Artificial Intelligence thread

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Point One: The supply channels extend far beyond the four countries you mentioned (including those Middle Eastern states). This explains why the U.S. has now implemented comprehensive restrictions across all nations.

Point Two: Specific operational details cannot be publicly disclosed here, particularly given this forum thread's persistent high visibility in Google search results.

I can reveal this much: We're discussing a nine-figure transaction between parties with no prior business history, necessitating intermediary escrow arrangements and technical verification capabilities to bridge mutual distrust. The Chinese recipient imposed location constraints favoring jurisdictions with Chinese legal protections. Through understanding their non-negotiable terms and standard transaction protocols, I gained rare insights into such high-stakes operations. Ultimately, the deal collapsed due to unique technical complications (consider: How does one validate quality assurance for components priced at 20k−300k per unit? The intermediary's financial exposure became untenable).

Regarding forum participants' professional backgrounds - while identifiable, what purpose would detailing serve? Should we disclose:

Additional unpublicized channels beyond known countries?

The shadow-state counterparts in these transactions?

Suppliers' operational blueprints and channel networks?

Any explicit discussion here would essentially gift-wrap intelligence for U.S. Commerce Department containment strategies. Given America's expanding restrictions, future countermeasures will inevitably adapt within predictable parameters.

If my expertise seems questionable, so be it. I'll maintain discretion accordingly. Let me emphasize: My statements here carry no guarantee of 100% accuracy, as the information itself contains inherent ambiguities. This case represents a commercially unsuccessful experimental transaction - precisely the kind of adaptive innovation constantly emerging to circumvent new restrictions.

Yes, I understand that supply channel goes through many countries. That's quite obvious. But it's also well known that nearby East Asian regions are some of the more popular conduit for these trades. Realistically, the only thing US can limit this trade is just forbid Nvidia chips from being exported.

And as we discussed already, it's pretty much pointless. Since the majority of demand going forward is going to be inference and China should be putting more effort on developed specialized inference chips.

As for your other stuff, I'm not really sure why you are mentioning them.

How smuggling happens is not particularly relevant but it's just relevant to know that they are happening. And the volume is quite high. My Singapore post is just one illustration, because we suddenly went from having no Nvidia sales to Singapore to probably $25B last year after H100/800 ban got imposed. And I think the commerce department probably has a good idea of just how much smuggling is getting through. Hence the new rules.

I think if you don't know how to be discrete with what you are saying, you shouldn't be talking about it. I wasn't trying to get you to talk but rather telling you to shut up.

So stuff like how transaction happens, this is not relevant. Some people might care, but people really just want to know that China has enough AI chips. I've been telling people "relax, they have enough AI chips". That's all people really need to know.
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member

huemens

Junior Member
Registered Member
Letter from US lawmakers to Waltz, requesting more chip restrictions
Please, Log in or Register to view URLs content!
People’s Republic of China (PRC) AI firm, DeepSeek, released a sophisticated model that made extensive use of Nvidia’s H20 chip, which is currently outside the scope of U.S. export controls. We ask that as part of this review, you consider the potential national security benefits of placing an export control on Nvidia’s H20 and chips of similar sophistication while cracking down on chips designed specifically for AI inference, not just training.
Some are already using DeepSeek’s latest model to cast doubt on the effectiveness of U.S. export controls, an argument we respectfully reject. DeepSeek’s own founder, Liang Wenfeng, acknowledged that the U.S. “embargo on high-end chips” was a major roadblock to the company’s development. Furthermore, DeepSeek made extensive use of Nvidia’s H800 chip, the first chip that Nvidia designed specifically to fall outside U.S. export controls.
Third, as part of the review, we ask that you look for ways to strengthen controls on shipments through third countries that pose a high risk of diversion. For example, Singapore represented 22% of Nvidia’s revenue in its most recently quarterly statement, despite the company itself revealing most of these shipments ultimately went to users outside of Singapore. Countries like Singapore should be subject to strict licensing requirements absent a willingness to crack down on PRC transshipment through their territory.
 

Temstar

Brigadier
Registered Member
don't make me laugh. There is like nothing that gets posted on this thread that the commerce department doesn't access to.

anyhow, my point about inference. More high speed inference server hosting R1
How does AI specialized chips work? Is it similar to how there were ASIC mining chips and because their logic was implemented in hardware they were faster than GPU at the one task they were designed to do?
 

9dashline

Captain
Registered Member
This is the true genius of making it open source. It enables countries who are not at the forefront of AI to have a fighting chance. The amount of geopolitical good will from this cannot be understated.
Anglo's GATEkeeping of the AI STARS and evil plans of global External AI Revenue Tax schemes just went up in smoke....

"Open"AI was hoping to milk that $200/mo into perpetuity but alas 2 month later DeepSeek dropped a hydrogen bomb on that....

If I was an Anglo/Jewish elite I would be pissed too. Their carefully crafted world domination would have worked had it not been for pesky Chinese and some quant geeks who decided to embark on a small side project.....
 

tokenanalyst

Brigadier
Registered Member
don't make me laugh. There is like nothing that gets posted on this thread that the commerce department doesn't access to.

anyhow, my point about inference. More high speed inference server hosting R1
I still think Advanced Packaging and low level optimizations will beat the slowing and expensive dimensional scaling at least for a while. A good design 7nm or even a 28nm large scale integrated circuit could deliver good AI speeds and performance a at "cheaper" cost. People forget is not just the machines and shiny CFET transistors, is about how optimized is your design. And more considering that 28nm is dirt cheap right now.

Yes looks like 3DSEMI do Wafer Level Packaging

View attachment 128302
The question is "large chip" means Wafer Scale Chip?

View attachment 128303

Zhejiang Lab is making really good progress on large scale integration.​


12 Inch Wafer Scale Chip:

View attachment 139840

4 Inch Large Scale Wafer "Packaging": I imagine that they built connections and additional circuitry direct in the wafer.

View attachment 139841

First time seeing this, a Software Defined System of Wafer:

View attachment 139842

tches, which limits the injection/local performance and introduces latency/energy/cost overhead. The new wafer-scale packaging and high-speed wireline technologies provide high-density, lowlatency, and high-bandwidth connectivity, thus promising to support direct-connected high-radix interconn
 

Temstar

Brigadier
Registered Member
As others have pointed out, R1 is much more eloquent in Chinese than English, if you ask for it to write up a passage trash talking ChatGPT in Chinese you can get bangers like this:

69e3d825gy1hy3d4jwhirj20u01t1ahp.jpg

005R9eTagy1hy3cw8ntt1j30u01v347g.jpg
Somehow it has conceptualized generalising humans as "my carbon based friends" and by that extension AIs as fellow "silicon based lifeforms". So it was able to come up with that creative insult of "ChatGPT? His mum is probably smeared in some server as silicon based thermal paste". Surely you can't just find that insult by scrapping Chinese internet?
 

coolgod

Colonel
Registered Member
As others have pointed out, R1 is much more eloquent in Chinese than English, if you ask for it to write up a passage trash talking ChatGPT in Chinese you can get bangers like this:

View attachment 144700

View attachment 144699
Somehow it has conceptualized generalising humans as "my carbon based friends" and by that extension AIs as fellow "silicon based lifeforms". So it was able to come up with that creative insult of "ChatGPT? His mum is probably smeared in some server as silicon based thermal paste". Surely you can't just find that insult by scrapping Chinese internet?
I think this just show the superiority of Chinese netizens over Western netizens. Garbage in Garbage out.
 
Top