USA Vs China: Who Will Win The Gen AI Battle?

 

With all the political turmoil that is happening today, the news headlines do not seem to be coming out as quickly about Generative AI as it once did, say, going until the end of last year.  The biggest fear is China, not just from the standpoint of tariffs, but also in terms of competition. 

In fact, if you recall, they came out with something remarkable like ChatGPT.  It was developed by a company called Deep Seek, and the cost of running the algorithms and the hardware needed (such as the GPUs) is much lower.

Also,  Nvidia took a decent hit with a financial charge of over $5  billion,  with the restrictions that have been put into place on sending GPUs to China.  But despite all this turmoil, there is yet another headwind that both produce and make use of Generative AI must contend with:  Data Privacy, and Compliance that comes along with it.

As I have written before, the fuel that runs Generative AI models are the datasets that are fed into it.  Not only do they need it to train, but they also need them to create the output you are seeking when you ask  it a specific query.  Generative AI Compliance will come across three different  angles:

Making sure that the right controls have been implemented on the training datasets.

The same with the above, but for the output that has been generated.

Also, the same with the above, but making sure that any data which is submitted by the end user  is  also as secure as possible.

To this end, the trends for this year are expected to be as follows:

1)     Efforts From The EU:

They have produced a new piece of legislation called the “NIS2”.  It is an acronym that stands for the “Network and Information Security”.  Just like the GDPR, it applies to any entities that conduct business in the EU, even if they are not physically located there. The tenets and the provisions are almost similar, but they also take a strong stance to Generative AI.  But, the financial penalties are very harsh for non-compliance:  It can be up to 2% of the revenue that has been generated on a global basis.

2)     The DORA:

This is an acronym that stands for the “Digital Operational Resilience Act”.  It was created and enacted by the EU as well.  But apart from Generative AI compliance, it has two key specific focuses:

Ø  Proving that you cannot just only create backups, but that you can also restore the mission critical data from them, if you are ever impacted by a disaster, natural or man-made.

Ø  That the backups which have been created are segregated in terms of the physical and logical ones.  The goal here is to make sure that businesses are storing their backups in various locations, such as On Premises or in the Cloud.

 

3)     More From China:

Take for example, that you have a hosting account with a domain registrar that is located here in the United States (such as GoDaddy, Namecheap),  etc.).  You decide to host your application in a datacenter that is in the US.  Although this may be technically correct once you launch your web application, the datasets that it uses could be stored at a datacenter in entirely different country that you may not even be aware of.  So, the hot topic of debate here is who takes custody of it?  Well, the Chinese Government is making this even clearer now, especially when  it comes to Generative  AI.  To this extent, they have passed two distinct laws:

Ø  The Personal Information Protection Law ( also known as the “PIPL”).

Ø  The Data Security Law ( also known  as the “DSL”).

Ø  The Cybersecurity Law (also known as the “CSL”). 

The result is that China is now relaxing its restrictions on storing “foreign datasets” on the datacenters that are located there and are now highly encouraging businesses from all  over the world to even put there backups as well. 

4)     The Rise of E2EE:

This is yet another acronym that stands for “End to End Encryption”.  Encryption has always been a favored tool in the arsenal to protect anything data related.  After all, it scrambles it so that if anything was to be intercepted by a third party, there is nothing  that they can do with it unless they have the appropriate key to decode it.  But with the E2EE, the IT Security team will have no choice on what can be encrypted, by default, everything will be.  While this is heavily targeted towards  the Generative AI algorithms and the datasets they use,  using the E2EE can be a bad thing as well.  For instance, even a Network or Database Administrator with the right permissions can be denied access.

My Thoughts on This:

On a theoretical level, all of this sounds great, taking more steps that the datasets that Generative AI use are now even more protected.  But in the real-world sense, just how enforceable is all of this?  Normally in a world where there is not much chaos or confusion, this all could very well be done.  But once again, given the political climate that we now have in the United States, who knows how this will all come together.

Then there is the issue of China.  They are the second largest economy in the world, and in fact, their manufacturing and supply chain logistics far surpass that of the United States.  For example, we are still trying finish construction on the next Ford class aircraft carrier, the “USS John F Kennedy”. 

During this time, the Chinese are already working on I believe, their third carrier, which would be quite compatible.

So, there are still many complexities and uncertainties which lie ahead because of these tariffs.  But one thing is for sure in this regard:  Given their sheer dominance, I bet they will far  outpace the United States when it comes to Generative AI development and production.  Not only can they do it faster and cheaper, but the quality  in the end may prove to far superior in the end.