NEW YORK, Feb 12, 2026, 06:34 ET — Premarket
Nvidia (NVDA.O) edged up nearly 1% ahead of the bell Thursday, with the stock priced at $191.87 after closing at $190.05 the day before. The move came on word from Samsung Electronics that it’s begun delivering its latest HBM4 memory chips—crucial for Nvidia’s AI accelerators. Samsung’s chip division head, Song Jai-hyuk, described customer feedback as “very satisfactory.” Reuters
High-bandwidth memory, or HBM, stacks DRAM right up against the processor, shuttling data at speeds needed by expanding AI models — but it’s a tight spot for supply. Two sources told Reuters TikTok parent ByteDance is working on its own AI chip and has been in discussions with Samsung, looking to nail down memory that’s “in exceptionally short supply.” ByteDance pushed back on that version of events. Reuters
Nvidia faces another immediate wild card in the form of U.S. export rules, as it tries to comply with licensing mandates on certain chip shipments to China. Commerce Secretary Howard Lutnick described the licensing for Nvidia’s H200 chips as “very detailed. They’ve been worked out together with the State Department, and those terms Nvidia must live with.” Speaking to the Senate, Lutnick added: “We agree that demand by American firms is vast for these chips, and we don’t want to do anything that stands in their way.” Reuters
Representative Ro Khanna, the ranking Democrat on the House Select Committee on China, said he’s willing to consider exports of Nvidia’s older Hopper-generation chips, but that’s where he draws the line. “We certainly shouldn’t be sending them Rubins. We shouldn’t be sending them Blackwells,” Khanna told reporters. Committee chair John Moolenaar put it bluntly: “Chinese AI companies are training their models on U.S. chips then giving these models to the PLA.” Reuters
Separately on Thursday, Samsung said it’s begun mass-producing HBM4 chips, touting steady data transfer at 11.7 gigabits per second and potential to hit 13 Gbps. Sang Joon Hwang, executive vice president for memory, pointed to “the most advanced nodes like the 1c DRAM and 4nm logic process” used for HBM4. Samsung Global Newsroom
Taiwan Semiconductor Manufacturing Co booked January revenue at roughly NT$401.26 billion, marking a 36.8% jump year-on-year. The number was also up 19.8% from December. For investors, that monthly figure tends to serve as a quick gauge of supply chain momentum tied to AI chips.
Vertiv reported its quarterly backlog soared 109% year-over-year, reaching $15.0 billion, a surge the company attributed to demand fueled by AI infrastructure. Those figures carry weight for Nvidia, since such orders tend to reflect how quickly data centers are ramping up power and cooling to handle GPU-intensive servers.
The story could change fast. Major clients leaning further into their own chips—or a regulatory squeeze on Nvidia’s China sales—would make recent supply gains less relevant. Margins might still get dented if memory prices don’t let up.
Nvidia is set to release its fourth-quarter numbers Feb. 25, 2 p.m. PT, per the investor relations calendar. Wall Street’s likely to focus on any new guidance around AI data-center demand, memory supply, and updates on China licensing.
