With its investment in OpenAI, Microsoft has become one of the biggest winners in this round of AI wars.
From Bing, to Office, to Windows, Microsoft continues to incorporate OpenAI's technology into its products. Not only has the limelight overwhelmed Google, Jefferies analysts said AI will help Microsoft's cloud business beat Amazon.
But betting on a startup that can disrupt the entire industry, how can such a good thing be called Microsoft to meet?
Does Microsoft rely on luck or strength?
Let go of pride
In December, Peter Lee, who oversees Microsoft's vast research efforts, briefed CEO Satya Nadella and her deputy on a series of tests Microsoft conducted on GPT-4, telling Nadella that Microsoft researchers were shocked by the model's ability to understand conversational language and produce human-like answers, and they believed that GPT-4 burst with the spark of general artificial intelligence comparable to human thinking ability.
But Nadella abruptly interrupted Lee to ask to know how OpenAI surpassed the capabilities of Microsoft's 1,500-person research team for more than a decade on AI projects.
Microsoft CEO Satya Nadella
According to The Information, Lee, executive vice president and head of Microsoft Research, said:
OpenAI built GPT-4 with 250 people. Why do we have Microsoft Research? ”
It's a stark question that goes to the heart of how big tech companies like Microsoft try to keep up with the pace of innovation that is shaking up the industry, but often fail to keep up. Due to turf battles and egos, established companies often pass up the opportunity to embrace promising new technologies, only to pay for their decisions years later.
Microsoft has made the same mistake in the past, but with OpenAI, Microsoft has taken a different approach: It has put aside its pride and gone all out to integrate the startup's technology into its own products. The decision also gives OpenAI the money and computing resources that most companies don't use in their lifetimes, and the opportunity to share in the profits it makes from its own customers.
In fact, there were doubts within Microsoft at the time about investing in OpenAI.
Back in 2009, Microsoft's internal research team began developing AI models for speech recognition and text generation.
In 2019, Microsoft made its first $1 billion investment in OpenAI.
At first, Lee said, he and his team were skeptical of OpenAI and worried that Nadella had made a mistake. Because it's hard to believe that OpenAI can do something in just a few years that Microsoft researchers haven't been able to do in more than a decade.
Lee said:
"I joke that there are five stages of grief, the first of which is doubt." "I was very skeptical, and then when I saw the people around me excited about it, I was really worried and angry, and I thought they were fooling themselves."
Lee wasn't alone in expressing skepticism. According to The Information, even Bill Gates warned Nadella not to invest in OpenAI.
Over time, these doubts began to fade.
When Microsoft researchers compare OpenAI's language model side-by-side with Microsoft's internal model (collectively known as Turing), there's no denying that OpenAI has built something much more complex. Lee said that while the self-esteem of Microsoft researchers has taken a hit, OpenAI's technology has breathed life into the department.
Earlier this year, Microsoft merged OpenAI's popular chatbot, ChatGPT, into Microsoft's search engine, Bing. Nadella joins Microsoft CTO Kevin Scott to push the 230,000-employee company's 230,000-employee business units to use OpenAI's big-language model. They believe that OpenAI will fundamentally reshape the way people interact with software.
At its annual software developers conference on Tuesday, Microsoft announced the results of that effort: OpenAI-powered "copilot" products built into Windows, Edge, email and Office software that help users automatically generate documents, summarize long texts and search for files with simple questions.
Microsoft CTO Kevin Scott at Tuesday's developer conference
It's too early to tell whether Microsoft's bet on OpenAI will pay off for the company. For now, though, the rapid pace of its AI program makes its biggest rival look clumsy. For example, infighting between Google's AI teams has hindered the company's AI development; Apple's head of AI is also grappling with similar internal turf battles between AI teams for employees and resources.
Sheila Gulati, managing director of Tola Capital and former Microsoft manager at Seattle-based venture capital firm, said:
"Five years ago, everyone would have said that the clear leader in AI was Google. Now most people would probably say Microsoft. ”
Tempering the "secret weapon"
With OpenAI, is the huge Microsoft Research still necessary?
The answer is that even as Microsoft began to integrate OpenAI into Microsoft's product architecture, Nadella decided not to abandon Microsoft's own AI research work.
During Nadella's tense exchange with Lee in December, other executives came out to defend the work of Microsoft researchers, including Mikhail Parakhin, who leads Microsoft's Bing Search and Edge browser teams.
After cross-examining Lee at the meeting, Nadella called him privately to thank him for his work in understanding and implementing OpenAI technology and getting it through enterprise customer audits. Nadella said he sees Lee's team as a "secret weapon."
In an interview with The Information, Lee said:
"Nadella is both a left-brain and a right-brained person, and he knows when to put pressure, but there's no doubt that you end up feeling valued and supported."
In 2019, after investing in OpenAI for the first time, Nadella commissioned Microsoft researchers to double-check whether OpenAI's technology was as good as some AI experts claimed. According to Lee's recollections, Nadella wanted Microsoft Research to be the company's "gatekeeper" for OpenAI, telling him "if the emperor was not wearing clothes."
At the same time, Nadella is always urging researchers to be bolder in their plans to use OpenAI. At the weekly AI 365 conference, he is actually the product leader of Microsoft's AI initiative, providing feedback to the product group to determine how Microsoft will use OpenAI's models. As an example, before Microsoft first invested in OpenAI, Nadella vetoed an early plan proposed by the Office product team to build AI products that relied less on emerging big language models.
Eric Boyd, vice president of AI platforms at Microsoft, said:
"When we first started working with the Office team, they would only show something simple. Nadella has been challenging them to be more ambitious, which has driven a lot of collaboration across the company. ”
Build supercomputers
On the road to gambling on OpenAI, Microsoft is not without "growing pains".
Shortly after Microsoft's first investment in OpenAI, Microsoft executives for the first time began communicating their ambitions for OpenAI technology to other company leaders. At the weekly AI 365 conference hosted by Scott and Nadella, Nadella made her priorities clear. He told them that Microsoft's investment in OpenAI gives it the right to reuse the startup's technology, and if OpenAI's model delivers on its promises, the proceeds could eventually make up for almost every product Microsoft sells.
To do that, though, Microsoft would have to make a very, very large computer for OpenAI.
In early meetings, OpenAI executives told Microsoft that they needed amazing computing power to support its AI models at scale. In a meeting with Scott Guthrie, president of Microsoft's cloud computing division (Azure), and his deputy, Eric Boyd, vice president of Microsoft's cloud AI platform company, Nadella said that this means Microsoft needs to build the largest computer ever built to meet the needs of OpenAI.
Boyd said in an interview:
"It's definitely a bold attempt, but we're very confident that it will be a rewarding one."
Guthrie explained that building state-of-the-art server hardware will also ultimately help Microsoft build a profitable business serving other customers using the same type of system, so he's willing to bear the expense.
Azure has budgeted $1.2 billion over several years to build OpenAI supercomputers, according to people with direct knowledge of the project. This dwarfs Microsoft's early spending on building server clusters for its own products, the person said.
Other sacrifices were needed to build the supercomputer.
To get the job done, Microsoft pulled teams of engineers away from maintaining other servers within Azure, reducing Microsoft's ability to conduct large-scale reliability testing on the largest computer clusters it operates.
At the time, one of Microsoft's biggest AI customers was Meta, which was using Azure servers to test large-scale AI products. Azure leaders have reportedly warned Meta that these servers may have reliability issues, given that Azure engineers are also building large clusters of graphics processing units for OpenAI (the type of server chip best suited to run AI models) (these warnings haven't damaged Microsoft's relationship with Meta, according to people familiar with the matter).
At the same time, Microsoft invested a lot of resources on this supercomputer to cause a commotion within the company.
According to a person familiar with the matter, product leaders inside Office 365 complained to Rajesh Jha, the company's executive vice president for experience and devices, that they had to apply and get approval to use a small amount of server space for product development and testing. In contrast, OpenAI's leaders can ask for as much compute as they need, and Azure builds specifically for them, the person said.
But Jha told the product team that there was nothing he could do about it and advised them to see their investment in OpenAI as an investment in the Office product, as the technology would ultimately enhance their work.
Lots of rebalancing
Microsoft has warned investors that it expects its capital spending to increase significantly as it invests money in the hardware it needs to run AI workloads. Meanwhile, Microsoft CEO Nadella sent a company-wide notice this month that full-time employees won't get a raise next year.
Still, Microsoft's leaders say they've seen a striking unity within the company in pushing AI, which is no easy feat for a tech giant once notorious for bitter battles between departments.
Lee said:
"Our feeling is that at least we are on the winning side. This makes it easier for everyone to cope and turn to thinking about what will happen next. ”
Sumit Chauhan, Corporate Vice President in charge of the Office 365 product team, said:
"For the first time ever, we're all working toward a unified AI architecture that everyone in the company can leverage." "We saw this as our opportunity and we worked really hard."
It wasn't until this year that Nadella's bet on AI began to reshape Microsoft's product line. In apps like Bing, Office, and security, the company has rolled out new features that leverage OpenAI's model to perform basic tasks based on a user's simple written prompts.
It remains to be seen to what extent these new products will boost Microsoft's business.
Chauhan said:
"To get there, we have to do a lot of rebalancing." "But I can't remember the last time I experienced this level of excitement."
This article does not constitute personal investment advice, does not represent the views of the platform, the market is risky, investment needs to be cautious, please make independent judgment and decision-making.