Big Tech's AI Dominance: Challenges for Startups and Regulatory Concerns

Big Tech is busy wrapping up AI talent to dominate the market

Big Tech is following a crafty playbook to hoover up AI talent: Instead of buying the hottest AI startups, the giants hire their leadership and license their intellectual property, essentially sucking the life out of them. So long as they leave behind the shell of a company, antitrust scrutiny can be avoided. Alphabet's Google [which just had bad news in the form of a court ruling against it for its search market monopoly], is the latest to follow this template by hiring the co-founders of Character.ai, Noam Shazeer and Daniel De Freitas, along with a few other employees. Most of the startup's staff will remain with the smaller company while its general counsel will become CEO. Why a lawyer? Because Google is entering a licensing agreement to use Character.AI's technology -- in addition to buying out the startup's investors, principally Andreessen Horowitz, who'd put in more $150 million. Microsoft last year hired the founding team behind Inflection, a high-flying startup that was also creating an AI companion. Amazon.com then hired the team at Adept, a San Francisco-based firm co-founded by former OpenAI and Google researchers. It's a sleight of hand that reaps big benefits for tech giants, who can afford the vast computational power and data needed to build Generative AI, but struggle to attract the best talent. Now they can do the latter too. Shazeer is a big hire for Google, which has grappled with glitches in its flagship AI model Gemini that could hinder its efforts to catch up to Microsoft and Amazon in the cloud wars. So venerated is Shazeer that Google may be paying him in the tens of millions of dollars (or more) to rejoin the company. He was one of the lead inventors of the Transformer, a powerful blueprint for AI systems that formed the basis of the recent boom. Back in 2017, he was perhaps the most respected Google scientist named on a landmark research paper detailing how the Transformer could exploit powerful AI chips to quickly process and generate information. Google was slow to capitalize on the technology, but OpenAI made the Transformer the basis of ChatGPT (it's what 'T' stands for). All eight authors of the research paper left Google, mostly to start companies that raised billions of dollars thanks to their pedigree, including two who co-founded Adept. When Shazeer co-founded Character.ai, he was continuing work that Google had shut down, to build lifelike chatbots that humans could use as companions. Character.ai went on to become one of the most popular AI apps outside of ChatGPT, particularly with teenagers. The app lets you chat with a bot specially-trained to emulate a celebrity, fictional character or historic figure, and several teens have told me they spend hours a day on the app, role-playing or chatting with an artificial confidante. Shazeer's goal was to create an AI model that could remember everything about you. "A person has probably heard or read hundreds of billions of words in their lifetime, so that's about the scale of data that you need -- a quarter of a gigabyte -- to have a lot of context on a person's life," Shazeer told me in January. "This is not beyond the realm of feasibility as we improve our algorithms... The model would know billions of things about you when it's talking to you." However intriguing that vision is, Shazeer will now be bringing it back to his old bosses at Google, where he'll likely have more sway than he did before, perhaps by working directly with Google DeepMind CEO Demis Hassabis. The question is whether the likes of Meta or Elon Musk's X.ai will follow a similar acqui-hiring strategy. Both reportedly looked into picking up Character.ai, before Shazeer went to Google, but there are other AI companies in the market that are likely grappling with the costs of building their technology amid unclear revenue prospects. Cohere, a Canadian AI start-up founded by another author of the Transformer paper, is likely in play, along with Perplexity, a San Francisco startup that is competing with Google on AI search. Even the founders of Anthropic and OpenAI, the two AI startups that have raised $8.8 billion and $11.3 billion respectively, must surely be glancing at the exit doors. OpenAI could lose $5 billion this year, according to an analysis by The Information, which looked at the company's internal financial data, meaning it will need to raise more cash one way or another. The US Federal Trade Commission is scrutinizing what Microsoft and Amazon are doing, but there's no indication they will turn into formal probes or hit deals. More likely: We'll see tech giants gobble up more leaders of the GenAI industry, consolidating power as they ride out the current market rout right under the noses of regulators. Perhaps this is what AI startups had expected. ©bloomberg

As regulators close in, Nvidia scrambles for a response

Nvidia's dominance in AI chip sales, with a 90% market share, has led to inquiries from the EU, Britain, China, and the U.S. Justice Department. The company is now focusing on building its Washington presence and legal defenses to address antitrust concerns, while managing high demand for its GPUs and new acquisitions.Nvidia rocketed to the top of the tech industry by providing the computer chips essential to building artificial intelligence. By the end of last year, it had more than a 90% share of those chips sold around the world. That success has quickly brought government scrutiny. Authorities with the European Union, Britain and China asked the company for information about its sales of those important chips, allocation of supplies and investments in other companies, according to Nvidia's financial filings. The Justice Department has also started investigating Nvidia's sales practices and will review one of the company's most recent acquisitions, said three people familiar with the inquiries, who asked for anonymity because it is early in the process. Nvidia wasn't ready for the attention, and is now racing to build the staffs and offices needed to respond. Just last year, Nvidia started searching for an office in Washington and hired four public policy employees. This year, it added its first in-house competition attorney to work alongside a legal team that has been addressing antitrust questions over the past decade. And it has begun developing a strategy to respond to government interest. Nvidia's scramble speaks to the narrow window it has to head off the kind of regulatory attention that has hamstrung other tech giants. Google, Apple, Amazon and Meta had at least a decade to build up sophisticated Washington operations before antitrust lawsuits, which have cast a shadow over their businesses. On Monday, a federal judge issued a landmark ruling that Google had violated antitrust law by abusing a monopoly over internet search. For Nvidia, the business stakes are enormous. Over the past two years, its quarterly profits have risen ninefold to $14.88 billion. Investors have responded by making it the world's third-most-valuable technology company after Apple and Microsoft, with a value of about $2.5 trillion. Nvidia's speedy ascent has given it little margin for error. Its stock fell more than 6% Monday after The Information, a tech news site, reported that Nvidia would ship its newest AI chip three months later than planned, and there is mounting skepticism about AI's potential to transform businesses. Nvidia dominates sales of chips known as graphics processing units, or GPUs, which make it possible to create AI systems in data centers. Customers want more of those chips than Nvidia can produce. As a result, antitrust authorities are concerned about Nvidia's power to determine how a scarce but essential technology is being allocated. "Regulators want to find out if Nvidia is putting their thumb on the scale of the deal or if the product is so good that it sells itself," said Daniel Newman, CEO of the Futurum Group, a tech research firm that pegged Nvidia's GPU market share at 92%. "There's no evidence they're doing anything monopolistic or anticompetitive, but the conditions are right because of their market leadership." In the eyes of many customers, Nvidia, a 31-year-old Silicon Valley company, has earned its success. The company bet its future on using GPUs to train the neural networks that power the large language models that are behind popular chatbots like OpenAI's ChatGPT. Now, it is expanding its business to include the sale of connected supercomputers and a cloud computing service. Nvidia plans to counter mounting antitrust concerns by responding to government requests on a case-by-case basis. The company argues that it is conducting its business fairly and in accordance with competition law. Tim Teter, Nvidia's general counsel, is responsible for making Nvidia's case to investigators. He joined the company in 2017 after more than two decades at the law firm Cooley, where he worked on commercial litigation, patent and technology cases. He'll work closely with Jensen Huang, Nvidia's CEO. "Regulators need not be concerned," said Ken Brown, a company spokesperson. "But we're happy to provide any information regulators need." This year, in the wake of complaints about Nvidia's chokehold on the AI market, Washington's concerns have shifted from China to competition, with everyone from startup founders to Elon Musk grumbling about the company's influence. Those complaints have spiked as Nvidia has increased its ownership of critical pieces of the AI supply chain. It offers cables and switches that control high-performance computers, software that manages data center performance and a computer language, CUDA, that developers use to control Nvidia's chips. In 2020, it spent $7 billion to buy Mellanox, the maker of high-speed cables and switches that transmit the data between servers to create AI systems. Two years later, it introduced its own central processing unit, or CPU, to deliver better GPU performance. By bundling its products for customers like cloud computing providers and server manufacturers, Nvidia enjoys more pricing flexibility and power than rivals, four people familiar with its business said. Nvidia's dominance in AI has complicated its plans to acquire other companies that could complement its business. When it offered to buy Arm Holdings, which licenses the designs used to create chips, for $40 billion in 2020, competitors and customers complained that the move would give Nvidia too much power over the industry. Nvidia abandoned the deal in 2022 after the Federal Trade Commission sued to block it. That was more than a year before Nvidia began building out its Washington presence. At the semiconductor industry's annual gala in the fall of 2021, Huang joked that he had been outmaneuvered on the acquisition by Cristiano Amon, CEO of Qualcomm, a chipmaker with a more established legal and public affairs team. "I connected some dots tonight," Huang said. "I was trying to figure out how is it possible that Cristiano knew every single regulator on the planet, that by the time I got there to tell them about my story on Arm, he was already there advocating against it?" Today, Nvidia is facing another challenge to its acquisition plans. It has offered $700 million for Run.ai, a startup that controls the scheduling of AI systems. But the Justice Department plans to review the deal alongside its inquiry into Nvidia's business practices, a person with knowledge of the plans said. The review and details of the broader Justice Department investigation were reported earlier by The Information and Politico. "Nvidia has been welcomed with open arms to the regulatory party," Newman of the Futurum Group said. "The scrutiny will be substantial, frequent and endless." This article originally appeared in The New York Times.