& so: this is nuts! When’s the crash? I watch the race to build ever-smarter machines and think those hoping for immense profits for themselves are highly likely to wind up very disappointed. Think...
Claude: "OpenAI's annualized revenue run rate surged to $10 billion which doubles from the $5.5 billion it reported in December 2024, company foresees revenue tripling this year to $12.7 billion."
How long until Huawei or similar creates reasonable competing AI chips? They know the chips are possible, they can study them, they can look at almost good enough lithograph machines. They can poach employees and recruit spies from the NVDA and TSM. NVIDIA is running a 56% profit margin, TSM's is 41%. If that isn't incentive enough, I bet Huawei would sell AI chips at a loss for prestige and strategic reasons.
The day that Huawei sells anything close to a competing chip, what is the value of NVIDIA?. How high would the barrier to entry be for new AI modelers? They better dig that moat fast because that day is coming.
Huawei strikes me as a Chinese version of IBM or Boeing, which is to say, too big, old, and insular to innovate. But with enough money over a couple of years either them or someone else can at least copy NVDA.
The argument that AI is embedded in something - hardware or software, that the producer controls is the right solution for profits. Microsoft might have to give away value by embedding AI into its Office Suite, but it will still make profits on the suite. It also has the benefit that the AI can run both locally and, if needed, in the cloud. OpenAI's ChatGPT can only run in the cloud (which currently costs OpenAI ROI losses). If they create local models, can they even charge for them beyond a nominal amount (I don't believe so)?
Palantir may well add AI to their data analysis and content service. Again, AI is part of the offering with its costs hidden from the buyer, even as the value is demonstrated as a sales proposition or benefit.
Once the architecture of a "thinking" machine is discovered, it may well offer 2nd mover advantage. We also don't know if hyper-scaling is the way to go, rather than a much smaller, curated corpus. If the model is small, computationally lightweight, and accurate, and importantly, able to reason well, this may be the way to go, especially if it is the means to create useful [humanoid] robots that learn new tasks well, and can follow instructions and learn from experience the owner's idiosyncrasies.
Claude: "OpenAI's annualized revenue run rate surged to $10 billion which doubles from the $5.5 billion it reported in December 2024, company foresees revenue tripling this year to $12.7 billion."
Gotta be a pony in there somewhere.
https://www.wheresyoured.at/make-fun-of-them/
Includes some financial details re OpenAI. Quite a few in fact.
How long until Huawei or similar creates reasonable competing AI chips? They know the chips are possible, they can study them, they can look at almost good enough lithograph machines. They can poach employees and recruit spies from the NVDA and TSM. NVIDIA is running a 56% profit margin, TSM's is 41%. If that isn't incentive enough, I bet Huawei would sell AI chips at a loss for prestige and strategic reasons.
The day that Huawei sells anything close to a competing chip, what is the value of NVIDIA?. How high would the barrier to entry be for new AI modelers? They better dig that moat fast because that day is coming.
I don’t think Huawei can catch NVidia and AMD in anything close to the near term.
Huawei strikes me as a Chinese version of IBM or Boeing, which is to say, too big, old, and insular to innovate. But with enough money over a couple of years either them or someone else can at least copy NVDA.
The argument that AI is embedded in something - hardware or software, that the producer controls is the right solution for profits. Microsoft might have to give away value by embedding AI into its Office Suite, but it will still make profits on the suite. It also has the benefit that the AI can run both locally and, if needed, in the cloud. OpenAI's ChatGPT can only run in the cloud (which currently costs OpenAI ROI losses). If they create local models, can they even charge for them beyond a nominal amount (I don't believe so)?
Palantir may well add AI to their data analysis and content service. Again, AI is part of the offering with its costs hidden from the buyer, even as the value is demonstrated as a sales proposition or benefit.
Once the architecture of a "thinking" machine is discovered, it may well offer 2nd mover advantage. We also don't know if hyper-scaling is the way to go, rather than a much smaller, curated corpus. If the model is small, computationally lightweight, and accurate, and importantly, able to reason well, this may be the way to go, especially if it is the means to create useful [humanoid] robots that learn new tasks well, and can follow instructions and learn from experience the owner's idiosyncrasies.