My only question - which will sound curmudgeonly - is in reference to 'But at least, Nvidia is supporting a customer that has a product that is actually useful'.
How confident are we that the product is useful relevant to the cost?
But I suppose you are either a believer or non-beliver on this topic!
What I meant by that was that during the dot com, a lot of the companies getting huge valuations didn't actually DO the thing they wanted to do yet. They all claimed that they would be XYZ, but most of them barely did it, had few paying customers, or the thing didn't really work yet.
OpenAI probably has close to a billion users who are each, on average, spending increasing amounts of time on the product. I think the utility is already way ahead of anything during the dot-com.
On the question of costs, there are multiple ways to look at it. I think that ifthe frontier labs didn't feel like they were in an arms race with others and had to invest as fast as possible, they could pretty easily serve their users profitably. It's training the next models and building infrastructure for the rapid growth that is most expensive, but at a stable run rate, I think they would be fine, and have pricing power and opportunities to make money in other ways (ads, affiliates, more API, etc).
Thanks for the detailed answer. I appreciate your writing on this - most of the AI stuff is too wide-eyed, but I’m much more likely to change my view with a balanced explanation like yours.
I take your point here, as someone who can still recall the dial-up noise in my mind! I don’t know if you read Ed Zitron, who I find a bit verbose & belligerent but also occasionally quite funny with it - I don’t agree with him on everything but like his point that AI doesn’t seem to have a ‘killer App’ to justify the cost/energy useage.
But ultimately it’s a nuanced situation: 1) we just don’t know yet, as ML sagely put it, 2) there are clearly some wonderful uses e.g. DeepMind with health breakthroughs 3) some training data appears problematic (not enough on women, racial differences e.g. where health issues vary).
Anyway I didn’ t mean to go big picture as I’m no expert - thanks again - I think you were one of the first CSU drum-bangers I found on CoBF back in the day - it’s not so well-known in the UK.
The Oracle buyback analysis here is fasinating - buying back 40% of shares between $40-90 and potentially reissuing at $300+ would be impressive financial enginering. But the update about borrowing $15B with $100B in net debt already on the books is concerning. At some point Oracle hits a balance sheet limit, and then they'll need to issue equity anyway, probaly at less favorable terms. The capex jump from 4% to 46% of revenue shows how capital intensive this AI infrastructure bet really is. If OpenAI doesn't deliver on those revenue projections, Oracle's financial engineering will look a lot less clever in hindsight.
I think we have to consider the second level effect of Nvidia so openly choosing one customer over others. You might force alliance that would not have happen to counter this triumvirat or other detrimental réaction from hyperscalers (msft amzn goog)
I simply stating that by investing directly in OpenAI - 10B once the deal is signed.. NVDA can no longer be viewed as a neutral merchant of arms. This might have LT implications.
Google: OpenAI is the number ONE foe, and existential threat. Google was happy to buy both TPU and NVDA chips. Now that NVDA is supporting Open AI cash burn problem a top priority for Google will be to make sure that NVDA will be off the list. If there is one aspect that I am confident is Google engineering and architectural knowhow. Google search have been running on custom SW from the get go. CUDA lock in not a problem for them.
Amazon: AWS is so critical for Amazon. OpenAI is a pain. First it helped microsoft and Azure taking market share. Then OpenAI stargate, Coreweave, Nebius are all pain points. So NVDA helping Open AI after investing with Coreweave is making it a real foe. Trainium becomes an existential product as AWS cannot afford to finance NVDA so that profits get funnel to OpenAI and coreweave.
Apple: Apple has a vesting interest in developing a edge computing approach to LLM AI processing so that the phone does not become a dumb phone. My understanding is that current AI capability from Apple chips are not necessarily related to LLM. So Apple must be working like crazy to put AI LLM capability into phones and also figure out how what LLM logic to put into the phone and in the cloud. This hybrid logic does not exist in the cloud neither in the edge computing. But this is a major technology shift (hybrid approach) thus creating a time gap. The recent judgment opens the door for more intertwined collaboration with Google so that Apple gains time to develop the hybrid cloud/edge AI solution. The now probably 25B revenue deal - think its a 37% rev sharing of safari google search - will be expanded to siri agent (gemini inside) solution. Why it is important? As such the #1 foe of your ally will become your foe.
So this decision to help out OpenAI cash problem with direct equity will cement a major alliance between Amazon, Google and Apple. Disclosure: Google is 30% of my holding and been a holder since 2004. Amazon is a top 10 holding. I was an early investor in Apple. So as such I am quite bias. But at the same time, this gives me a different perspective that NVDA bull. I am looing forward to get your thoughts.
Nice job as always. Love the unraveling of NVIDIA’s game plan here.
On free speech, I agree that it’s better for someone to believe in the principle, but I modestly disagree with the harm by showing the downside. As humans are risk adverse, I think sometimes it can be more effective to show the unintended future downside than only the principle to adhere to. Maybe we can do both?
Thanks - very detailed & balanced explanations.
My only question - which will sound curmudgeonly - is in reference to 'But at least, Nvidia is supporting a customer that has a product that is actually useful'.
How confident are we that the product is useful relevant to the cost?
But I suppose you are either a believer or non-beliver on this topic!
Thanks for reading, Tom, and for the comment!
What I meant by that was that during the dot com, a lot of the companies getting huge valuations didn't actually DO the thing they wanted to do yet. They all claimed that they would be XYZ, but most of them barely did it, had few paying customers, or the thing didn't really work yet.
OpenAI probably has close to a billion users who are each, on average, spending increasing amounts of time on the product. I think the utility is already way ahead of anything during the dot-com.
On the question of costs, there are multiple ways to look at it. I think that ifthe frontier labs didn't feel like they were in an arms race with others and had to invest as fast as possible, they could pretty easily serve their users profitably. It's training the next models and building infrastructure for the rapid growth that is most expensive, but at a stable run rate, I think they would be fine, and have pricing power and opportunities to make money in other ways (ads, affiliates, more API, etc).
I guess time will tell!
Thanks for the detailed answer. I appreciate your writing on this - most of the AI stuff is too wide-eyed, but I’m much more likely to change my view with a balanced explanation like yours.
I take your point here, as someone who can still recall the dial-up noise in my mind! I don’t know if you read Ed Zitron, who I find a bit verbose & belligerent but also occasionally quite funny with it - I don’t agree with him on everything but like his point that AI doesn’t seem to have a ‘killer App’ to justify the cost/energy useage.
But ultimately it’s a nuanced situation: 1) we just don’t know yet, as ML sagely put it, 2) there are clearly some wonderful uses e.g. DeepMind with health breakthroughs 3) some training data appears problematic (not enough on women, racial differences e.g. where health issues vary).
Anyway I didn’ t mean to go big picture as I’m no expert - thanks again - I think you were one of the first CSU drum-bangers I found on CoBF back in the day - it’s not so well-known in the UK.
The Oracle buyback analysis here is fasinating - buying back 40% of shares between $40-90 and potentially reissuing at $300+ would be impressive financial enginering. But the update about borrowing $15B with $100B in net debt already on the books is concerning. At some point Oracle hits a balance sheet limit, and then they'll need to issue equity anyway, probaly at less favorable terms. The capex jump from 4% to 46% of revenue shows how capital intensive this AI infrastructure bet really is. If OpenAI doesn't deliver on those revenue projections, Oracle's financial engineering will look a lot less clever in hindsight.
They're definitely being very aggressive. Chances are, Ellison sees this as a legacy move and is ready to gamble a lot on it
I think we have to consider the second level effect of Nvidia so openly choosing one customer over others. You might force alliance that would not have happen to counter this triumvirat or other detrimental réaction from hyperscalers (msft amzn goog)
That’s certainly a possibility. Do you have something specific in mind or are you just generally thinking about it?
I simply stating that by investing directly in OpenAI - 10B once the deal is signed.. NVDA can no longer be viewed as a neutral merchant of arms. This might have LT implications.
Google: OpenAI is the number ONE foe, and existential threat. Google was happy to buy both TPU and NVDA chips. Now that NVDA is supporting Open AI cash burn problem a top priority for Google will be to make sure that NVDA will be off the list. If there is one aspect that I am confident is Google engineering and architectural knowhow. Google search have been running on custom SW from the get go. CUDA lock in not a problem for them.
Amazon: AWS is so critical for Amazon. OpenAI is a pain. First it helped microsoft and Azure taking market share. Then OpenAI stargate, Coreweave, Nebius are all pain points. So NVDA helping Open AI after investing with Coreweave is making it a real foe. Trainium becomes an existential product as AWS cannot afford to finance NVDA so that profits get funnel to OpenAI and coreweave.
Apple: Apple has a vesting interest in developing a edge computing approach to LLM AI processing so that the phone does not become a dumb phone. My understanding is that current AI capability from Apple chips are not necessarily related to LLM. So Apple must be working like crazy to put AI LLM capability into phones and also figure out how what LLM logic to put into the phone and in the cloud. This hybrid logic does not exist in the cloud neither in the edge computing. But this is a major technology shift (hybrid approach) thus creating a time gap. The recent judgment opens the door for more intertwined collaboration with Google so that Apple gains time to develop the hybrid cloud/edge AI solution. The now probably 25B revenue deal - think its a 37% rev sharing of safari google search - will be expanded to siri agent (gemini inside) solution. Why it is important? As such the #1 foe of your ally will become your foe.
So this decision to help out OpenAI cash problem with direct equity will cement a major alliance between Amazon, Google and Apple. Disclosure: Google is 30% of my holding and been a holder since 2004. Amazon is a top 10 holding. I was an early investor in Apple. So as such I am quite bias. But at the same time, this gives me a different perspective that NVDA bull. I am looing forward to get your thoughts.
1 man's closed circular economy is another man not seeing dividends from unproven durable AI profit models.
Nice job as always. Love the unraveling of NVIDIA’s game plan here.
On free speech, I agree that it’s better for someone to believe in the principle, but I modestly disagree with the harm by showing the downside. As humans are risk adverse, I think sometimes it can be more effective to show the unintended future downside than only the principle to adhere to. Maybe we can do both?
Yes, my argument was that we should do both. Apologies if that was unclear.