After 2 Years of GenAI, the Industry's Explosive Growth Has Some Unexpected Fallout

Artificial intelligence entered its adolescence last year, as evidenced by one hell of a growth spurt. Investors piled on in 2023 as companies across industries looked to hop on the AI train. In the U.S., investment in the sector grew to $67.2 billion, with a third going directly to makers of generative AI products, the technology popularized by OpenAI's ChatGPT. According to Stanford University's 2024 AI Index Report, genAI investment jumped to more than $20 billion, up from the $2.21 billion invested in 2022. For AI entrepreneurs, the enthusiasm is double-edged. Interest in their tools has never been greater, as nearly half of the Inc. 5000 honorees who took our CEO Survey cite the use of at least one AI service. OpenAI was the top provider. But genAI hype has also led to misconceptions about what these tools actually do. As AI zips to the top of investors' port­folios, founders say the biggest factor limiting their growth isn't fundraising; it's overcoming a towering knowledge gap. Benjamin Plummer understands this implicitly. He's the CEO of Invisible Technologies (No. 152), a San Francisco-based software and data services provider. Invisible helps clients such as Microsoft and Cohere create high-quality data to train their AI models. To do so, Invisible uses a roster of more than 5,000 contractors, all experts in their fields, who help fine-tune or stress test different models. "You might have a health care company training a chatbot that needs 100 doctors to test and evaluate the model," explains Plummer, 38. By creating complex workflows with multiple experts testing and grading models, richer training data is collected. Plummer says helping organizations like OpenAI--a client since before the launch of ChatGPT--train models has been a "huge part of our growth over the past year." Brandon Tseng, 38, has a different method for improving his AI capabilities--acquisition. The co-founder of San Diego-based defense technology maker Shield AI (No. 634) has been in the business of enabling drones and aircraft for AI operation since 2015, but a pivotal acquisition boosted his company's ascent. In 2020, the defense agency Darpa held a series of simulated dogfights between different AI models that had been trained to operate F-16 fighter jets. Aviation startup Heron Systems bested the slate, as well as an experienced human F-16 pilot. Less than a year later, Shield AI acquired Heron and used the tech to update Hivemind, its software framework for piloting drones and fighter jets. The service, which is in use by the U.S. Air Force and the Coast Guard, among others, is today a digital maverick, more than ready to take risks that most ­human pilots wouldn't dare. "It's not afraid to die," says Tseng. Shield AI has raised over $1 bil­lion, but Tseng still struggles to explain to potential investors and customers that he doesn't offer a language model like ChatGPT. For a certain set of investors, says Tseng, AI is just chatbots and generative art. He encourages them to think bigger. By pigeonholing a tool capable of "perceiving, thinking, and acting" as just a chatbot, he says, investors are "missing the forest for the trees." Even companies that use generative AI have met hurdles explaining what they do. While Eric Yang's Dallas-­based image and video enhancement company, Topaz Labs (No. 2,697), uses genAI technology, the result is a far cry from that of text-to-image tools like OpenAI's Dall-E. Topaz's software enhances the quality of digital images, sharpening blurry shots and removing grain. In 2018, Topaz introduced Gigapixel, an app that uses AI to change the dimensions of a digital image without losing detail. Gigapixel uses generative adversarial networks (GANs), which consist of two AI models, a generator and a discriminator. The generator creates fake data, which is compared by the discriminator with the original training data. You can think of the discriminator as a bouncer at a nightclub, letting in only clubbers with the right look. As this process goes on, the generator learns how to make data that the discriminator will perceive as real. Topaz uses GANs to generate believable textures and detail in images. Beyond that, says Yang, 36, Topaz's products are all about fidelity to the original image. "People assume that because we're an AI company, we're seeking to replace the need for photographers. But the really useful tools will be the ones that give people superpowers." Even with superpowers, business leaders should be ready to recalibrate on a moment's notice, suggests Daniel Berlind, co-founder of Snappt (No. 41), an AI-powered fraud-detection platform. Founded in Los Angeles in 2016, Snappt provides landlords with an AI model trained to sniff out signs of tenant fraud by looking at pay stubs and bank statements. Last year, Berlind, 36, learned that enterprising criminals had developed an entirely new method of fraud to get around Snappt whereby scammers would register LLCs and use free trials from payroll providers like ADP and Gusto to create legitimate-looking pay stubs. Snappt's forensics team proceeded to train its AI to counteract the scheme. It's clear from Snappt's experience that success in the world of AI requires constant adjustments. It also requires a willingness to create demand. Just ask John Dean, 26, co-founder and CEO of WindBorne Systems (No. 585). In 2019, Dean, then an undergrad at Stanford, co-founded the company alongside classmates Andrey Sushko and Kai Marshland. They developed long-duration weather balloons that can fly above the open ocean, where traditional balloons can't, to collect temperature and atmospheric-pressure readings. This data is useful for world governments, which spend approximately $10 billion annually on weather observation. WindBorne's growth is slated to kick up with this past February's launch of WeatherMesh, an AI-forecasting model that's set accuracy records. Today, the company, based in Palo Alto, California, credits its Weather­Mesh model with expanding its clientele beyond the public sector. "Every big tech company working on AI-based weather modeling has reached out to us about purchasing our datasets to improve their models," claims Dean, pointing out that when AI enters an industry, the data ­becomes much more valuable. The gold rush these execs have witnessed has shown no signs of slowing down in 2024, with rivals OpenAI and Anthropic one-upping each other with faster and less- expensive technology. But matur­ity comes for us all. And if 2025 features a little less hype and a little more understanding of how this incredible technology works, it might not be such a bad thing. You're Not Hallucinating Any conversation about AI in the past year has included OpenAI, the Sam Altman-run startup behind ChatGPT. So why isn't the company on the Inc. 5000? We spoke with OpenAI several times, but executives declined to verify its revenue, which has been reported by The Information to be running at an annualized rate of $3.4 billion. Considering the company transitioned to a capped-profit structure in 2019, it is likely Open­AI would have placed at or near the top of the list, if that figure is accurate. Maybe next year, Sam.