Nvidia has been summoned for a talk. This matter may be more serious than people think.

In the past two days, NVIDIA has been in an awkward position.

Just recently, Jensen Huang came here and publicly announced that NVIDIA could sell H20 graphics cards to the Chinese market. Finally, the graphics cards worth billions of dollars could be sold.

Shortly afterwards, NVIDIA was summoned by relevant authorities due to the security risks of the backdoor in the H20.

When I first heard this news, I thought it was just groundless rumor. But when I saw that major official media were all reposting it, I realized that this was definitely a serious matter.

So, I did some research and really dug out the big story behind it. To sum it up in one sentence: Jensen Huang can really blame the US government for NVIDIA being summoned.

I guess Jensen Huang is so angry now that his leather jacket can't contain his rage.

Let's first talk about what exactly happened?

Although the information reposted by the official media is not long, it contains a huge amount of information, especially the following two sentences I marked for you ↓↓↓

First of all, our official media actually put it a bit euphemistically.

In fact, US congressmen are not only calling for advanced chips exported by the US to be equipped with "tracking and positioning" functions, but also directly writing this requirement into the ongoing bill, namely the "Chip Security Act".

Although it's not new that the US government secretly implants backdoors in chips and software, it's the first time in my impression that they so blatantly expose such underhanded means through legislation.

Those guys really don't care about their reputation anymore...

Actually, the proposal of this bill is also related to the popularity of DeepSeek at the beginning of this year.

Maybe because DeepSeek performed too well and challenged the perception of US government officials, they issued a security report on DeepSeek in April this year.

The report cited a survey by the social media research institution Graphika, saying that the model technology of DeepSeek might have been developed using banned graphics cards.

In other words, the US government would rather believe that we developed advanced AI using smuggled graphics cards than believe that we bypassed the computing power limitation to some extent through innovative technology.

So they focused on the point that "advanced graphics cards were smuggled into China" and rushed to promote various chip security bills.

In April, a US congressman proposed the "Stop Stealing our Chips Act".

This bill is as straightforward as its name. It encourages the public to report the smuggling of high-computing-power US chips to China by setting up an informant incentive mechanism.

In May, Democratic congressman Bill Foster proposed the "Chip Security Act".

Compared with the previous version's informant mechanism, this version strengthens the technical restrictions.

The core of it is to implant location verification technology in chips to check if the chips flow into restricted sales areas. And manufacturers must provide evidence of location transfer to relevant authorities. Once a chip is found in a restricted area (such as China), the manufacturer has to remotely lock it.

That is to say, if this bill is truly legislated and implemented, every chip using US technology will wear an "electronic shackle" that can be remotely controlled.

Bill Foster was able to propose such a bill because he is a technical expert. He was originally a particle physicist and understands chip design and AI application development.

We guess that the US AI expert mentioned in the official media news probably refers to him.

As some previous reports have mentioned, Bill has openly pointed out that most of the positioning technology is already built into NVIDIA's current chips.

Then why did we summon NVIDIA at the end of July when the incident happened in May?

I guess it's not because we wanted to give Jensen Huang face when he came to China for exchanges. It's because in June, the US government held a meeting to discuss why the US must lead in the AI field.

After the meeting, they decided to promote the "Chip Security Act" again in July, aiming to further suppress our AI industry. Although the bill is still in the proposal stage, it has already received bipartisan support from both houses of Congress.

That is to say, there is almost no obstacle from the two major US political parties in the process of this bill becoming law.

So, at this time, it's a correct and reasonable thing for our National Cyberspace Administration to summon NVIDIA.

But don't simply think that our action is just to give NVIDIA a warning.

After I investigated the technical principle behind it, I found that the situation may not be as optimistic as we thought.

First of all, different from what many people think, this bill neither uses built-in GPS in chips for tracking and positioning nor uses software technology like "geofencing" to define a monitoring area.

Instead, it uses a mature technology that has been commercially available for many years and is not easy to crack - Ping-based positioning.

As for the reason, in the article "Location Verification for AI Chips", the author (Asher Brass) gave the answer:

1. Built-in GPS is easy to crack, and those already sold AI chips don't have built-in GPS receivers.

2. Geofencing requires defining an area where US AI chips can be used worldwide and then disabling chips outside the area. The workload would be very large and it would be more difficult to implement.

So, what is Ping-based positioning?

You can understand it as a device (such as a mobile phone or computer) sending a message to another device (such as a server) and then receiving a response message from the other side (a round-trip of the signal).

With this feature, we can calculate the distance using the signal response delay time:

Assume that it takes 1 millisecond for device A to send a "Ping" to server B and get a response. And we know that the transmission speed of network fiber is equivalent to the speed of light, which is about 300,000 kilometers per second. Then we can calculate that the round-trip distance of the signal is about 300 kilometers. Divide this by 2, and we can get that the one-way distance between the device and the server should be 150 kilometers.

Of course, just calculating the distance between two devices is not enough to complete the positioning. More devices need to be involved to lock the location.

For example, to lock the location of device A, we at least need to calculate its distances to server C and server D respectively, and then use the triangulation method we learned in the second grade of primary school to get the approximate location.

Based on this technical principle, the "Chip Security Act" requires establishing trusted landmark servers at known locations, such as servers B, C, and D mentioned above. Then, AI chips need to send verification information to these landmark servers with their built-in keys to verify their identities and then calculate the approximate location information.

Some of you may ask: What if I don't connect to the Internet? Can the US still know where my graphics card is?

Well... According to the requirements of the bill, they will force all AI chips to send verification information to the landmark servers. So your graphics card may not work after disconnecting from the Internet.

After understanding the principle, I believe you have realized that this technology can be completely implemented through software and can also be used on old graphics cards.

An engineer team has already implemented a preliminary version of location verification on the H100 graphics card. The following screenshot demonstrates that a prototype landmark server in Singapore can verify if an AI chip in Singapore is within the pink coil area.

For us, the risks behind this are really hard to predict...

Then, for Jensen Huang, what price does NVIDIA have to pay for the sudden idea of the congressmen?

In the article mentioned above, the author has calculated the cost for NVIDIA and proposed measures that can be completed within six months (PS: There is a six - month buffer period from the legislation and promulgation of the bill to its actual implementation).

This includes firmware and software updates to enable AI chips to quickly perform location verification, with a cost of about $1 million; and building a network of trusted landmark servers near all countries and regions where smuggling activities may occur, with an annual cost of about $2.5 million to $12.5 million.

If I were Jensen Huang, I would have been cursing in my heart. They are adding so many unnecessary operating costs to me for no reason.

But the author of this article also said that the cost of a set of location verification systems can be recovered by selling just 500 more AI chips for Jensen Huang.

Jensen Huang must be even more angry. NVIDIA's China-exclusive H20 graphics cards are already in a difficult situation of "unsalable products", and the US government is still causing more trouble.

So, going back to the point we mentioned at the beginning, Jensen Huang is completely a victim in this incident. It's not us but the US government that is harming him.

He came to China happily to do business and find a market for the graphics cards. But the US government's actions have put him in an awkward position, and it will be more difficult for him to do business in the future.

<