The most telling set of numbers in tech tonight comes from Hon Hai Tech Day, which kicked off today.
First, speed: development of some products that takes others three months has been cut to three weeks by Hon Hai. Assembly of the compute tray for NVIDIA’s Vera Rubin platform has been reduced from over two hours to 20 minutes, thanks to a cable-free design and fully automated assembly.
Second, precision: high-speed rotating shafts spin at 40,000 rpm, with machining precision below one micron. The spacing error between optical-fiber tracks is kept within 0.3 microns, and the total error across 40 laid-out fibers is just 0.7 microns.
Third, partnerships: Chairman Young Liu revealed that he had met with OpenAI’s Sam Altman that morning, and that new collaborations are under discussion. Details will be announced once confirmed. He also said the industry is not in a bubble and still expects double-digit growth next year.
My take: In the second half of the AI race, the competition isn’t about who announces a model first, but who can actually build computing power and deliver it on time. Every hour saved in manufacturing means faster deployment. Where do you think the next bottleneck will be: chips, electricity, or the assembly line?
#AI算力 #Hon Hai Tech Day
First, speed: development of some products that takes others three months has been cut to three weeks by Hon Hai. Assembly of the compute tray for NVIDIA’s Vera Rubin platform has been reduced from over two hours to 20 minutes, thanks to a cable-free design and fully automated assembly.
Second, precision: high-speed rotating shafts spin at 40,000 rpm, with machining precision below one micron. The spacing error between optical-fiber tracks is kept within 0.3 microns, and the total error across 40 laid-out fibers is just 0.7 microns.
Third, partnerships: Chairman Young Liu revealed that he had met with OpenAI’s Sam Altman that morning, and that new collaborations are under discussion. Details will be announced once confirmed. He also said the industry is not in a bubble and still expects double-digit growth next year.
My take: In the second half of the AI race, the competition isn’t about who announces a model first, but who can actually build computing power and deliver it on time. Every hour saved in manufacturing means faster deployment. Where do you think the next bottleneck will be: chips, electricity, or the assembly line?
#AI算力 #Hon Hai Tech Day