|
Thanks for reading The Briefing, our nightly column where we break down the day’s news. If you like what you see, I encourage you to subscribe to our reporting here. Welcome back! Two months after Microsoft’s conscious uncoupling from OpenAI, Microsoft wants to prove it’s thriving as an AI provider that doesn’t have to rely on the ChatGPT maker’s technology. Microsoft’s annual Build conference for roughly 2,500 app developers, which kicks off Tuesday in San Francisco, will be a “debutante ball” of sorts for its team developing AI models that aim to be an alternative to those from OpenAI and Anthropic. As I reported last week, Microsoft is preparing to unveil a suite of new homegrown AI models for specialized tasks like transcription, image generation, reasoning and—crucially—coding. Microsoft wants to convince developer customers that the models are good enough to replace those from Anthropic and OpenAI for simpler tasks. The strategy mirrors that of Microsoft’s cloud rivals Google and Amazon, but swaying developers won’t be easy. It’s also a departure from Microsoft’s strategy with the Build event. Two years ago, Microsoft primarily used the event to talk about subscription AI features powered by OpenAI models in apps like Word and Excel, and how developers could purchase OpenAI models from its Azure cloud. Last year, Microsoft executives at Build emphasized new partnerships with Anthropic and xAI to add their models to its GitHub Copilot coding tool and to Azure. Such partnerships are as important as ever, and Microsoft will have the right to use OpenAI’s intellectual property for its own products, for free, until 2032. But Microsoft’s homegrown AI models are expected to get more airtime during this year’s build than in any of the company’s previous conferences. Microsoft has good reason to funnel its developer customers toward the new models. Its GitHub unit is facing stiff competition in the AI coding race from the likes of OpenAI, Anthropic and Cursor, and that race could undermine GitHub’s core business, not just its Copilot AI coding assistant. Microsoft can also drive better profit margins using its own models than paying for Anthropic’s. Notably, Microsoft reduced the number of people it invited to Build so it could focus more on app developers rather than consultants, resellers and corporate customers that have populated Build events in the past, GitHub Chief Operating Officer Kyle Daigle told me in April. Daigle said he also wanted this year’s Build to appeal not just to Microsoft loyalists but also to “people who are going to show up with MacBooks.” In other words, he said, Microsoft wants to win over new customers based on the strength of its developer tools alone. I also expect Microsoft to show off new tools for its Copilot AI to work in a more autonomous, around-the-clock way, similar to OpenClaw-powered AI agents. Microsoft is also preparing to simplify Copilot to unify various features—like coding tools as well as functions that automate work in Office 365 apps—in one app, which is also a plank of its effort to cut Copilot “bloat.” Whether developers will care remains to be seen. I’ll be there to find out, and to see how many Macbooks are on the convention floor. Jensen Huang is back in Taiwan, and the crowds have followed. The Nvidia CEO, who drew cameras at the night markets and dined with local executives, is there partly on a charm offensive and partly for a supply check, as shortages spread across the parts needed to make the AI servers and chips his firm designs. That pressure will change the conversations at Computex, Asia’s largest electronics trade show, which starts Tuesday. Previously, the AI hardware race was mostly about whose system was faster and more power-efficient. This year, the sharper question is whose designs can be built in volume and who has secured enough factory capacity to deliver them. The shortage is giving customers more reason to look beyond Nvidia. Google, Amazon and others have been building more of their own AI chips to lower costs and reduce dependence on one supplier. Computex will give that shift a bigger stage, as more companies show boards, servers and networking gear built for a world with more than one kind of AI processor. Nvidia’s response is to make its technology harder to avoid. That is why Huang’s planned appearance with Marvell CEO Matt Murphy is worth watching. Nvidia invested $2 billion in Marvell in March and made it a partner for NVLink Fusion, its high-speed technology for connecting chips inside AI systems. Marvell can help cloud companies build their own chips, while Nvidia works to make sure those chips still plug into its broader system. As AI systems get bigger, the network is becoming as important as the processor. Thousands of chips have to move data fast enough to act like one machine. That makes networking the next battlefield, a theme likely to show up in executive speeches and vendor prototypes. Nvidia will still set the pace for many of the technologies showcased at Computex. Among the most watched will be co-packaged optics, or CPO, which puts light-based connections next to the networking chip so data can move faster and use less power. Nvidia has already announced two CPO product lines, and Computex will show how quickly Taiwan’s server supply chain is starting to build around that future. • The U.S. Space Force awarded a $4 billion contract to SpaceX as part of a program to deploy space-based sensors to track airborne threats. • Meta Platforms plans to start testing an AI "pendant" in the next year as part of an ambitious roadmap for wearable devices aimed at reversing the huge losses in its hardware division.
| 
