Technology + Businessweek
These Are the 10 AI Startups to Watch in 2024

OpenAI introduced the world to what AI could be with the introduction of the ChatGPT chatbot in 2022. Since then the company has continued to push the boundaries of the technology. It’s developed software that can generate impressively realistic-looking videos, and its AI can respond almost in real time to a user’s question with a human-sounding voice. (These products have yet to be rolled out to general audiences.) OpenAI has also provided one of the most engrossing tech dramas in modern history: firing Chief Executive Officer Sam Altman in November, only to bring him back five days later, a series of events that unnerved some AI safety advocates.

Founded by former staffers at OpenAI, Anthropic is its chief rival in the world of large language models, the technology that powers this era’s most dazzling AI achievements. The startup has closed in on OpenAI both in terms of performance and its massive haul of funding from the likes of Alphabet Inc.’s Google and Amazon.com Inc. Anthropic says its most recent model outperforms OpenAI’s in key evaluations in coding and text-based reasoning. With a focus on making tools for businesses, it’s expanded its enterprise business to include Bridgewater Associates, Salesforce and Pfizer, among other customers. Led by an unusual brother-and-sister team, with Daniela Amodei as president, Anthropic has been vocal about its emphasis on safety and has recently published research containing insights into the still mysterious ways generative AI actually works.

Give Suno a command—for example, “Sing me a reggae ballad about anchovies on pizza”—and its software will deliver, churning out a surprisingly complex, human-sounding song in seconds, complete with vocals and rhyming lyrics. Suno is a leader among new AI startups building tools to automate the music-making process. Its technological innovation has made it popular with users, and, predictably, the company has drawn the wrath of the music industry. Generative AI must be trained on massive troves of data, which in this case means Suno has ingested a lot of songs. In June the world’s biggest record labels sued Suno and competitor Udio, alleging that their training data included a large number of copyrighted sound recordings, kicking off what could be a landmark case in the AI industry.

Founded in 2022, Perplexity has speed-run the journey from tiny upstart to one of the most closely watched startups in Silicon Valley, stirring up fierce controversy along the way. Hailed as a Google-disruptor, the company offers an AI-powered search engine and is in late-stage talks for new funding that would value it at $3 billion, making the startup the poster child for AI tools that rethink internet services as basic as search. Perplexity’s investors include big names in tech such as Jeff Bezos, SoftBank and Nvidia. With products that deliver conversational search results and news summaries, the company has also panicked the publishing world, garnering accusations of plagiarism and stoking fears that writers’ work could be subsumed into traffic-hogging AI products.

The splashiest AI companies with the most famous models generally hail from the US or China: Mistral is an exception. The Parisian startup, formed last year by French alums of Google DeepMind and Meta Platforms Inc., has released a flurry of popular large language models, primarily as open-source technology, along with developer tools, a chatbot (Le Chat) and an AI programming product (Codestral). It’s all part of a pitch Mistral is making as the independent alternative to Silicon Valley—though the startup has also inked deals with US tech giants including Microsoft Corp. and IBM Corp. in a bid to get more traction across the Atlantic.

Elon Musk’s new startup raised several billion dollars earlier this year on the back of his pledge to build an anti-woke chatbot “with a rebellious streak.” Grok, as the bot is called, is trained on posts from X, the site formerly known as Twitter, and is currently available to the network’s paying subscribers. “We want to try to be the funniest AI,” Musk said at a conference earlier this year. “If we are gonna die, at least we should die laughing.” It’s personal for Musk, a former co-founder of OpenAI whose acrimonious falling out with the startup included a lawsuit, which he later dropped. Next, xAI is using some of its multibillion-dollar war chest to develop a supercomputer facility in Memphis.

An AI system’s output is only as good as the data it’s ingesting. Scale’s platform cleans up troves of data using a combination of software and human workers to sift through and label the images, text and audio that underpin large language models. The startup counts AI giants such as OpenAI, Google and Meta as customers, in addition to the US Department of Defense. That seemingly insatiable demand for data has made Scale one of the world’s most highly valued AI startups.

Cohere doesn’t have a cool consumer chatbot that can discuss the weather or write a poem. The company develops large language models almost exclusively for its corporate customers, which include Salesforce Inc. and Accenture. CEO Aidan Gomez is an AI celebrity, one of the former Google engineers who co-authored “Attention Is All You Need,” a 2017 seminal AI paper, helping usher in the current age of AI. Cohere’s investors include major names such as Nvidia Corp. and Oracle Corp. The company’s latest model, Command R+, is cheaper than OpenAI’s GPT-4 Turbo, and though it doesn’t perform as well in terms of raw capabilities, Cohere says it’s better for certain business-focused tasks.

No one—not OpenAI, not Anthropic, not Google—has enough graphics processing units for all the AI work they want to do. Enter CoreWeave, a provider of cloud computing horsepower, offering chips for hire to a new generation of companies desperate for capacity. According to reports, even Microsoft inked a deal with CoreWeave to access extra GPUs. CoreWeave more than doubled its valuation in April, vaulting it into the ranks of the US’s most valuable startups as it raked in billions of dollars in debt and equity. Formerly focused on Bitcoin mining, the company has made a lucrative transition.

In a critical year for elections, ElevenLabs’ technology is freaking people out. The company is at the forefront of voice-cloning AI, with software that can replicate a person’s voice, turn text into human-sounding utterances and translate speech from one language to another. The technology is used by AI video startups such as HeyGen and Captions as well as individuals. Virginia lawmaker Jennifer Wexton, for instance, created a copy of her voice to help her communicate after a neurological disease affected her ability to speak. Not all use cases are altruistic. Early this year, ElevenLabs banned a user for creating faked audio of President Joe Biden urging people not to vote in the New Hampshire primary.