Nvidia's New Chip Tech: Connecting Chips for Faster AI

So, Nvidia (NVDA.O) just dropped some cool news, like, yesterday. They're gonna sell a tech thingy that basically glues chips together to make them talk faster. This is a big deal for building and using all those fancy AI tools everybody's buzzing about.

They're calling this new version of their tech NVLink Fusion, and they're gonna let other chip makers buy it. This way, those companies can build their own super-powerful AI systems by linking up a bunch of chips. It's pretty neat, right?

Apparently, companies like Marvell Technology (MRVL.O) and MediaTek are planning to jump on this NVLink Fusion bandwagon for their own custom chips. Nvidia actually came up with the original NVLink ages ago, and it's already used to move tons of data between different chips, like in their GB200, which is like a mashup of two graphics chips and a main processor.

Jensen Huang, the boss man at Nvidia, spilled the beans at this big tech show in Taiwan called Computex. He was talking about how Nvidia got into the whole AI chip thing and all the software they've built to make it work. He even joked that back in the day, his talks used to be all about graphics chips, but things have totally changed.

Now, Nvidia isn't just about making video games look awesome anymore. They're seriously the go-to company for the chips that are powering this whole AI craze since ChatGPT showed up. You know how Microsoft (MSFT.O) computers use that Windows stuff? Well, Nvidia's been working on chips for those too, using tech from Arm Holdings (O9Ty.F), which is kinda interesting.

Remember last year at Computex? Jensen Huang was like a rockstar in Taiwan! People were seriously following him around and getting all excited. It was kinda wild, they even called it "Jensanity" (https://www.reuters.com/technology/like-pop-star-nvidias-ceo-huang-stirs-up-jensanity-taiwan-2024-06-05/).

Back in March, at their big developer shindig, Huang talked about how Nvidia was gonna keep up with the shift from building huge AI models to actually using them in everyday stuff (https://www.reuters.com/technology/artificial-intelligence/nvidia-expected-reveal-details-latest-ai-chip-conference-2025-03-18/). He showed off some new AI chips, like the Blackwell Ultra, which is coming out later this year. And get this, after the Rubin chips, they're planning even more advanced ones called Feynman processors way out in 2028! They even launched a smaller version of their AI chips for people doing AI research, called DGX Spark. Huang said those are being made right now and should be ready in just a couple of weeks.

Computex is a pretty big deal with like 1,400 companies showing off their stuff. It's the first time a bunch of computer and chip bigwigs are getting together in Asia since that time when President Trump was talking about putting tariffs on things to get companies to make more stuff in the US. So yeah, lots going on!

Hey, if you're looking to get some tech work done or maybe find someone to help you with some of this AI stuff, you could totally check out Fiverr. It's a cool platform where you can find freelancers for all sorts of things! Just a thought.