Everyone knows Nvidia for its GPUs. Fewer people know that the company has been quietly building a networking business that brought in over $31 billion last year and grew 267% year over year in its most recent quarter alone. That is $11 billion in a single quarter from a division that barely gets mentioned in the same breath as the chip business that made Nvidia famous.

The networking division exists to connect the data centers that run AI. When you are training a massive language model, the chips doing the computing need to talk to each other constantly, and doing that efficiently at scale requires serious networking infrastructure. Nvidia sells exactly that, through products like NVLink, which handles communication between GPUs on a data center rack, and Spectrum-X, its ethernet platform built specifically for AI workloads. Together these products form what Nvidia calls an AI factory, a complete data center setup designed for training AI models at scale.

The roots of this business go back to 2020, when Nvidia acquired Mellanox, an Israeli networking company, for $7 billion. At the time it was not entirely clear why. By owning the networking layer alongside the compute layer, Nvidia can sell customers a full package rather than just chips that then need to be connected using someone else's hardware. One analyst put it plainly: Nvidia's networking division does in a single quarter what Cisco's entire networking business does in a full year.

What makes the division even more interesting is how it goes to market. Nvidia does not sell the networking tech directly to end customers. It sells through partners, and it only offers the full stack rather than individual components. That approach has helped it build deep relationships throughout the data center supply chain without competing directly with the companies it relies on to distribute its products.

At GTC 2026, Nvidia announced a fresh round of updates to the networking platform including the Rubin platform, new memory storage systems, and more efficient ethernet switches.

Meta was planning to pull Horizon Worlds from its Quest VR headsets. Then someone reached out to the company's CTO Andrew Bosworth on Instagram saying they were heartbroken about it, and Bosworth reversed the decision the same day. Just like that, Horizon Worlds stays on VR.

That sequence of events tells you a lot about where the metaverse currently stands.

Earlier this year Meta had confirmed it would move Horizon Worlds to web and mobile only from June 15, effectively ending VR support for the social platform that was once central to Mark Zuckerberg's entire vision for the future of the internet. The reversal is good news for the small but apparently vocal community of users who actually use the app in VR. But the fact that the shutdown was planned in the first place is hard to spin as anything other than an admission that VR socializing never really took off the way Meta hoped.

The numbers back that up. Quest headset sales were down 16% from 2024 to 2025. Meta's Reality Labs division, which covers VR, AR, and some AI research, has lost $73 billion since 2021, the year the company rebranded from Facebook to Meta. In January, Meta cut over 1,500 Reality Labs employees and shut down several game studios. Reports now suggest another round of layoffs could hit as much as 20% of the broader company.

Meanwhile, Horizon Worlds on mobile is actually doing reasonably well in terms of downloads, with 45 million total worldwide installs and a 53% year over year growth in downloads so far in 2026. But consumers have spent just $1.1 million on the app in total, which is a very small number given the scale of investment behind it.

Anthropic has rolled out a Projects feature for Claude Cowork, the desktop tool that lets users automate tasks and run Claude directly from their computer. The update gives Cowork users a way to group related tasks into dedicated workspaces, each with its own files, instructions, memory, and context. Think of it as a focused environment for a specific area of your work, separate from everything else you have going on.

Setting up a project is straightforward. You can start one from scratch, import an existing Claude project from the chat interface and bring its files and instructions over automatically, or simply point Cowork at a folder already sitting on your computer. Each approach takes just a few clicks and gets you into a working environment quickly.

What makes projects genuinely useful is the memory piece. Claude remembers what has happened within a project and carries that context forward into future tasks in the same workspace. So if you have given Claude specific instructions, shared relevant documents, or built up a working pattern over time, it does not have to start fresh every session. That memory stays scoped to the project it belongs to, so nothing bleeds across into unrelated work.

Each project can also hold scheduled recurring tasks, custom instructions for how Claude should behave, and context pulled from a local folder, a linked chat project, or a URL. For anyone running repetitive workflows, the scheduling feature alone could save a meaningful amount of time each week.

The feature is available now for paid plan users on the latest version of Claude Desktop for both macOS and Windows. A few limitations are worth noting: projects currently work in Cowork only and not in Claude Code, there is no cloud sync yet so everything lives locally, and project sharing is not supported for Team and Enterprise plan members at this stage. Anthropic says Claude Code support is planned for a future update.

More than a decade after the Fire Phone quietly faded into tech history as one of Amazon's most expensive failures, the company is reportedly working on a new smartphone. According to report, the device is codenamed Transformer and is being developed by a relatively new internal team called ZeroOne, which sits within Amazon's Devices and Services division and is led by J Allard, the former Microsoft executive who helped create the Xbox.

The Fire Phone launched in 2014 and was pulled from sale within a year. It was widely seen as a product that solved no real problem, loaded with features nobody asked for and priced against competitors it had no business going up against. Amazon took a massive write-down and walked away from phones entirely. So why try again?

The answer, by all accounts, is Alexa. The new device is being designed around Amazon's AI assistant, which the company spent over a year completely rebuilding with generative AI before relaunching it as Alexa+ in February. The upgraded Alexa can now do most things you would expect from a modern AI assistant, from planning trips and updating calendars to making recommendations and helping with homework, while keeping its original strength as a smart home controller. Amazon sees a phone as a natural hardware home for all of that, and a way to pull more customers deeper into its ecosystem of shopping, Prime Video, Prime Music, and now AI.

The timing makes sense given where Amazon is putting its money. The company recently put $50 billion into OpenAI as part of that record funding round, and is projecting $200 billion in capital spending on AI, chips, and robotics in 2026 alone. A smartphone built around Alexa would give Amazon a direct line to users throughout their day, something that Echo speakers and Fire tablets have never fully delivered.

AI-first pull request reviewer with context-aware feedback, line-by-line code suggestions, and real-time chat.