What has actually been achieved on this video call? It takes Jared Spataro just a few clicks to find out. Microsoft’s head of productivity software pulls up a sidebar in Teams, a video-conferencing service. There is a 30-second pause while somewhere in one of the firm’s vast data centres an artificial-intelligence (AI) model analyses a recording of the virtual meeting so far. Then an impressively accurate summary of your correspondent’s questions and Mr Jared’s answers appears. Mr Jared can barely contain his enthusiasm. “This is not your daddy’s AI,” he beams.
Teams is not the only product into which Microsoft is implanting machine intelligence. On March 16th the software giant announced that almost all its productivity software, including Word and Excel, were getting the same treatment. Two days earlier, Alphabet, Google’s parent company, announced a similar upgrade for its productivity products, such as Gmail and Sheets.
The announcements add to a spate of similar ones in the past month or so from America’s tech titans. OpenAI, the startup which is part-owned by Microsoft and which created ChatGPT, an AI conversationalist that has taken the world by storm, released GPT-4, a new super-powerful AI model. Amazon Web Services (AWS), the e-commerce giant’s cloud-computing arm, said it will expand its partnership with Hugging Face, another AI startup. Apple is reportedly testing the use of new AI models across its business, including with Siri, its virtual assistant. Mark Zuckerberg, the boss of Meta, said he wants to “turbocharge” Meta’s products with AI. Adding to its productivity tools, on March 21st Google launched its own AI chatbot to rival ChatGPT, called Bard.
The frenzy of activity is the result of a new wave of AI models, which are making their way from lab to the real world. No group of companies stands to benefit or lose out more than big tech. All five giants claim to be laser-focused on AI. What that means for each in practice, though, differs. Two things are already clear. The race for AI is heating up. And even before a winner emerges, the contest is changing the way that big tech deploys the technology.
AI is not new to tech’s titans. Amazon’s founder, Jeff Bezos, quizzed his teams on how they planned to embed it into products in 2014. Two years later Sundar Pichai, Alphabet’s boss, started to describe his firm as an “AI-first company”. The technology underpins how Amazon sells and delivers its products, Google finds stuff on the internet, Apple imparts its smarts on Siri, Microsoft helps clients manage data and Meta serves up adverts.
The new GPT-4-like “generative” AI models nevertheless look like a turning point. The firing gun sounded in November, with the release of ChatGPT, which became hugely popular thanks to its uncannily human-like ability to generate everything from travel plans to poems. The thing that makes such AIs generative is “large language models”. These analyse content on the internet and, in response to a request from a user, predict the next word, brush stroke or note in a sentence, image or tune. Many technologists believe they mark a “platform-shift”. AI will, on this view, become a layer of technology on top of which all manner of software can be built. Comparisons abound to the advent of the internet, the smartphone and cloud computing.
The tech giants have everything they need—data, computing power, billions of users—to thrive in the age of AI, and consolidate their dominance of the industry. But they recall the fate of once-dominant firms, from Kodak to BlackBerry, that missed previous digital platform shifts, only to sink into bankruptcy or irrelevance. So whether or not the AI evangelists are correct, big tech isn’t taking any chances. The result is a deluge of investments. In 2022, amid a tech-led stockmarket crunch, the big five poured $223bn into research and development (R&D), up from $109bn in 2019 (see chart 1). That was on top of $161bn in capital expenditure, a figure that had also doubled in three years. All told, this was equivalent to 26% of their combined annual revenues last year, up from 16% in 2015.
Not all of this went into cutting-edge technologies; a chunk was spent on prosaic fare, such as warehouses, office buildings and data centres. But a slug of such spending always ends up in the tech firms’ big bets on the future. Today, the wager of choice is AI. And the companies aren’t shy about it. Mr Zuckerberg recently said AI was his firm’s biggest investment category. In its next quarterly earnings report in April, Alphabet plans to reveal the size of its AI investment for the first time.
To tease out exactly how the companies are betting on AI, and how big these bets are, The Economist has analysed data on their investments, acquisitions, job postings, patents, research papers and employees’ LinkedIn profiles. The examination reveals that serious resources are being spent on the technology. According to data from PitchBook, a research firm, around a fifth of the companies’ combined acquisitions and investments since 2019 involved AI firms—considerably more than the share targeting cryptocurrencies, blockchains and other decentralised “Web3” endeavours (2%), or the virtual-reality metaverse (6%), two other recent tech fads. According to numbers from PredictLeads, another research firm, about a tenth of big tech’s job listings require AI skills. Roughly the same share of big tech employees’ LinkedIn profiles say that they work in the field.
These overall numbers conceal big differences between the five tech giants, however. On our measures, Microsoft and Alphabet appear to be racing ahead, with Meta snapping at their heels. As interesting is where the five are deciding to focus their efforts.
Consider their equity investments, starting with those that aren’t outright acquisitions. In the past four years big tech has taken stakes in 200-odd AI firms. And these investments are accelerating. Since the start of 2022, the big five have together made roughly one investment a month in AI specialists, three times the rate of the preceding three years.
Microsoft leads the way. One in three of its deals has involved AI-related companies. That is twice the share at Alphabet (one of whose venture-capital arms, Gradient Ventures, invests exclusively in AI firms and has backed almost 200 startups since 2019) and Amazon. It is more than six times that of Meta, and infinitely more than Apple, which has made no such investments at all (see chart 2). Microsoft’s most important bet is on OpenAI, whose technology lies behind the giant’s new productivity features and powers a souped-up version of its Bing search engine. The $11bn that Microsoft has reportedly put into OpenAI would, at the startup’s latest rumoured valuation of $29bn, give the software giant a stake of 38%. Microsoft’s other notable equity investments include those in D-Matrix, a firm that makes AI technology for data centres, and in Noble.AI, which uses algorithms to streamline lab work and other R&D projects.
Microsoft is also a keen acquirer of whole AI startups; nearly a quarter of its acquisition targets, such as Nuance, which develops speech recognition for health care, work in the area. That is a similar share to Meta, which is a more eager buyer than piecemeal investor. As with equity stakes, AI’s share of Alphabet acquisitions have lagged behind Microsoft’s since 2019. But these, plus its equity stakes are shoring up a formidable AI edifice, one of whose pillars is DeepMind, a London-based AI lab that Google bought in 2014.. DeepMind has been behind some big advances in the field. It developed AlphaFold, a system which can predict the shape of proteins, understanding of which is both critical in drug discovery and notoriously hard to ascertain.
But it is Apple that is the most single-minded AI acquirer. Nearly half of its buy-out targets are AI-related. They range from AI.Music, which generates tailor-made tunes, to Credit Kudos, which uses AI to assess the credit worthiness of loan applicants. Apple’s acquisitions have historically been small, notes Wasmi Mohan of Bank of America. But they tend to be quickly folded into existing products.
As with investments, big tech’s AI hiring, too, is ramping up. A greater share of jobs listed by Google, Meta and Microsoft require AI expertise than they did, on average, over the past three years (see chart 3). Data from PredictLeads suggest that since 2019 nearly one in four of Alphabets’s job listings have been AI-related (see chart 4). Meta came second, at 8%. According to data from LinkedIn, one in six of Alphabet’s employees mention AI skills on their profile—a touch behind Meta but ahead of Microsoft (Apple and Amazon lag far behind). Greg Selker of Stanton Chase, an executive-search company, observes that demand for AI talent continues to be red-hot, despite big tech’s recent lay-offs.
All these AI boffins are not twiddling their thumbs. Zeta Alpha, a company which tracks AI research, looks at the number of published papers in which at least one of the authors works for a given company. Between 2020 and 2022, Alphabet published about 9,000 AI papers, more than any other corporate or academic institution. Microsoft racked up around 8,000 and Meta 4,000 or so.
Meta, in particular, is gaining a reputation for being less tight-lipped about its work than its fellow tech giants. Meta’s AI-software library, called PyTorch, has been available to anyone for a while; since February academic researchers can freely use its large language model called LLaMA, the details of whose training and biases have also been made public. All this, says Joelle Pinneau, the head of Meta’s open-research programme, helps it attract the brightest minds (who often make their move to the private sector conditional on a continued ability to make the fruits of their labours public).
Indeed, if you adjust Meta’s research output for its revenues and headcount, which are much smaller than Alphabet’s or Microsoft’s, and only consider the most-cited papers, Mr Zuckerberg’s firm tops the research league table. And, points out Ajay Agrawal of the University of Toronto, openness brings two added benefits besides luring the best brains. Low-cost AI can make it cheaper for creators to make content, including texts and videos, that draw more eyeballs to Meta’s social networks. And it could dent the business of Alphabet, Amazon and Microsoft, which are all trying to sell AI models through their cloud platforms.
The AI frenzy is, then, in full swing among tech’s mightiest firms. Promisingly for them, their AI bets are already beginning to pay off, by making their own operations more efficient (Microsoft’s finance department, which uses AI to automate 70-80% of its 90m-odd annual invoice approvals, now asks a generative-AI chatbot to flag dodgy-looking bills for a human to inspect) and by finding their way into products at a pace that seems faster than for many earlier technological breakthroughs.
Barely four months after ChatGPT captured the world’s imagination, Microsoft and Google have introduced the new-look Bing, Bard and their AI-assisted productivity programs. Alphabet and Meta offer a tool which automatically generates an ad campaign based on the advertiser’s objectives, such as boosting sales or winning more customers. Microsoft is making OpenAI’s technology available to customers of its Azure cloud platform. Thanks to partnerships with model-makers such as Cohere and Anthropic, AWS users can access more than 30 large language models. Google, too, is wooing model-builders and other AI firms to its cloud with $250,000-worth of free computing power in the first year, a more generous bargain than it offers to non-AI startups. It probably won’t be long before AI.Music and Credit Kudos pops up in Apple’s music-streaming service and its growing financial offering, or an Amazon chatbot will recommend purchases uncannily matched to shoppers’ desires.
If the platform-shift thesis is correct, big tech could yet be upset by newcomers, rather as they themselves upset an earlier generation of technology giants. The mass of resources that big tech is ploughing into the technology reflects a desire to remain not just relevant, but dominant. Whether or not they succeed, one thing is certain: these are just the modest beginnings of the AI revolution. ■