I'm increasingly baffled by the inconsistencies in how the administration views AI.
On one hand, it's the most important technology of the century, and will definitively decide the outcome of the U.S.-China competition. Yet we're willing to outsource and offshore data center development. I realize the analogy is flawed, but it has the same feeling as saying we want to outsource production of the B-21 because we have too many regulations, it will take too long to get permits and build, we can't afford it, and BTW we'll be able to provide oversight and monitoring of the manufacturing plant. Not something we would do in a million years.
So if AI is that critical, why would we ever agree to this chip sales/data center deal--purported to result in what will become the largest cluster in the world? (While cutting BIS resources.)
Rather than being about national self-interests, this has all the earmarks of a pure transactional agreement, centered on money and power.
And while it's easy to be skeptical of the claims about AGI and ASI, if the assertion is that AI is going to be the most powerful, game-changing tech of the century, why should we ever stop states from trying to implement reasonable oversight and regulations? The states are doing it because there's a major gap at the federal level. The hyperbolic claims about over-regulation of AI during the previous administration are belied by the facts on the ground. In reality, the previous EOs directed very little regulation, outside mandating a number of actions within the government itself. I'm all for smart, tailored, less intrusive regulations. But it's really hard to justify scant or no regulatory actions, at the same time the administration is arguing that this technology will forever change the world.
Call me very naive, but it seems to me that the strategy to win the AI war against China will be to build the most compute-demanding and powerful foundation models in a single highly-secured US-based data center with dedicated power supply (off-grid) and fully isolated data paths, built and operated via a public-private project (i.e., "Manhatten-style"). Then, the more "routine" data centers for b-to-b and b-t-c AI services can be serviced via a geographically-diversified data center deployment strategy with various levels of security as appropriate to the product being sold. I think this is what Sacks has in mind. And he knows perfectly well that you can't simply count server racks, and just likes to say that because he doesn't want to spend his time explaining the details of what it would really take to describe the actual assessment to non-specialists (and, much of that planning detail is classified in any case).
As an Australian this deal reminds me of the AUKUS submarines Australia has purchased but will probably never get. Big money contracts where nothing actually happens for the next 10 years. Australia is expected to get its first nuclear sub in 2035.
if the Saudi's have not even build these AI labs then how does that help Invidia in the near future?
I suspect this deal, in the short term, is a deal designed to slow the fall of the NASDAQ. It will also allow for a back door so China can obtain the chips it needs although the fabrication tech is far more important then the chips themselves.
I may be naive but I don't believe Chinese investment would be attractive to Saudi and the UAE in this area because of the geopolitical backdrop and their defense dependencies- that is, for as long as China remains 'friends' with Russia and Iran.
I read the Jensen Huang interview and his argument was "they might become better than the US in AI but it's better if we help them be better than the US (somehow)"
Excellent commentary.
I'm increasingly baffled by the inconsistencies in how the administration views AI.
On one hand, it's the most important technology of the century, and will definitively decide the outcome of the U.S.-China competition. Yet we're willing to outsource and offshore data center development. I realize the analogy is flawed, but it has the same feeling as saying we want to outsource production of the B-21 because we have too many regulations, it will take too long to get permits and build, we can't afford it, and BTW we'll be able to provide oversight and monitoring of the manufacturing plant. Not something we would do in a million years.
So if AI is that critical, why would we ever agree to this chip sales/data center deal--purported to result in what will become the largest cluster in the world? (While cutting BIS resources.)
Rather than being about national self-interests, this has all the earmarks of a pure transactional agreement, centered on money and power.
And while it's easy to be skeptical of the claims about AGI and ASI, if the assertion is that AI is going to be the most powerful, game-changing tech of the century, why should we ever stop states from trying to implement reasonable oversight and regulations? The states are doing it because there's a major gap at the federal level. The hyperbolic claims about over-regulation of AI during the previous administration are belied by the facts on the ground. In reality, the previous EOs directed very little regulation, outside mandating a number of actions within the government itself. I'm all for smart, tailored, less intrusive regulations. But it's really hard to justify scant or no regulatory actions, at the same time the administration is arguing that this technology will forever change the world.
We're trying to create another Taiwan we have to protect lol
Call me very naive, but it seems to me that the strategy to win the AI war against China will be to build the most compute-demanding and powerful foundation models in a single highly-secured US-based data center with dedicated power supply (off-grid) and fully isolated data paths, built and operated via a public-private project (i.e., "Manhatten-style"). Then, the more "routine" data centers for b-to-b and b-t-c AI services can be serviced via a geographically-diversified data center deployment strategy with various levels of security as appropriate to the product being sold. I think this is what Sacks has in mind. And he knows perfectly well that you can't simply count server racks, and just likes to say that because he doesn't want to spend his time explaining the details of what it would really take to describe the actual assessment to non-specialists (and, much of that planning detail is classified in any case).
As an Australian this deal reminds me of the AUKUS submarines Australia has purchased but will probably never get. Big money contracts where nothing actually happens for the next 10 years. Australia is expected to get its first nuclear sub in 2035.
if the Saudi's have not even build these AI labs then how does that help Invidia in the near future?
I suspect this deal, in the short term, is a deal designed to slow the fall of the NASDAQ. It will also allow for a back door so China can obtain the chips it needs although the fabrication tech is far more important then the chips themselves.
I may be naive but I don't believe Chinese investment would be attractive to Saudi and the UAE in this area because of the geopolitical backdrop and their defense dependencies- that is, for as long as China remains 'friends' with Russia and Iran.
I read the Jensen Huang interview and his argument was "they might become better than the US in AI but it's better if we help them be better than the US (somehow)"