While Silicon Valley take AI capex like an AGI moonshot, China's approach to innovation and emerging tech is much more methodical and long-term based. Some of the cultural and political differences actually do favor China.
More efficiency and innovation relative to less capital - more model optimization and innovation as a direct result of AI chip export controls and bans. Fairly predictable trajectories tbh.
Each year the argument for technological primacy gets more tenuous. The correct answer for a long time, is to increase the quality of the people. Instead the US system avoids this at all costs.
Long gone is any confidence that the neolib managers of the nation's wealth and power are interested in sharing its benefits. A common expectation in much of the West is these tools will be used as a more efficient way to exploit the people and enforce control.
The advent of AI actually is a vital subject, in my opinion. Seeing some of the latest advances, I'm starting to come around to the immense transformative potential of AI. But tech as a rule is readily duplicated, so chances are distant of AI plugging the geopolitical-power gap. I.e. the dream is fallacious, of "fixing" the slower US and G7 rate of development. That gap is fundamentally due to the West - and US especially - having first wasted and later atrophied much of its human capital.
But given the ability to combine parallel calculation in massive centralized servers, ability to reason, and naturally amazing powers of pattern matching, AI will nonetheless enable great transformations. It seems particularly well matched tool to create something akin to the "panopticon". That was once the dream of high modernism. Now rightly seen by most as dystopian, a "worst practices" pattern in the realm of thinking about how to structure society.
Question people might want to ask is, do we really want to have this technology emerge under a combination of (1) proto-fascist Trump administration (2) egged on by believers in the high-modernist panopticon vision of humanity such as Thiel or Musk - and (3) that techno vision paired with the boundless greed philosophy, exemplified by price gouging basic medicine like insulin, because hey, if your customer doesn't give you all their money they're free to die aren't they? Is that 3-way power combination really going to be the one to oversee deployment of AI in the US? When they fail to beat international rivals, they'll inevitably turn their tools inward. Probably sooner than later. Is that what we want in the next 3 years?
Thanks for the report Lennart. One important Q: how do you we analyze the tradeoff whereby export controls incentivize Chinese indigenous chipmaking innovation vs. keeping China dependent on Western/Intl chip makers for a longer period of time?
There are a bunch of very tenuous assumptions in this piece that essentially render it mute. For starters, volume-wise, China has way more chips deployed than the US. Yes, the US retains an edge computationally (50% vs 20% in China). However, measuring an edge just in terms of brute capacity is very treacherous. As it's been demonstrated, Huawei's latest AI cluster (CloudMatrix 384) is actually better than anything built with NVIDIA. Granted, as mentioned, this entails way more chips (5.3x more) and power (3.9x more to be precise). Nonetheless, that's the biggest logic fault here, China has energy in spades, it's not only the leader in renewables, but has already surpassed the US in nuclear power generation (with way more plants being built as we speak). So, considering that China controls most of the semiconductor supply chain (except top-of-the-line fabs for nodes <5nm of TSMC), on top of the energy one, and has way more installed power and growing, the supposed computational advantage becomes, eventually, irrelevant.
Despite this gross misconception, the author digs deeper into AI deployment in the industry. Well, anyone who has visited China recently would passionately disagree with that statement. From robotics to AVs, drones, surveillance, and most manufacturing plants, the level of AI integration in China's industrial and social fabric (mostly edge cases) is orders of magnitude higher than in the US. Let me give you a specific number to drill the point. Palmer Lucky (Anduril) states that the US capacity for Naval Army shipbuilding is 1 to 350. For every US boat, China can crank 350 (civilians included with dual-use parameters). Now, one of the reasons for this is the speed at which they adapt to design changes for the ships. One of the issues with constant redesign (a typical US Navy problem) is the rewiring. China has several labs that employ AI models to rewire ships in 1 day, vs the usual 1 year of the US. So, no, excuse me, the US still has many sectors (except the tech industry) that lag considerably when it comes to AI deployment. I get the whole rhetoric, but bring your numbers, models, and scale up to date, because if not, this is just an exercise of screaming at the clouds.
While Silicon Valley take AI capex like an AGI moonshot, China's approach to innovation and emerging tech is much more methodical and long-term based. Some of the cultural and political differences actually do favor China.
More efficiency and innovation relative to less capital - more model optimization and innovation as a direct result of AI chip export controls and bans. Fairly predictable trajectories tbh.
Each year the argument for technological primacy gets more tenuous. The correct answer for a long time, is to increase the quality of the people. Instead the US system avoids this at all costs.
Long gone is any confidence that the neolib managers of the nation's wealth and power are interested in sharing its benefits. A common expectation in much of the West is these tools will be used as a more efficient way to exploit the people and enforce control.
The advent of AI actually is a vital subject, in my opinion. Seeing some of the latest advances, I'm starting to come around to the immense transformative potential of AI. But tech as a rule is readily duplicated, so chances are distant of AI plugging the geopolitical-power gap. I.e. the dream is fallacious, of "fixing" the slower US and G7 rate of development. That gap is fundamentally due to the West - and US especially - having first wasted and later atrophied much of its human capital.
But given the ability to combine parallel calculation in massive centralized servers, ability to reason, and naturally amazing powers of pattern matching, AI will nonetheless enable great transformations. It seems particularly well matched tool to create something akin to the "panopticon". That was once the dream of high modernism. Now rightly seen by most as dystopian, a "worst practices" pattern in the realm of thinking about how to structure society.
Question people might want to ask is, do we really want to have this technology emerge under a combination of (1) proto-fascist Trump administration (2) egged on by believers in the high-modernist panopticon vision of humanity such as Thiel or Musk - and (3) that techno vision paired with the boundless greed philosophy, exemplified by price gouging basic medicine like insulin, because hey, if your customer doesn't give you all their money they're free to die aren't they? Is that 3-way power combination really going to be the one to oversee deployment of AI in the US? When they fail to beat international rivals, they'll inevitably turn their tools inward. Probably sooner than later. Is that what we want in the next 3 years?
Thanks for the report Lennart. One important Q: how do you we analyze the tradeoff whereby export controls incentivize Chinese indigenous chipmaking innovation vs. keeping China dependent on Western/Intl chip makers for a longer period of time?
There are a bunch of very tenuous assumptions in this piece that essentially render it mute. For starters, volume-wise, China has way more chips deployed than the US. Yes, the US retains an edge computationally (50% vs 20% in China). However, measuring an edge just in terms of brute capacity is very treacherous. As it's been demonstrated, Huawei's latest AI cluster (CloudMatrix 384) is actually better than anything built with NVIDIA. Granted, as mentioned, this entails way more chips (5.3x more) and power (3.9x more to be precise). Nonetheless, that's the biggest logic fault here, China has energy in spades, it's not only the leader in renewables, but has already surpassed the US in nuclear power generation (with way more plants being built as we speak). So, considering that China controls most of the semiconductor supply chain (except top-of-the-line fabs for nodes <5nm of TSMC), on top of the energy one, and has way more installed power and growing, the supposed computational advantage becomes, eventually, irrelevant.
Despite this gross misconception, the author digs deeper into AI deployment in the industry. Well, anyone who has visited China recently would passionately disagree with that statement. From robotics to AVs, drones, surveillance, and most manufacturing plants, the level of AI integration in China's industrial and social fabric (mostly edge cases) is orders of magnitude higher than in the US. Let me give you a specific number to drill the point. Palmer Lucky (Anduril) states that the US capacity for Naval Army shipbuilding is 1 to 350. For every US boat, China can crank 350 (civilians included with dual-use parameters). Now, one of the reasons for this is the speed at which they adapt to design changes for the ships. One of the issues with constant redesign (a typical US Navy problem) is the rewiring. China has several labs that employ AI models to rewire ships in 1 day, vs the usual 1 year of the US. So, no, excuse me, the US still has many sectors (except the tech industry) that lag considerably when it comes to AI deployment. I get the whole rhetoric, but bring your numbers, models, and scale up to date, because if not, this is just an exercise of screaming at the clouds.
Assuming the enormous investments actually generate the necessary increase in worker productivity 😃
we shall see!