OpenAI Unveils GPT-Live: The Voice Model That Actually Listens While It Talks
On July 9th, OpenAI dropped GPT-Live, their next-gen voice model built on a full-duplex architecture. What does that mean? It can actually listen and speak at the exact same time—just like a real human conversation. They’re rolling out two versions: GPT-Live-1 and the lighter GPT-Live-1 mini.
Meta’s Making Its Own AI Chips Starting September, Locks in Supply Deals with Samsung & SanDisk
According to a memo spotted by NUPIAO on July 9th, Meta is planning to start producing its custom “Iris” AI chips as early as September. They’re aiming to deploy 7 gigawatts of compute infrastructure this year, with plans to double that to a massive 14 gigawatts by next year.
To pull off this massive infrastructure expansion, Meta has already locked in several long-term supply agreements. These include memory chip deals with Samsung Electronics, flash storage agreements with SanDisk, and fiber optic equipment partnerships with Sumitomo Electric.
Starbucks is Using AI to Build Its Own Tools and Ditch Microsoft & IBM
Word on the street from NUPIAO on July 9th is that Starbucks is leveraging AI to build in-house tools. The goal? To replace a Microsoft inventory tracking system and an IBM maintenance management tool. If the tests go well, some of Starbucks’ homegrown software could be fully rolled out by the end of next year.
Meta Drops $10 Billion to Build Its Very First Data Center in Canada
On July 8th, Meta officially announced they’re pouring 13 billion Canadian dollars (around $10 billion USD) into building their very first data center in Canada. This is all about expanding their infrastructure to back up those ambitious AI goals.
According to Meta’s statement, this massive facility will be located in Sturgeon County, Alberta, and will boast a whopping 1-gigawatt power capacity. Meta is even funding a new power generation facility that will hook right into Alberta’s grid.
SK Hynix’s US IPO is Getting Crazy Demand: Over 7x Oversubscribed
Insiders told NUPIAO on July 9th that SK Hynix’s US IPO has been oversubscribed by more than seven times. The Korean memory chip giant is planning to price its shares this Thursday. Offering 177.9 million American Depositary Receipts (ADRs), the deal has attracted massive interest from global long-only funds, tech-focused funds, sovereign wealth funds, and global investors eyeing the Asian market. Of course, talks are still ongoing, so the final details of the US listing could still shift a bit.
Zhipu Raising a Record 3.14 Billion HKD: The Biggest Hong Kong Tech Placement This Year
On July 9th, Zhipu dropped a filing on the Hong Kong Stock Exchange, announcing a placement of new H shares under general authorization at a price of 1,588 HKD per share. If all shares are fully placed, they’re looking at a total haul of roughly 31.41 billion HKD, with net proceeds (after knocking off commissions and expenses) hitting around 31.375 billion HKD. This smashes the record for the largest single placement by a Hong Kong tech stock this year, and stands as the biggest post-IPO equity financing in the domestic large model sector.
Nvidia & Hugging Face are Teaming Up to Build Open-Source Robotics Models
Reported by NUPIAO on July 9th, Nvidia just announced a killer partnership with Hugging Face. They’re teaming up to develop open-source foundation models specifically for robotics. By merging Nvidia’s GPU ecosystem and CUDA stack with Hugging Face’s massive model library and developer community, they’re looking to drastically lower the barrier for AI training and deployment in the robotics world.
Tencent Cloud Officially Launches ‘Agent Bucket’
On July 9th, Tencent Cloud officially rolled out Agent Bucket. Think of this “bucket” as a dedicated cloud space capable of handling hundreds of millions of AI agents. It acts as a unified hub to store everything—from files you upload, to the reports, images, code, and other work artifacts generated by the agents themselves.
