Rene Haas was on an exercise bike two weeks ago watching CNBC like he does every morning, when the tech sector took a beating. The TV screen had “red everywhere”, recalls the chief executive of Arm Holdings. “Everything is crashing and I’m thinking to myself: seriously? Really? What are people thinking?”
The market freak-out had been triggered by the release of a new model from DeepSeek, a Chinese-owned artificial intelligence start-up that has developed a large language model capable of results comparable to those of OpenAI’s ChatGPT — for what it claimed was a fraction of their cost. As investors took fright that day, Nvidia, the dominant maker of chips that power AI applications, lost almost $600bn of market value. Shares in Arm, which Haas has run since 2022, fell about 10 per cent, equivalent to losing about $17bn (they have since recovered).
Arm designs and licenses the essential architecture in almost all smartphones and increasingly works with chipmakers such as Nvidia, so is likely to be affected by any anxiety about AI’s trajectory. Does Haas agree with the venture capitalist Marc Andreessen, who hailed DeepSeek as AI’s “Sputnik moment”? “No,” he says firmly. “This is moving so fast, by the time you write this article, there could be something different.”
But he acknowledges DeepSeek was a surprise. “Number one, an open-source model has caught up with, in theory, some of the best closed-source reasoning tools.” DeepSeek published the research behind its model and made some of its workings publicly available; models such as ChatGPT are closed source, using proprietary code. The fact that the model originated in China is also “a big deal”, he says, given advancements in AI have, to date, been led by Silicon Valley. Haas is less impressed with claims DeepSeek was developed for a bargain $5.6mn, a tiny fraction of the amount used in the training of US models. He says he does not believe “the rumours” that “they did this on a shoestring budget . . . I think that’s where people just over-indexed on ‘maybe the world’s coming to an end’.”
Anyone worried about the future of AI should look at who is investing money, he says. “The canary in the coal mine to look at is when [tech bosses] Satya Nadella or Sundar [Pichai] or [Mark] Zuckerberg say, ‘You know that $80bn of capex I said I was going to do? I think I’m going to cut that by two-thirds.’ That’s what you need to look for.”
Haas also doubts DeepSeek’s approach was particularly revolutionary, saying he believes the company used a process called “distillation”, whereby it learns from other AI models. OpenAI agrees: days after the launch of DeepSeek’s performance data, the US company said it had evidence that its Chinese rival had used ChatGPT’s proprietary data to train its models.
He does not predict a rosy future for DeepSeek, saying he thinks it will “get shut down”. Washington is “scrambling on what to do with this thing. Think about it . . . if you’re not going to allow a TikTok, why would you allow this?” This is only his opinion, he adds. “I’m not operating on any knowledge [here].”
Arm occupies a different but related part of the tech ecosystem. Its roots stretch back to the BBC Micro, a computer that was a fixture of most 1980s UK school classrooms and which featured the first Arm processor. Arm’s founders launched their own company from an old turkey barn in Cambridgeshire, licensing their chip design to Apple for its now defunct Newton handheld device, and then to the early generation of mobile phones. When Apple fired the gun on the smartphone revolution with the iPhone, it turned to Arm. Since the UK-based company’s launch in 1990, close to 300bn devices using its chip designs have been shipped.
Haas is a tall man, and taller today in nifty Cuban heels. This month is his three-year anniversary as Arm CEO but he first arrived at the company in 2013 after a seven-year stint at Nvidia — where he worked closely with chief executive Jensen Huang — and a spell starting several companies. He began his career at Texas Instruments but has his father to thank for introducing him to computing. The elder Haas fled Nazi Germany with some of his Jewish family for Portugal as a child in the early 1930s. It is “the classic American immigrant story”, Haas says. His father met his mother in Portugal and the couple moved to America, eventually settling in upstate New York, where Haas senior worked for the research arm of Xerox.
The division had a sister research group on the west coast: the Palo Alto Research Center which, according to Silicon Valley legend, is where a young Steve Jobs first saw a graphical user interface in 1979. Haas had visited the facility as a young boy with his family a few years earlier and was blown away by what he saw. “It was like walking into a science-fiction movie. Computers, games, communicating with others . . . and this is 50 years ago.”
In 2006, Haas landed at Nvidia, which at the time was generating about $4bn in revenue and had a market capitalisation of about $10bn (these days, even after the DeepSeek turmoil, it is worth $3tn).
He developed a close relationship with Huang (who recently appeared on a podcast hosted by Haas) and says the company at the time was “scrappy”, working in the shadow of then market leader Intel. They would eventually swap places as Nvidia’s growth exploded. “Intel had a board mentality, meaning it moved at a slow pace relative to making decisions.” At Nvidia, “one of their superpowers is they are able to pivot and change strategy and direction. More importantly, Jensen is able to do that and the company mobilises very quickly behind him.”
When Haas jumped to Arm, it was jointly listed in London and on the Nasdaq and three years later was acquired by SoftBank for $32bn. SoftBank’s chief executive, Masayoshi Son, wanted to split the company in two, with one half focused on the “internet of things” and services and the other on its classic chip design business, which Haas was asked to run.
“Masa had a general view that [we] had a really high market share, but didn’t really extract a lot of value for that share.”
Son, though, “got distracted with all this other stuff. He did Vision Fund. He bought WeWork. He was trying to get the T-Mobile Sprint deal over the line. He had a million things going on.”
This turned out fortuitously, allowing Haas and his team to try out new strategies. He decided to flip Arm’s business model, tying higher royalties to the devices that used its chip architecture, rather than focusing on upfront licence fees. Arm was charging roughly the same royalty on a chip that appeared in a blender as one that went into a high-end data centre, which he says was “crazy”. He reorganised the company along vertical lines, creating a business for servers, and a business for cars, for example. Arm’s designs had to be priced “commensurate with the value”.
But bringing new products to market takes time and, although Haas made the sales model more favourable to Arm, the results were not immediate. “It wasn’t really growing but I knew it was going to.”
With little growth on the immediate horizon, Son (who was in the Rotunda with the other tech bros for Donald Trump’s inauguration last month) decided to sell the company.
The only bidder willing to pay the price he wanted was Huang at Nvidia, although his offer was ultimately unsuccessful and blocked by regulators on antitrust grounds. Haas says the regulators “got it right. Absolutely. Arm’s broad market share in the hands of one of its customers would have put a very, very significant disadvantage against the others.”
Events since suggest not selling to Nvidia was the right move. SoftBank ultimately decided to relist Arm on Nasdaq, rejecting the overtures of the UK government for a listing in London. Less than two years later and not yet three years since Haas was made chief executive, its value has skyrocketed to a market capitalisation of about $175bn.
Market fervour for AI may have helped but Haas’s business model switch certainly transformed Arm. The question now is where the next boost will come from. There have been reports that Arm will start building its own chip, a move that would be a radical departure from its royalty and licensing-based business model.
I press Haas on when this might happen but he does not want to say more.
More immediately, Arm is, via SoftBank, part of Stargate, the $500bn AI infrastructure investment project unveiled by Donald Trump on his second day in office. OpenAI is also part of the consortium and, separately, is working with Arm on providing the platform for a new generation of AI “agents” to improve workplace productivity.
There may be bumps in the road but the AI revolution is real and will change the world, says Haas, pointing to the first dotcom bubble and the companies that emerged when it burst. He is also a member of the AstraZeneca board and becomes most animated when talking about how AI can be harnessed to benefit medicine.
“Think about what you do today with brand new drugs. You actually do trials on animals before you do trials on humans. That’s 1950s kind of stuff, right?” AI means existing “paradigms . . . can be completely shattered”.
He mentions using AI on DNA and RNA research. The improbable can happen, it seems — much like turning a company that started in a barn into a $175bn powerhouse. There is, he says, an “opportunity to cure cancer in our lifetime. It’s pretty real.”
