Sources at Ctee reported that the B300 will be built into Bianca compute boards, featuring one CPU and two GPUs. Mass production is expected to begin before the end of the year, with analysts predicting a boost across supply chains including TSMC, Machtech, Inventec, Chipset, Quanta, Wistron and Foxconn.

Nvidia’s B300 is intended to fill the supply gap left by the now banned H20 GPU in China. Ctee said that a shipment from Nanya Advanced Packaging was noted in early April to support CoWoS-L packaging for the new chips.

Strong customer demand has forced TSMC to ramp production quickly. Nvidia chief scientist Bill Dally recently said at TSMC’s North American Technology Symposium that the B200 AI chip uses CoWoS packaging to combine two GPUs, overcoming the traditional single reticle size limit.

Dally said: “The B200 uses CoWoS advanced packaging to package two GPUs, breaking the limitations of the single reticle limit size.”

Semiconductor analysts say TSMC’s efforts to push advanced packaging technologies are key to moving beyond Moore’s Law, packing ever more transistors into larger chip packages.

Nvidia is expected to unveil full details of the B300 at Computex 2025 next month, alongside its teased Nvidia and MediaTek Arm-based AI PC processor. Robotics technologies and other products are slated for launch at the event.