.
“The company was originally founded as C3 Energy to develop analytics solutions for public utilities preparing for the emergence of cap-and-trade and smart grids,” Adrangi wrote in a recent report on the company. “But management’s master stroke was rebranding operations as C3.ai in 2019 and going public with the ‘AI’ stock ticker, thus securing its place as the default artificial intelligence stock play for the undiscriminating investor despite the bulk of its business coming from relatively dated analytics models built for a very small number of utility, energy, and government customers.”
As Adrangi tells II, “I don’t think you can claim to be an AI business just by calling yourself AI.”
C3.ai, helmed by software entrepreneur and Silicon Valley luminary Tom Siebel, “does have a real business. It’s just not worth where it’s currently trading at,” Adrangi explains. “In the most recent 12 months, it basically nearly burned as much cash as it had revenue.”
Axler says investors are “clearly making a bet” on Siebel, who sold his previous company, Siebel Systems, to Oracle for $5.85 billion in 2005. For his part, Siebel has lashed out at the short-sellers. In a Bloomberg interview, he called them “scumbags who should be in jail.”
And in a response posted on its website, C3.ai calls the Kerrisdale report “a highly creative and transparent attempt by a self-acclaimed short-seller to short the stock, publish an inflammatory letter to move the stock price downward, then cover the short and pocket the profits.” It also says Kerrisdale erred on two specific accounting criticisms in the report — gross margin and unbilled receivables.
The stock recovered some of the losses it incurred after the latest Kerrisdale report following a better-than-expected earnings report. But that has not mollified the short-sellers.
“We’re probably in the early-innings stage of where we are in the AI cycle, and it still remains to be seen who are going to be the winners and losers,” says Axler, who first began shorting C3.ai in February 2022. “But ultimately, we don’t see C3.ai as being a big winner.”
It has been a tough slog for the short-sellers so far. As of August 25, C3.ai short sales were 34 percent of the float, and this year, short-sellers had lost almost $380 million, a 51 percent decline, betting against Siebel, according to S3 Partners. The stock is up more than 150 percent this year, although it’s down almost 40 percent from its peak in June.
Applied Digital is also down about 40 percent from its high point — which came in July, when the company announced the Character.ai deal and shortly after insiders had registered plans to sell their shares. Short-seller reports by David and Koppikar also dinged the stock. Then, on August 14, investors sued the company in a Dallas federal court, repeating short-sellers’ claims that Allied Digital had misled investors about its business model and the independence of its board members, based on close ties to B. Riley, its underwriter and Applied Digital CEO Cummins’ employer. (Applied Digital has denied the allegations.)
Despite the recent setbacks for some of these stocks, AI is still driving the markets higher. Notably, last week Nvidia reported record quarterly sales of $16 billion, far ahead of market expectations. How long the momentum can last is another question.
Cautions Adrangi: “With all these overvalued, overhyped stocks, it’s hard to call when the bubble will burst. But one day it does.”
