Llama is the AI model, not the animal. It is a piece of open-source software created by Meta, the social media giant that has pivoted hard to AI and is competing with Microsoft and others to dominate the foundations of the emerging generative AI economy. “Both” is a reference to the fact that this Malaysian agriculture technology company—chosen to show off its use of Microsoft’s technology to Microsoft’s CEO— is using Meta’s rival AI model in addition to GPT-4, the large language model (LLM) created by Microsoft’s strategic partner, OpenAI. Nadella wants his Redmond, Wash.–based software giant to have the most capable, popular AI models on the market. 

“Um, yep, so we are, we are using Llama also,” says Adrian Lee, the chief technology officer at Agroz, the Malaysian startup, a hint of embarrassment in his reply. Agroz, which builds hydroponic farms, has created an AI chatbot to answer farmers’ questions about how best to tend their lettuce and bok choy.  

“What are you using Llama for?” Nadella asks pointedly, standing before Lee at Agroz’s exhibition kiosk.

Agroz, Lee explains, eventually wants to use humanoid robots for farming, and the robots may need to operate offline. Some versions of Meta’s Llama model are compact enough to be embedded in robots or phones, unlike the larger GPT-4. 

“Take a look at Phi,” Nadella says, switching from CEO-as-chief-market-researcher to CEO-as-chief-salesman. Phi is a family of small language models Microsoft has built in-house and offered, like Llama, as open-source software. Nadella apparently knows the parameter count of the smallest Phi model off the top of his head, and he rattles it off—just 3.8 billion parameters, making it small enough to run “on the edge.” 

Lee says Agroz will experiment with Phi. Meanwhile Agroz’s CEO Gerard Lim offers Nadella some Agroz-grown bok choy. “Oh, you want me to eat it?” Nadella says. He takes a small piece. “Mmmm,” he says, chewing vigorously, and, no doubt, chewing over the lesson about how perilously competitive the AI race is. 

It’s a telling exchange. Few companies have benefited as much from the generative AI boom as Microsoft. Investor fervor for the technology has helped make Nadella’s company a perennial contender for the title of the world’s most valuable corporation, with a market cap that consistently bobs somewhere in excess of $3 trillion. 

When Nadella took the reins at Microsoft in 2014, the company was floundering. Under his predecessor, Steve Ballmer, it had missed the smartphone revolution, was lagging on tablets, and was even losing market share in the PC operating system business that made it a household name. Microsoft’s shares had fallen more than 40% during Ballmer’s tenure. In 10-plus years as CEO, Nadella has reinvigorated the company, successfully steering it through two technological transformations: from PCs to the era of cloud computing, and now, to the age of AI. Nadella’s prescient and early bet on OpenAI and its technology, and the fruitful-if-sometimes-tense relationship that ensued, have given Microsoft as good a shot as any company at supremacy in this new era. The company arguably hasn’t been this powerful since it dominated the PC market in the 1990s’ “Wintel” era. 

But as Nadella begins his second decade at the helm, there’s no guarantee Microsoft will retain its lead. Regulators, hackers, and rivals each present threats serious enough to potentially undermine its industry leadership. Above all, it must reckon with its own leviathan size—and avoid becoming a victim of bureaucracy and bloat when AI’s protean nature calls for speed, agility and finesse.

Nadella’s approach to leadership reflects his acute awareness of these risks. Even when seemingly ahead of the pack on AI, he and his team are constantly listening, their corporate antennae sensitive to the tiniest pivots in users’ needs and preferences. Microsoft is continually investing in technologies and talent that might someday supplant OpenAI’s models—or even supplant generative AI altogether. In other sectors, such hypervigilance might seem like overkill, even paranoia. But as Nadella knows as well as anyone, in tech, platform shifts happen frequently and fast. Blink for a second too long and you’re chasing the future, not making it. 

---

Nadella’s jet-setting tour of Southeast Asia in late April and early May—Indonesia, Thailand, and Malaysia in just three days—showcases AI’s rapid progress. At each stop, he meets government leaders eager to understand how AI can boost their economies and software developers building AI products—from personal tutors to “copilots” for fish farmers—as sophisticated as any of their counterparts in the U.S. or Europe. 

“The thing that is striking to me is the rate of diffusion,” Nadella tells me when I sit down with him in Jakarta. “If I had come to Indonesia in the second year of cloud, or even the server, obviously, there would have been some adoption, but it wouldn’t have been this broad and this fast.” 

The pace of AI’s diffusion is only possible, he says, because it builds on previous innovations, in particular cloud computing, the area where Nadella spent much of his career prior to becoming CEO. It is hard to democratize technology, he notes, when people can’t actually access it in large swaths of the world. But today that’s no longer true. The internet, smartphones, and the cloud are omnipresent.

In one sense, Microsoft navigated the platform shift to the cloud so well, and has so deftly played the opening round of the platform shift to AI, that its future success seems assured. If AI increases economic productivity, which most analysts think likely, global GDP growth should accelerate too, while tech will represent a bigger piece of this enlarged economic pie. Combined, Nadella says, this ought to mean that Microsoft—whose top line has been accelerating at a double-digit annual pace even with economic growth fairly constant at 3%—has nothing to worry about. It helps explain why investors love Microsoft—it looks like a no-lose bet. (Its share price has risen 11-fold under Nadella.) “Microsoft’s growth prospects are pretty straightforward,” Nadella says. “Just do a good job of what we do.” 

But, of course, “just doing what we do,” isn’t simple. There are a thousand ways a colossus like Microsoft can lose its way. It has before—it largely missed mobile, after all. And of the three big technological innovations Nadella famously said in 2017 would shape Microsoft’s future—AI, quantum computing, and mixed reality—the company has really only hit the jackpot with AI. It sank billions into its HoloLens mixed-reality glasses, billing them as “augmented reality for business,” before discontinuing work on the hardware and laying off most of the mixed- and virtual-reality teams in early 2023. It invested $1 billion in a swing-for-the-fences drive to develop a quantum computer based on an exotic subatomic particle called a Majorana fermion that has so far failed to result in a commercially viable machine. For now, Microsoft has fallen back on offering quantum services through a partnership with Quantinuum, a company that is partly a spinoff of Honeywell’s quantum tech.

Nadella says the idea of missing the next big technological leap keeps him up at night. “When the paradigm shifts, do you have something to contribute?” he asks. “Because there is no God-given right to exist if you don’t have anything relevant.” That in turn, he says, requires “a culture that allows you to build capability long before it is conventional wisdom that you need that, to come up with new concepts.” 

“When the paradigm shifts, do you have something to contribute? Because there is no God-given right to exist if you don’t have anything relevant.”

Satya Nadella, CEO, Microsoft

Nadella says build. But find might be more accurate. For an executive who spent nearly his entire career at Microsoft, Nadella is surprisingly willing to go outside the organization to obtain the innovation on which the company’s future depends—whether through acquisitions, partnerships, or hiring. Nadella tells me he also closely tracks how many senior and junior people Microsoft is hiring from other tech companies, seeing the influx as critical to ensuring the company doesn’t become sclerotic. “The only way to keep yourself honest is to bring senior talent that comes from the outside and keeps you intellectually grounded,” he says. 

Those who have worked closely with Nadella say he acts decisively when he senses a big strategic opportunity. His record as acquirer-in-chief supports that idea. To cement Microsoft’s position as the hub for all digital business activity and to expand the company into social networking, he spent $26 billion to acquire LinkedIn in 2016. Two years later, he bought the open-source software code repository GitHub for $7.5 billion in stock. That deal gave Microsoft a better window into what legions of software developers were working on, helping it spot trends, and also gave the company a purchase point for selling those developers on other Microsoft products. More recently, he spent $75 billion to buy video game giant Activision Blizzard. The deal was alternately couched as Microsoft’s “metaverse” play or about acquiring content for its Xbox hardware franchise, but as some analysts pointed out, the more compelling strategic logic was ensuring Microsoft’s cloud is where the world’s most valuable entertainment medium gets built and run.

Microsoft’s dependency on a pipeline of innovation—and innovators—from outside the ranks of its own 221,000 employees has been most glaring when it comes to AI. In 2019, Kevin Scott, a veteran engineer who came to Microsoft when the company acquired LinkedIn, and who Nadella subsequently elevated to chief technology officer, became concerned the company wasn’t making enough progress in AI—that its efforts lacked ambition. Like Nadella, Scott thought AI would be transformative, not just for Microsoft but for society. Yet despite spending hundreds of millions of dollars over a decade, and having a research arm stuffed with leading computer scientists, Microsoft had not achieved the kinds of headline-grabbing breakthroughs Alphabet was regularly churning out from its two AI labs, DeepMind and Google Brain. 

Nadella green-lit Scott’s plan to look for startups that Microsoft could invest in and partner with that might help the company grab a lead in the AI race. Microsoft faces “top competitors who are more vertically integrated,” Nadella says. “And if you want to compete with those players, we felt like partnering is the right approach.” 

Scott homed in on an unusual San Francisco outfit called OpenAI. The startup’s ambition couldn’t have been greater. OpenAI was committed to achieving artificial general intelligence, or AGI, software as capable at cognitive tasks as a person—AI’s holy grail. As it turned out, Nadella had met OpenAI CEO Sam Altman at Allen & Co.’s Sun Valley conference—nicknamed “summer camp for billionaires”—and come away intrigued. Altman, meanwhile, twigged that Microsoft might be the deep-pocketed backer OpenAI needed to achieve its goals.

Since then, Microsoft has invested at least $13 billion into the AI company, and the startup’s technology has become the bedrock of Microsoft’s AI offerings. OpenAI’s models underpin the company’s “Copilot”-branded AI products, including GitHub Copilot, its AI software coding assistant, and generative AI features it has built into its Microsoft 365 office productivity software. They also power “Copilot for Bing,” its free, consumer-facing chatbot and generative AI search engine. Microsoft’s integration of AI Copilot features into its ubiquitous Office 365 products gives the company a lot of ways to hook customers. Still, some of those customers have balked at the hefty $30 per user per month Microsoft charges for Copilot. “No one budgeted a 50% lift on their M 365 license,” says Jason Wong, a distinguished vice president, analyst at technology research firm Gartner.

Microsoft’s Azure cloud computing business has gotten the biggest financial boost from OpenAI’s models so far, showing the symbiosis between Nadella’s two big strategic transformations. The desire of companies to tap OpenAI’s GPT models through Azure has contributed meaningfully to the company’s revenues. In the quarter that ended in March 2024, Microsoft attributed 7% of Azure’s 31% overall sales growth to AI. Morgan Stanley estimates that Azure AI services are on track to generate $4 billion in annual sales. Overall, Microsoft’s cloud revenues were $35.1 billion in the quarter, up 23% from a year earlier. That has also helped Microsoft, which ranks second in cloud market share, narrow the gap with the industry leader, Amazon’s AWS. 

Lee Sustar, principal analyst at technology research firm Forrester, says that after spending years chasing AWS, the OpenAI partnership has given Microsoft “the initiative for the first time ever in cloud computing.” The premium Microsoft can charge for AI offerings gives it a chance to recoup the costs of building its cloud. The question, he says, is whether Azure can maintain this momentum given that many customers have piled into AI without knowing how best to generate a return on their investment. “We just went through the ‘buy before you try’ phase of AI development,” he quips.

Of course, Microsoft is also spending big to purchase graphics processing units (GPUs), the specialized chips on which AI applications run, and to construct still more data centers. Its capital expenditures in the first three months of 2024 surged 79% from a year earlier, to $14 billion, and the company has told investors those costs will keep rising at least through 2025. Microsoft has begun designing its own AI chips to lessen its reliance on Nvidia’s GPUs, but those chips are only expected to begin arriving later this year.

---

Meanwhile, Microsoft’s dependency on OpenAI carries risks. This became painfully obvious when, days before Thanksgiving 2023, OpenAI’s board suddenly fired its charismatic CEO and cofounder, Sam Altman, saying he had not been “consistently candid,” throwing the startup’s future into doubt.

Nadella scrambled to reassure customers and investors that Microsoft’s access to OpenAI’s tech wasn’t in jeopardy. He also helped catalyze Altman’s reinstatement by offering to hire him, OpenAI cofounder Greg Brockman, and any other OpenAI employees who wanted to join them. The threat of mass resignations forced OpenAI’s board to rehire Altman. As a condition of his return, most of the board members involved in his firing were replaced, and Microsoft gained an observer’s seat on the body. Nadella claimed in a television interview in January that Microsoft’s board position was unimportant, but that he wanted to see “good governance and real stability.”

Beyond that blowup, there have been periodic reports hinting at tensions between the partners: OpenAI has been too slow delivering new, more powerful AI models; it has tried to sell its tech directly to big Microsoft customers. Scott acknowledges occasional friction on both sides. “Everybody just sort of feels this incredible sense of urgency. And so you’re just putting everything that you’ve got into it, and we’re all occasionally getting frustrated with each other because we just want things to move forward so quickly,” he says. 

Top Microsoft execs insist they are exuberant about OpenAI overall. When I catch up with Nadella a few days after his return from Southeast Asia, he emphasizes that OpenAI delivered Microsoft GPT-4, which by most benchmark tests remains the most high-performing AI model in the market, as well as the model most companies are using for generative AI applications. Work on an even more powerful GPT-5 model is well underway. “We feel very good about where we are, with what we have been able to do on top of GPT-4,” he says. 

Altman describes a close relationship with Microsoft, and in particular with Scott, whom he calls “one of the most important partners, colleagues, whatever you want to call it, of my entire career.” He and Scott speak multiple times a week, and Altman says he leans on the Microsoft CTO for thoughts on strategy and even AI’s societal implications. 

OpenAI’s sweeping ambitions and Microsoft’s practical focus on execution don’t always harmonize neatly. In Malaysia, I watched Nadella deflect a question from a local CEO about what AGI—OpenAI’s ultimate goal—might mean for humanity. When I ask Altman about Nadella’s seeming discomfort, Altman says he and the Microsoft CEO “have different jobs and different roles, and we are kind of focused on different time horizons. But it’s on the same vector, so I think it’s fine.” Outside of those differences, Altman describes tight collaboration between Microsoft and OpenAI at every level. “We don’t do ropes course together or anything,” he says, but notes that executives and engineers from the two companies are in constant communication, frequently work at one another’s offices, and bond at post-work dinners.

“We have different jobs and different roles, and we are focused on different time horizons. But it’s on the same vector.”

Sam Altman, OpenAI, on how he and Nadella approach the goal of Artificial General Intelligence

That said, competitors are also innovating at AI’s frontier, and customers are still working out how best to derive value from AI—so Nadella wants options beyond OpenAI. “Microsoft is hedging its bets,” notes Arun Chandrasekaran, a distinguished vice president, analyst for tech innovation, cloud, and AI at Gartner. With $80 billion in cash and short-term investments sitting on Microsoft’s balance sheet as of the end of March 2024, Nadella can afford to place multiple wagers. In addition to helping cloud customers implement sometimes-pricey OpenAI models, Microsoft offers them free, open-source AI models, including some from rivals such as Meta’s Llama. These models are generally less capable than GPT-4, but for some uses will do the trick at a potentially lower cost. 

In April, Microsoft made a $1.5 billion investment in Abu Dhabi–based G42, a technology conglomerate closely tied to Abu Dhabi’s royal family that has trained open-source Arabic language AI models. It has invested $16 million in Mistral, a French AI startup that has built highly capable open-source language models. Microsoft has also created its own family of open-source small language models, Phi, that pack an outsize punch.

“Customers want choice and optionality,” Chandrasekaran says. Microsoft’s goal is to ensure “whichever way the customer goes they will still spend money on Azure.” If they do, he explains, Microsoft can build lucrative long-term relationships with them. Most companies need to run AI applications on the same cloud that houses their data. Once Microsoft has that data, it is difficult for customers to leave—“data has gravity,” Chandrasekaran says—and Microsoft can sell them additional services around data management and security. 

Microsoft’s most significant AI play aside from OpenAI came in March, when it hired DeepMind cofounder Mustafa Suleyman and much of the staff from his AI startup Inflection to form the core of a new consumer AI division. Microsoft also agreed to pay $620 million to license Inflection’s technology—it had built one of the best LLMs outside of OpenAI’s—a move that Microsoft executives told me was intended to forestall possible trade secret theft lawsuits from Inflection and its investors. (Microsoft execs dispute speculation that the hiring and license deal was designed to evade antitrust scrutiny.) 

What Microsoft really wanted from Inflection was talent. Tech publication The Information has reported that the former Inflection team is hard at work on a new LLM that might rival OpenAI’s systems. When I asked Nadella about this, he said only that Microsoft will continue to innovate on AI models in-house. Chandrasekaran, the Gartner analyst, says Microsoft needs its own expertise training the most capable, large frontier models. A key provision of its partnership with OpenAI is that Microsoft loses its automatic right to license OpenAI’s most powerful AI models if the San Francisco company achieves its stated goal of AGI—a determination that is at the sole discretion of OpenAI’s nonprofit board. Nadella likes to have a plan B. Suleyman and the AI experts he brought from Inflection give him one.

The quest to keep tabs on where AI is heading extends beyond LLMs. One warm, overcast April afternoon, I rode through London’s congested streets in a self-driving Jaguar I-Pace powered by software from a Microsoft-backed startup called Wayve. Watching a car steer, stop, and start itself had all the wow factor you’d expect, but the tech powering it is where Microsoft’s deeper interest lies.

Chris Young, Microsoft’s executive vice president for business development, strategy, and ventures, tells me investments like Wayve are about gaining “market signal”—in this particular case, a view into the emerging world of “embodied intelligence,” the term for AI systems that operate in the physical world, controlling cars, drones, or robots. Microsoft makes investments in startups both directly, from its cash reserves, and through its corporate venture capital arm, M12. Since arriving at Microsoft in 2020, Young has, with Nadella’s blessing, shifted M12’s strategy from generating financial returns to focusing on startups that can help Microsoft expand its business, gain market intelligence, or keep up with key innovations. Helping to train Wayve’s AI will give Microsoft a better understanding of the auto sector’s future cloud-computing demands as carmakers adopt ever more sophisticated driver assistance technology.

Young says that Wayve is not Microsoft’s only investment in embodied intelligence. It has also backed humanoid robotics company Figure. And the company is putting money into synthetic data creation, cybersecurity, and even biotech and drug discovery that are seen as key potential future markets for Microsoft’s applications or its cloud services.

---

A good surfer rides a wave as much by feel as sight, sensing the water passing beneath the board. Nadella’s Microsoft surfs technological waves by keeping as close as possible to developers—the engineers and coders who build software within companies—and trying to remain responsive to what they want. Nadella’s trip to Southeast Asia was a C-suite-level exercise in developer dialogue.

Among the cultural changes Nadella instituted at Microsoft shortly after becoming CEO was the idea of “a growth mindset”—that employees should stop being “know-it-alls” and seek to become “learn it-alls.” That philosophy now guides the company’s interactions with developers. Scott Hanselman, a Microsoft vice president who looks after its global developer community, says that in the pre-Nadella era, Microsoft had a reputation for dictating to developers and forcing them to use the company’s entire tech stack. Now it allows developers to pick and choose components, and seeks constant feedback on features they want Microsoft to build. “Azure is beholden to them; they are the customer, and they will keep using it if they find it helpful and delightful,” he says. 

Microsoft’s latest effort to delight developers centers on GitHub Copilot, its AI coding tool. Amanda Silver, corporate vice president of product design, user research, and engineering systems, rattles off a long list of stats from studies that have shown how much faster and more accurately programmers code with Copilot. Coders have enthusiastically embraced the tool. In April, Microsoft went further, debuting GitHub Copilot Workspace. This software allows a developer to describe a feature they want to build, and then the software generates all the necessary code automatically. Copilot is binding Microsoft and developers in an ever-tighter vortex—one that has the potential to draw in more and more people as Copilot helps democratize software creation. 

Copilot’s success isn’t uncomplicated, however. It was trained on code developers had previously uploaded to GitHub. Some of those developers claim Microsoft’s use of their code violates the license terms under which they uploaded it, and they’ve sued Microsoft and OpenAI for breach of contract. The case—as well as copyright infringement lawsuits Microsoft and OpenAI face, including from the New York Times—point to something else that Nadella says keeps him up at night: the possibility the company might lose what he calls “its license to operate.” 

That license, he says, “rests on our social contract,” which makes several demands of Microsoft. “Trust is one of them,” he says. Another is whether Microsoft’s products and services create economic surplus everywhere it operates. “Why would anyone in Asia give us a license to operate there?” he tells me after his swing through the region. “Are we making small businesses there more productive, multinationals there more competitive, the public sector in those countries more efficient, or driving better health outcomes and education outcomes?” 

To ensure it can deliver those benefits—and to buy itself goodwill—Microsoft has announced a series of multibillion-dollar investments around the world. During his Asia tour, Nadella announced Microsoft would help train 2.5 million people across Southeast Asia in AI skills. It announced that it would spend $1.7 billion for data centers and training in Indonesia and $2.2 billion in Malaysia over the next four years, as well as announcing Microsoft’s first data center in Thailand at an undisclosed cost. The company is making similar investments in infrastructure and training in Europe and Japan.

But when it comes to trust, there are some ominous blips on the radar screen. The alleged plundering of data to train AI models may mar the company’s reputation. Microsoft has suffered a series of damaging cyberattacks over the past two years, making it seem suddenly vulnerable in a domain where it was once considered best-in-class: cybersecurity. A further potential vulnerability might be the proliferation of deepfakes, AI-assisted fraud, and AI-generated junk content flooding the internet, some of it created with Microsoft’s tech. Nadella tends to argue, conveniently, that deepfakes and disinformation should be policed at the point of distribution (the social media platforms) not the point of creation (Microsoft’s AI apps). It isn’t clear the public and governments agree. 

A loss of trust would feed into another big threat facing Microsoft: regulation. Big Tech companies, Microsoft included, are operating under unprecedented government scrutiny globally, and AI’s emergence has only increased that attention. “The days of the deregulated technology sector are over,” says Brad Smith, Microsoft’s vice chair and president, and the person who spearheads Microsoft’s relationship with governments around the globe. 

Last year, Microsoft agreed to unbundle its Teams workplace collaboration tools from Office, in a bid to avert a European Union antitrust fine. But the move has failed to assuage regulators. And while EU competition authorities declined to pursue action based on Microsoft’s OpenAI partnership, antitrust enforcers in the U.S. continue to probe the relationship. Smith acknowledges that acquisitions are more difficult to make than five years ago; the company still faces a Federal Trade Commission claim related to its Activision Blizzard purchase. Meanwhile, the company must comply with a raft of laws in Europe aimed at Big Tech, including a new AI Act.

A quarter-century ago, the U.S. government successfully sued Microsoft over its bundling of Internet Explorer with Windows, almost resulting in the company’s breakup. Smith tells me Microsoft has benefited from that experience—“with having tasted defeat,” he asserts—and come away wiser. Under Nadella, Microsoft has adopted a more conciliatory approach to regulators than founder Bill Gates did as CEO. “Everybody who’s in the technology sector is going to need to adapt to succeed in this combination of engineering and regulation,” Smith says. “I think we’re probably more willing to lean into the need to adapt than some others.”

---

Indeed, as Microsoft’s ambitions—and its commitment to achieving them—grow larger, it will need even more collaboration with governments. 

AI and cloud services demand data centers, for example, and data centers need energy. In Kuala Lumpur, Nadella told a group of CEOs that a lack of renewable energy sources was limiting where Microsoft could build them. If governments wanted Microsoft to invest in their countries, he said, they should fund green-energy generation and upgrade power grids. In the U.S., Microsoft faces similar constraints. It has even explored ways AI could be used to streamline regulatory approvals for new nuclear power plants. Microsoft also needs to lessen the carbon intensity of manufacturing AI chips and building data centers; the company revealed in May that its CO2 emissions have spiked 30% since 2020, in large part because of those activities, jeopardizing its pledge to become carbon negative by 2030.

Microsoft and OpenAI are also, according to a story in The Information, considering construction of a vast data center for OpenAI’s future models that might cost as much as $100 billion to equip. Scott says that report is “amusingly wrong,” but declines to say how. What he does say is that “we most certainly are not going to be outspent by rivals.”

This whatever-it-takes attitude makes Microsoft a formidable competitor. But it may not be enough to ensure it remains the world’s most valuable company. Staying on top, Nadella tells me in Jakarta, requires that Microsoft continue meeting “unmet, unarticulated needs of customers.” That in turn demands that the company “stay humble, stay hungry, and exhibit a growth mindset.” The company can never coast. “I don’t take it for granted,” he says. 

---

Beyond ChatGPT: Microsoft’s other bets

Microsoft’s partnership with OpenAI has allowed it to leap to the front of the generative AI race. But AI technology is rapidly evolving, and Microsoft’s cloud customers want to explore a variety of AI models for different use cases and to better control costs. So Microsoft is hedging its bets with investments like these:

1. G42 

Microsoft has invested $1.5 billion in this Abu Dhabi–based tech conglomerate, which has built some of the best large language models (LLMs) for Arabic. The investment is widely seen as a U.S.-brokered reward for G42 severing its once-close ties with China and Chinese chip companies.

2. Mistral 

Founded by former DeepMind researchers in 2023, the Paris-based startup has built a family of highly capable, open-source LLMs. Mistral has raised hundreds of millions of dollars in venture funding. Microsoft made a small $16 million investment in the company and offers Mistral’s models on Azure.

3. Wayve

London-based Wayve builds software for self-driving cars that doesn’t require specialized sensors or detailed street mapping. Instead, it takes in visual information and directly outputs the correct driving actions, a significant advance over many competing systems. Microsoft has joined multiple funding rounds for Wayve, including a $1 billion fundraise in May.

4. Figure 

A startup building humanoid robots for commercial use, Figure has won plaudits for impressive demos of its technology. It has also partnered with OpenAI to provide AI brains for the robots. Microsoft participated in a $675 million funding round for the Sunnyvale, Calif., company in February that valued it at more than $2 billion. 

Jeremy Kahn is Fortune’s AI Editor and author of Mastering AI: A Survival Guide to Our Superpowered Future (Simon & Schuster, July 9).

This article appears in the June/July 2024 issue of Fortune with the headline, “The vigilance of Satya Nadella.”