AI remains a hot topic, with the H100 chip not only being coveted by major companies but also potentially becoming a speculative tool for some businesses.
Last week, the media outlet Institutional Investor mentioned in a column that many startup AI companies are complete "scams," lacking scalable revenue models and burning through astonishing capital expenditures on AI with no results.
Short-selling firm Koppikar believes that in the AI wave, large tech companies are driving the development of AI. They fund startups, encouraging them to buy their products, creating a cash cycle. Nvidia's explosive revenue growth doesn't necessarily provide insights into the future of AI.
Now, US "scam" companies are all vying for the H100 chip, creating an impression that they will dominate the AI sector. In 2021, these "scam" companies added Bitcoin to their balance sheets; now, they're adding the H100.
This phenomenon is termed "Grift Shift," where companies move away from losing bets on cryptocurrencies and tech stocks to cash in on the AI era.
AI short-sellers have smelled blood and are targeting companies like SoundHoundAI, C3.ai, and Applied Sciences. But so far, the short-sellers haven't had it easy.
Dressed in AI's Cloak
In April 2022, a penny stock company named Applied Sciences went public on NASDAQ. It rebranded itself as a cloud hosting service provider for Bitcoin miners and changed its name to Applied Blockchain, hoping to ride the Bitcoin wave.
Unfortunately, luck wasn't on their side. At that time, Bitcoin was in a "crash" mode, with its price plummeting from over $60,000 to less than $20,000. Applied Blockchain's dream of hitching a ride shattered, with its stock soaring dream far from realization.
However, the company's management didn't lose hope. Within a few months, the company was renamed "Applied Digital," transforming into an "AI concept stock."
In May, Applied Digital's CEO, Wes Cummins, announced that the company had signed a cloud hosting agreement worth $180 million with a renowned AI company.
Subsequently, they signed another partnership agreement with an AI giant worth up to $460 million, causing the company's stock price to skyrocket.
By July, the company's stock price had surged by nearly 450%, becoming one of the biggest "winners" in the AI wave.
During this period, CEO Wes Cummins, who at one point held nearly 25% of the company's shares, boasted on Twitter (now known as X) that the company had ordered 26,000 H100 GPUs from Nvidia at a price of $40,000 each, aiming to reach new heights in the AI sector.
But is this really the case?
At least the short-sellers don't believe so-they've smelled blood.
In a report, short-seller Dan David analyzed that such a massive purchase would make Applied Digital a leader in high-performance computing, on par with Google, Meta, and Amazon. However, the cost of purchasing these devices would exceed $1 billion, far surpassing Applied Digital's market value of $600 million. In the end, David couldn't help but vent his frustration:
After the emergence of ChatGPT, people's interest in AI skyrocketed, attracting the worst salespeople and unscrupulous individuals to sell fake AI products to gullible investors, with Applied Digital being one of them.
Applied Digital is one of nearly ten companies short-sellers have targeted this year, seen as dubious beneficiaries of AI frenzy. As of August 25, short-sellers have bet against 19% of Applied Digital's circulating shares.
Nate Koppikar, co-founder of Orso Partners and also a short-seller of Applied Digital, termed this phenomenon as "Grift Shift." He believes companies are moving away from losing bets on cryptocurrencies and tech stocks, hoping to cash in on the AI era.
H100-A Chip for Speculation
In the AI era, Nvidia's H100 chip has become a hot commodity, thanks to its outstanding performance. Major tech companies are scrambling for it, with orders lined up for the next year, making it a tool for speculative companies.
According to the "Q2 AI and Machine Learning Report" released by PitchBook, generative AI and machine learning startups have raised about $39.4 billion this year, with $19.4 billion raised in the second quarter alone.
Despite funds continuously flowing into all AI sectors, some critics have started to wonder if the latest technology is a real revolution or just progress. Meanwhile, issues like "fabricating facts" and "degradation" in large language models have started appearing in tech literature, referring to various errors made by generative AI. Studies indicate that the output quality of these products seems to be deteriorating over time.
Koppikar points out:
It seems like a huge market for fraudsters, who apply half-functional prototypes and AI to products that are actually driven by humans.
Many startup AI companies are complete scams, never having a scalable revenue model. The capital expenditures they burn on AI will reach astonishing numbers, but with no results.
Last year, Koppikar was one of the earliest investors to predict a decline in tech stocks and growth stocks. Part of his prediction was based on the interdependence between these companies. He believes this interdependence has played an even more extreme role in the AI wave:
Many highly regarded AI startups have received support from big companies like Nvidia, and Nvidia needs these startups to buy its chips. It's a back-and-forth action.
Large tech companies are driving this process. They fund startups, encouraging them to buy their products, creating a cash cycle.
For instance, cloud service startup CoreWeave recently announced that it would receive a $2.3 billion collateral loan supported by Nvidia.
Looking further, Nvidia's explosive revenue growth doesn't necessarily provide insights into the future of AI, Koppikar explains:
US "scam" companies are all vying for the H100 chip, creating an impression that they will dominate the AI sector. In 2021, these "scam" companies added Bitcoin to their balance sheets; now, they're adding the H100.
Koppikar categorizes Applied Digital, which has large orders for Nvidia chips, in this group. Koppikar adds that the clients Applied Digital has signed up can't achieve the revenue they claim, calling it an "AI cautionary tale."
"AI is Just an Excuse to Buy Tech Stocks"
It's undeniable that generative AI has significant flaws, one of which is fabrication.
Software engineer and founder of the blog TechTalks, Ben Dickson, says that fabricating facts is a serious issue. LLM often generates text that seems plausible but is inconsistent with facts, such as fabricating the names of papers and journals.
Meanwhile, new research and evidence suggest that the quality of ChatGPT's outputs is deteriorating over time.
For example, professors from Stanford University and the University of California, Berkeley, studied ChatGPT's ability to identify prime numbers. Researchers found that in March, ChatGPT had an accuracy rate of 84% in identifying prime and composite numbers, but by June, the accuracy rate dropped to 51%.
Outside academia, users have also noticed some issues. Roblox's product manager, Alex Hicks, said that ChatGPT's output quality has been declining, and the model is more likely to generate incorrect or nonsensical answers.
However, these issues haven't stopped investors from pouring money into AI startups. According to PitchBook, generative AI and machine learning startups have raised about $39.4 billion this year, with $19.4 billion raised in the second quarter alone.
The question remains: Is the latest technology a real revolution or just progress?
