China prohibits enterprises from purchasing H20 chips, Huang Renqian suddenly visits Taiwan power supply chains for solid Rubin

NVIDIA executive chairman Huang Renxian visited Taiwan without warning, bluntly saying that he was visiting Taiwan for mass production of Rubin chips. In addition to foreign media reports, China prohibited large technology companies such as Teng and Word Skipping from purchasing NVIDIA's China-specific H20 chips on the grounds of information security. NVIDIA also notified Amkor and Samsung to stop producing H20. Market insiders pointed out that the two things have certain relationships.

Huang Renyi landed in Taiwan early this morning and was interviewed by the media when he was off the machine. This time he visited the production partner Taiwan's power-produced Rubin chip. After heading straight to Taiwan, I left Taiwan in less than a day and did not arrange other suppliers' itineraries. It can be seen that NVIDIA's multi-visual tyre-electrical power-produced Rubin chips are produced.

NVIDIA Rubin chips will be mass-produced at the end of the year and will be sold in early 2026. However, there were previously noises in the market, but due to design problems, there may be delays. However, according to external data surveys, Rubin's schedule has not been delayed. The first Rubin chip is expected to be tested in Taiwan's transistor round factory in October. Even if the chip design continues to be micro-tuned, the engineering products are still expected to be completed as scheduled in the fourth quarter. The chip and system design will be finalized in March 2026, and the production will be in the second quarter, and the server rack will begin to be shipped in the third quarter. Therefore, at the key point of preparing for production, Huang Rensha’s visit, it can be seen that the NVIDIA Rubin architecture chip is only allowed to succeed and not fail.

In April, the US government notified NVIDIA that H20 chips need to obtain special export permissions before they can be sold to China, and their sales are actually prohibited. NVIDIA's sales in the Chinese market amounted to about US$17 billion, accounting for about 13% of its total revenue, so the market is sensitive to NVIDIA's loss of the Chinese market. In response, Huang Rensha had previously said during an interview with the media that the H20 chip was exported under the control of the United States and that NVIDIA would claim a loss of up to US$5.5 billion (about NT$16.59 billion), which made him deeply painful. Therefore, NVIDIA later tried to say that the Trump administration opened sales and knew that it had previously sold 15% of the Chinese business to sell H20 to the Trump administration, which was the sales permit.

Just as NVIDIA has the conditions to obtain a license for selling H20 in China, the Chinese market has reported that the Beijing government has banned enterprises from purchasing H20 chips. The reason is that in addition to the previous official doubt that the NVIDIA H20 chip can fully realize the "far-closing" function, including setting voltage, fixed time and other conditions to automatically close the chip, doubting the situation of its installation door, and having information security doubts, it is also because of how the US Secretary of Commerce Howard Lutnick We will not sell our best products to them, we will not sell the second best, or even the third best. It also emphasized that it is necessary to sell many (products) to China, so that their developers will rely on the technology stacking in the United States. This is a comment from the thinking behind it, which angered Beijing's high-level, and therefore decided to ban Chinese technology companies from purchasing H20 chips.

As Huang Renfeng said before, the H20 chips were exported under the US control, and NVIDIA must claim a loss of up to $5.5 billion. Now that China boycotts the purchase of H20 chips, the single sales loss on H20 has already made Huang Ren's head hurt, and will it not expand to other projects in the future. Even if NVIDIA makes every effort to prevent blockage, the development of the matter is difficult to predict. Therefore, at this critical moment, Huang Renjun chose to visit Taiwan suddenly and go straight to the high-level of the Taiwan Electric Power Club to ensure that the mass production of Rubin architecture chips will not be lost and that additional risks cannot be added to the pressure of the H20 chip.

Rubin architecture chip tyre 3-N3P process, and is built with CoWoS-L with a packaging surface of four times the size of the Reticle size, and adopts a small chip architecture. This is very distinct from the traditional single chip GPU, with higher performance, scalability and cost efficiency. Small chips allow chip manufacturers to package multiple smaller semi-conductor chips into a single chip, increasing production volume and reducing production costs.

Semiconductors are becoming more and more popular, chip design is becoming more and more complex, and traditional process expansion limitations are becoming greater. The small chip architecture is made of telco CoWoS-L packaging process. NVIDIA can improve the energy efficiency and performance of the GPU, which will be very suitable for artificial intelligence, data centers and high-performance computing. Under this situation, NVIDIA also expects Rubin architecture chips to continue, even surpass the hot sales of the previous generation Blackwell architecture chips, to inject some time to face difficulties in the Chinese market. Therefore, it is the consideration that we will inevitably come to Taiwan to visit Taiwan's electricity solid foot for this level.