Microsoft’s Copilot Woes Are a Familiar Problem (Premium)

With Steve Jobs on medical leave in early 2009, a Barclays analyst asked Apple executives during a quarterly earnings call what would happen to the company if Jobs was unable to return. Apple CFO Peter Oppenheimer dismissed the question by noting that Jobs was still the CEO and that Tim Cook, then Apple’s COO, was responsible for day-to-day operations. But Cook stepped in and provided more information.

“The values of our company are extremely well entrenched,” he said. “We believe that we are on the face of the earth to make great products, and that’s not changing. We are constantly focusing on innovating. We believe in the simple not the complex. We believe that we need to own and control the primary technologies behind the products that we make, and participate only in markets where we can make a significant contribution.”

There was more–Cook also highlighted saying no and focusing, the deep collaboration between Apple’s product groups, not settling on anything less than excellence, having the self-honesty to admit when they’re wrong, and having the courage to change–but this overt and specific description of his leadership style was shocking to Steve Jobs, who believed only he was qualified to run Apple, and it became known as the “Cook doctrine” among those who follow the company.

Whether Apple under Tim Cook has met the tenets of this doctrine are debatable, but one of his points is relevant to this discussion. Cook’s desire that Apple “own and control the primary technologies” behind its products can be seen in Apple’s maniacal pushes to eliminate third parties like Qualcomm and Intel from the supply chain, and it may help explain the problems the company has had creating conversational Siri: Apple is so intent on going it alone that it never seemed to internalize the reality that it may never catch up with fast-moving AI pioneers like OpenAI and Google.

The issue for Apple in this instance is that a strategy that always worked well for it in the past suddenly and unexpectedly came up short. And it just doesn’t know how to respond to that. When something keeps working, you stop considering alternatives. Everyone is familiar with the notion that insanity can be defined as doing the same thing repeatedly and expecting different results. But sometimes you do the same thing, and you do get different results. It’s no wonder this is confusing.

This problem isn’t unique to Apple in any way. Indeed, it’s human nature. But when a company becomes dominant in certain markets, it creates a blind spot for itself that may one day undermine that success. Witness Intel, a chip designer and manufacturer that can no longer design or build competitive chips.

Or witness Microsoft, a company that saw so much success with Windows that it mimicked that strategy in other markets with diminishing results over time.

Microsoft’s current successes can be tied to leadership that saw this problem, instituted change, and made the right bets, just as Intel’s current failures can be tied to leadership that did not. But Microsoft, like Apple, isn’t immune to the blind spots that occur when one is too successful. And when I look at Copilot, I see a failure of strategy, not a failure of technology.

Whatever happens, Microsoft will always be credited for jump-starting the AI era we’re now navigating. Microsoft didn’t invent the technologies that made this possible, but it did invest in OpenAI, the company that did. And it orchestrated its investments–made the right bets–in such a way that it retained a level of control over OpenAI that some might describe as anticompetitive and illegal. That control lies at the center of the spiraling relationship between these two symbiotic companies. Which need each other to survive and but also can’t stand each other.

The record shows that OpenAI released ChatGPT along with its GPT‑3.5 models in November 2022, and that it was immediately a sensation. Thanks to its unprecedented capabilities, ChatGPT quickly became the fastest-growing Internet service of all time, gaining over 100 million users within two months and triggering a panic at Google, which had pioneered the transformer architecture underlying OpenAI’s models. ChatGPT also set off alarm bells inside Microsoft. But thanks to its investment in, and control over, OpenAI, the software giant was well positioned to act.

The record further shows that Microsoft did act, and decisively. CEO Satya Nadella and his senior leadership immediately did two things the Cook doctrine does not allow: They had the courage to change, and they embraced outside technology as being core to the company’s future and made it their own. In February 2023, Microsoft held a surprise AI event one day before a pre-announced Google AI event. There, it announced “a new, AI-powered Bing and Edge browser” that would include “a new chat experience and the ability to generate content.” These new experiences were powered by a “next-generation OpenAI model,” though Microsoft was careful to present its offering as unique thanks to the proprietary way it interacted with OpenAI technologies. And Microsoft would quickly overturn the obvious branding mistake and recast this work under the Copilot brand.

I’m not much of a prognosticator, but I did opine in January 2023, ahead of the Microsoft announcement, that maybe AI is the next wave. In the post, I outlined the history of Microsoft’s involvement with OpenAI as well as its own working internally on bringing AI capabilities to its hardware and software platforms. It’s an interesting read today, given what’s happened since, and I was right about Microsoft limiting certain capabilities to those PCs that included dedicated NPUs, though I referred to the result as Windows 12, while Microsoft eventually went with Copilot+ PC.

In any event, Microsoft’s rollout of Copilot capabilities across all the products and services it offers was both chaotic and broad, as it remains over two years later. This mirrored the pandemic-era push it gave to Microsoft Teams, which subsumed the Microsoft 365 organization at the time, but this time company-wide. That is unprecedented in Microsoft’s history. The only similar event was Microsoft’s early 2000s .NET push, in which it planned to rebrand all its products as .NET–Windows would become Windows.NET, Office would become Office.NET, and so on–though that effort died before it even began. Only one product, Visual Studio, used that branding before quickly reverting.

The Copilot brand is a good one as it’s immediately obvious and overtly descriptive: This is technology that you can use side-by-side with other apps and services. It was first used in 2021 with GitHub Copilot, which was also made possible by Microsoft’s OpenAI partnership. The transition from Bing branding to Copilot started with the Edge Copilot and Microsoft 365 Copilot, which were announced days apart from each other in March 2023. And then it accelerated from there. By the end of 2023, Microsoft had about 120 Copilot-branded products in development and in production.

This, too, is a common problem. It’s not unique to Microsoft, but it’s fair to say that Microsoft is particularly susceptible, in part because this company is terrible at branding. So when it arrives at a successful brand, as it did with Windows, it beats that brand to death. In the 1990s and early 2000s, Microsoft reused the Windows brand everywhere it could, and there was even a product called Windows Media Player for Mac, for F’s sake. Today, it’s overusing the Copilot brand. One might say it’s like riding a bike. (You may recall that Google suffered from its own branding issues with AI. It went to market first with Bard before switching all its AI solutions over to the Gemini brand.)

Microsoft’s successes over the years can be summarized in different ways. But at a high level, the company has seen great success with licensing software to third parties, as it does with Windows and PC makers; bundling individual apps into a suite, as it did with Office; subscription-based licensing, in which it gets paid monthly over time, starting with enterprises in the early 2000s; and cloud computing and online services infrastructure, with Azure. Each drove generational expansion for Microsoft, and the company today is one of the biggest in the world by market capitalization.

Microsoft’s biggest defeats can be described in a similar way. For example, Windows Phone failed for many reasons, but Microsoft’s desire to license the OS to hardware makers–at a time when Google was providing Android for free and its Google Play apps and services for very little–was a crucial blind spot moment for the company leadership at the time. Success breeds complacency. And it’s not always obvious until it’s too late.

With Copilot, the problems are likewise many.

As I’ve pointed out in the past, the new AI-powered features that users see in Microsoft 365, especially, but also in Windows 11 and elsewhere, were the sorts of updates that customers historically got for “free,” as part of an ongoing subscription or product purchase. But AI is expensive. And so Microsoft has created new payment tiers to help recover some of the costs. With Microsoft 365, that’s a $20 or $30 monthly fee per user, an untenable additional cost on top of a subscription that already costs $10 or so each month per user for consumers and $20 or more per month for business users.

This shift flies in the face of an established Microsoft success, product bundling. Just before this AI era, Microsoft created Teams to take on the threat from Slack, and while it eventually surpassed Slack’s capabilities, its real value was that it was part of Microsoft 365, something existing customers got for “free,” not something that needed to be acquired separately. Antitrust issues would later cast this practice in a new light, triggering changes, but that happened too late to halt Team’s rapid expansion, and after the AI era was underway.

Microsoft co-founder Bill Gates was fond of describing even minor advances as him “betting the company,” but Satya Nadella is inarguably making that phrase a reality with Copilot. This company has spent over $80 billion in the past year–literally, over $20 billion each quarter–building out the expensive infrastructure required to run the inefficient, power-hungry cloud-based AI models that Copilot and ChatGPT require. And while this was defensible to a point from an investment perspective, Microsoft’s board of directors and shareholders finally started asking questions. This led to Microsoft altering its OpenAI partnership so that OpenAI could use other infrastructure providers, allowing the software giant to scale back its spending levels in coming quarters. But make no mistake, $20 billion is still the new minimum on this spending.

The rationale for that spending is getting less defensible with each passing month. The problem is that Copilot has consistently underperformed despite all the investment, functional additions across the stack, and the resulting chaos internally and externally. One month after it announced the Bing chatbot, which would become Copilot, Microsoft claimed it had over 100 million daily active users, an important milestone if you think back to the ChatGPT 100 million user milestone. But it quickly became obvious that Copilot was not successful. It didn’t move the needle on Bing usage. And today, it hasn’t moved the needle at all. Copilot is many things. But one of those things is a Zune-level also-ran.

I’m not exaggerating. It’s only been two years, maybe two and a half years, since Microsoft announced what we now call Copilot. Since then, Every report that’s measured the relative success of AI chatbots, whether in usage or revenues, has either ignored Copilot entirely or relegated it to the bottom of the list. Microsoft has invested literally hundreds of billions of dollars over those years in just infrastructure costs, and that’s on top of the tens of billions its invested in OpenAI and the unknowable sums it’s spent pivoting the entire company to embrace AI and creating an in-house Microsoft AI group that operates like a secret police organization outside the control or oversight of the rest of the company. And it’s like Copilot doesn’t even exist.

The latest such report, from Sensor Tower, is damning: Where individuals have downloaded the ChatGPT app 938 million times, Copilot has seen less than one-tenth as many downloads at just 79 million. Google Gemini has been downloaded 200 million times, and even DeepSeek has more downloads, 127 million, than Copilot. Only Perplexity is, for now trailing Copilot, with 47 million downloads.

Yes, that’s just mobile downloads, a market in which Microsoft has no foothold. But it’s not just mobile, and, even worse for Microsoft, it’s not just consumers.

Microsoft claims that 70 percent of the Fortune 500 is using Copilot, while paid users have tripled in the past year. Both of these figures are nonsense: Only one user in a company needs to be using Copilot to meet the needs of the first claim, and big growth is easy when you’re starting with a tiny, almost non-existent user base. OpenAI, meanwhile, says that there are now 3 million business customers paying for ChatGPT, and many companies that pay for Microsoft 365 are skipping Copilot to pay for ChatGPT instead.

I noted that Copilot is a good brand for all the obvious reasons. But ChatGPT is a successful brand, one that resonates with consumers and businesses a like. It is the Kleenex or Xerox of AI, and it is popular with customers regardless of its merits compared to other AI solutions. That’s not a dig at ChatGPT, it’s routinely near or at the top of any AI chatbot comparisons and will likely remain there for the foreseeable future. But this is an additional hurdle for Copilot because customers just want ChatGPT. This is the same problem that Microsoft Edge faces. No matter how good it is, customers just want Chrome.

Microsoft has thrown money at all kinds of products that failed in the market, and it’s acquired companies like aQuantive and Nokia only to later write them down or off as complete failures. But aQuantive was a $6 billion write-down, while Nokia was a $7.6 billion write-off. The combined cost of those losses is exceeded by a single quarter of capital expenditures on AI infrastructure. Put another way, those were big bets. But Nadella is betting the company on Copilot and AI. And it is not paying off.

This ensures that the chaos continues. But it’s not just Microsoft’s customers who face uncertainty and change. As Microsoft’s relationship with OpenAI continues to unravel, it faces an unpredictable future in which the two companies will split for good at some point. Microsoft will or will not be ready to face that future when it happens. But I’m curious if Nadella and the current leadership will be around to plot the next phase. To date, we haven’t seen any serious dissention from the board or shareholders because of Microsoft’s meteoric stock price gains under Nadella. This mirrors a similar situation with Tim Cook and the Apple Intelligence fiasco he’s responsible for.

But if AI has taught us anything, it’s that things change, often quickly. The clock is ticking. And if Microsoft’s AI defeats continue unabated, hard questions will be asked. And change will come.