Interesting that the DHS "Exclusion and Removal Order #2026-01" that was made public yesterday was actually released July 24th, 2026 and appears to have taken effect then. Agencies would have started pulling deepseek at that time. Which means it's likely DS became aware of the memo around then.
The release of this repo hours after the DHS order is a tactical move they started preparing when the DHS order was issued almost 3 months ago, and they timed publishing the repo right as the order hit the federal register.
DHS: "China bad and scary. Don't send them data."
DS: "You don't have to. Happy xmass kids."
Link to order: https://www.federalregister.gov/documents/2026/10/20/2026-21...
> It may take a minute or two for access for 90.8.226.75 to become effective
Exciting... feel like I'm back in '96, having to wait two minutes for the page to load.
> The release of this repo hours after the DHS order is a tactical move they started preparing when the DHS order was issued almost 3 months ago, and they timed publishing the repo right as the order hit the federal register.
What does DS achieve with this "tactical move"? Or, are you saying this is a tactical move from a higher level (Chinese gov interests)?
“ Or, are you saying this is a tactical move from a higher level (Chinese gov interests)?”
What country produces all of the ASICs used in contemporary bitcoin mining?
The faster we get to a good enough open model, the sooner it gets burned to hardware, and the sooner the transition to (relatively) low tech, low margin, high volume cards to run it on …
… and those are built in China.
Maybe they are assembled in China?
You're not going to make money buying a new 14nm bitcoin miner, and that is where the China foundries are for cost effective silicon. They're great at 28nm+ and good at 22nm pricing.
Most new design modern sub 7nm ASICs are made in Taiwan/Korea by TSMC and SLSI. That is where miners are profitable.
Whatever speculation you have on China’s foundry capabilities or a gap to TSMC will be wildly out of date within a year or 2. I’ll bet on it.
Go bet on it @=^D It's not speculation, it's what they sell to the world and build processors for globally available Huawei/Honor phones. It's not a secret.
If they had a solution to get to >90% yields on sub 5nm they'd be doing it now. They don't, even the 7-10nm isn't there. Best hope is they get to buy the entire world's supply of EUV in 2027, then they install all of them, then they bring them up, then they build processes around them and make masks, then they ramp the yield on 3-4nm by copying exact TSMC.
The hilarious thing is that TSMC couldn't do this at a new fab in Taiwan, with full access to equipment, supply chain, and people. So, nope.
What is "@=^D"?
Thinking on, I agree 1-2 years was too optimistic. Even if they have the tech right now already, building a fab itself within 2 years is tight, but I think doable. Maybe not at full capacity but at least functioning. It's possible they can start building the fab infra in parallel with acquiring EUV tech speculatively to speed it up.
Like the other person said, within 4-5 years is more reasonable.
give them 4-5 years, given the amount of knowledge leak between China, Taiwan and South Korea and the amount of research budget put in replacing ASML tools, it's gonna come faster than we expect for sure.
From that point onwards, we'll buy Chinese hardware or be left behind
To steel man the argument, I think that there really is a Moore's Cliff coming. I don't think that resolution is going to matter as much, although process/design knowledge will still be a huge barrier. I don't see even 10A being as big a node as 14A or 2nm. Too many things fall apart all at once, and the cyclical nature of semi will raise it's bloody head above the industry as the AI trade bursts. So many $10Bs will be spent and margins will plummet... again, just in time for China to catch up.
Already, the big capabilities are becoming multi-layer/die integration. Whether it's BSPDN, stacked die, optical interconnects, etc it won't be as simple as Denard scaling. It will be dependent on experience and supply chains, which are harder to transfer than equipment or IP.
I do think the market for inference will drive a lot of innovation, and a lot of that will be at larger nodes. A single reticle die masked ROM with CIM can do 7B at 6nm. At 2-3nm you can do 27-32B which is a single expert or router. At that point there's not much reason to shrink for AI since it's mostly 1-hot die per token (except for the router). Update the parameters every 3-6mo when a new frontier comes out, and take the cost/power savings to the bank.
That’s part of the equation but there’s almost certainly interest in the open models guiding information trends for the interests of the Chinese government as well. So not only to help their economy in a variety of ways, but to help undermine the US AI efforts in the process and put them finger on the scale of what models ultimately do output.
oh I love the thought that there is a future where I get my models burned in hardware same as I used to get games in cartridges
US wants to ban non us models from as many places as possible. This move enables them to get users in more places.
Attackers usually have a far easier time against defenders. The effort is in vain
Cheap inference from foreign countries is bad.
Cheap labor from foreign countries is good.
Can anyone explain how this is not hypocrisy?
making AI bubble burst (even if it is just frontier labs that got hit) while chinese companies keep making money off cheap AI would be massive economical hit to US
Only because the US went insane and tried to build too much, which will not be built and we'll see HBM repurposed into consumer gear
Any drawbacks?
I do not understand this. Was the order secret for all this time?
I think there is also an unlikely but reasonable concern that there could be exploits intentionally trained into the model itself.
https://news.ycombinator.com/item?id=50157302
Except this is also banned under the DHS order. The order bans "information technology products and services developed by Hangzhou DeepSeek Artificial Intelligence Co., Ltd", and more pressingly, "third-party products and services developed using DeepSeek implementations".
I noticed that. Vendors will have to certify that neither they nor their upstream suppliers use anything made with Deepseek, even if it's provably innocuous ('Deepseek write 'Hello World' and 5 implementations of fizzbuzz'). So basically this is the government leveraging its monopsony power rather than a real safety requirement. They'll probably do the same for Qwen, MM, GLM etc.
This is potentially quite effective at corporate levels but I don't see it being effective over the long term because there are too many small-medium users that don't need to care about supply chain certification.
Does that mean US labs are banned from using or even learning from this? Assuming they still want to be used by the government.
Something not being allowed to learn from hasn't stopped them before
It's not US labs that it targets, but just US gov't civilian agencies.
... so far.
This is also the word inside Defense sector contractors as well. No PRC origin anything - no code, OSS or no, no open weight models, no assistance from them.
No regs set in stone yet, but DoD doesn't need regs - they can tell you what the rule is and that's that.
They last part is impossible to enforce really without forcing audits on everything.
Not sure why you're being downvoted. It's very clear to me in the order that even the open weights would be forbidden, even served through vllm or sglang etc.
Agreed, though this appears to be much more than open weights.
Next would be just saying no Chinese inference engines.
Probably wouldn't have been many agencies using DS to begin with anyways.
I think the draft NDAA blocks running Chinese models on American hyperscaler hardware, but it's principally because they were using it to bypass chip export restrictions.
I don't think any Federal agencies are using DeepSeek inference, but many State agencies would be, primarily through third party services. Of more concern is the critical or financially important infrastructure which is enabled by Chinese models. Banking, ecommerce, retail.