Can spikey enough energy prices spill over into valuations of participants in things like patent races in energy-intensive businesses?
AI training seems pretty energy-intensive to me. But what do I know? I ain't nothing but a superannuated energy economist with a strange obsession with forests carbon banks as a kernel for the urgently needed carbon economy.
Yes, but in practice radical increases in energy costs will hurt the companies playing catchup more than the leaders.
AI training is very energy intensive, but there are also diminishing returns for ongoing training of a particular model (by model think of ChatGPT 3.5, 4 or 5)
There is even such a thing as "overtraining", where quality starts diminishing with additional training.
This is why new models are developed - the old ones stop getting significantly better no matter how much power you put into them.
Model development requires training, but in many cases algorithmic improvements don't require the full training overhead that would be associated with a new model version release.
A leading edge AI company can slow down the training iteration cycle with no impact on current revenue until competing companies roll out superior models and take market share away. This is risky in case a competitor leapfrogs the leading edge dramatically, but it is an available business decision.
Of course a trailing company playing catch up is stuffed. They have to iterate the models and pour money into AI training, sales and marketing until they actually become competitive enough to get customers to move or find a new market to exploit.
In contrast to training, serving AI queries is pretty cheap near the state of the art. Google's published number is that a typical query uses the power of 9 seconds of TV.
As with many things, if energy costs go through the roof, the stronger companies are even more likely to win, while the weaker companies burn through their VC money.
Happy birthday! Rare to hear a person say I don't know v blah blah when they have the credentials to really scenario it out but cannot when there are too many known and unknown unknowns. It self-proves the unforecastable nature of event driven economics, with precious real time add factors unknowable. They do say the trend is your friend however. There is no Fed put option. Multiple trends are not looking good in multiple dimensions. Am glad your comments reflect this. That itself is comforting :)
Happy Birthday Dr. Sahm! Mine was just last week, as was Colin WP Lewis'.
.
“I am not on recession watch, but I am concerned,” I'm downright terrified. We have the most fickle, feckless president of all time who rules by caprice. We don't know what he's going to decide an hour from now. He doesn't know what he's going to decide until the moment he does. Then he'll rage tweet it in the middle of the night. We're living in The Twilight Zone.
We'll see how long inflation persists due to the Iran conflict and the subsequent rise in gas prices, but I don't believe the negative supply shocks from it cannot be solved by the federal reserve reversing course and having to hike rates. Hiking rates at a time when Americans already struggling with affordability and private debt would be a terrible move.
Can spikey enough energy prices spill over into valuations of participants in things like patent races in energy-intensive businesses?
AI training seems pretty energy-intensive to me. But what do I know? I ain't nothing but a superannuated energy economist with a strange obsession with forests carbon banks as a kernel for the urgently needed carbon economy.
Happy birthday, young'un.
Yes, but in practice radical increases in energy costs will hurt the companies playing catchup more than the leaders.
AI training is very energy intensive, but there are also diminishing returns for ongoing training of a particular model (by model think of ChatGPT 3.5, 4 or 5)
There is even such a thing as "overtraining", where quality starts diminishing with additional training.
This is why new models are developed - the old ones stop getting significantly better no matter how much power you put into them.
Model development requires training, but in many cases algorithmic improvements don't require the full training overhead that would be associated with a new model version release.
A leading edge AI company can slow down the training iteration cycle with no impact on current revenue until competing companies roll out superior models and take market share away. This is risky in case a competitor leapfrogs the leading edge dramatically, but it is an available business decision.
Of course a trailing company playing catch up is stuffed. They have to iterate the models and pour money into AI training, sales and marketing until they actually become competitive enough to get customers to move or find a new market to exploit.
In contrast to training, serving AI queries is pretty cheap near the state of the art. Google's published number is that a typical query uses the power of 9 seconds of TV.
https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference/)
As with many things, if energy costs go through the roof, the stronger companies are even more likely to win, while the weaker companies burn through their VC money.
Happy Birthday! All the best.
Perhaps some day we'll have rational coherent policy and leaders who think through the consequences of their actions. One can dream.
If we can survive this regime.
Happy birthday! Rare to hear a person say I don't know v blah blah when they have the credentials to really scenario it out but cannot when there are too many known and unknown unknowns. It self-proves the unforecastable nature of event driven economics, with precious real time add factors unknowable. They do say the trend is your friend however. There is no Fed put option. Multiple trends are not looking good in multiple dimensions. Am glad your comments reflect this. That itself is comforting :)
Cheers to the first and next 50 years!
Happy Birthday!! You Rock!
Happy Birthday Dr. Sahm! Mine was just last week, as was Colin WP Lewis'.
.
“I am not on recession watch, but I am concerned,” I'm downright terrified. We have the most fickle, feckless president of all time who rules by caprice. We don't know what he's going to decide an hour from now. He doesn't know what he's going to decide until the moment he does. Then he'll rage tweet it in the middle of the night. We're living in The Twilight Zone.
Good post, as always!
Happy Birthday, and I'm overjoyed that it has been a healthy one for you!
Happy birthday 🎉
Happy birthday!
Happy belated Birthday!
happy birthday!! life is precious !
Happy belated birthday Claudia! You’re so young. I very much appreciate your economic expertise and useful perspectives. I wish you all the best.
A dear friend turned 57 today, so hey I'm wishing you the most happiest of happy birthdays!!
Happy birthday!
Happy birthday! I enjoy your insights.
We'll see how long inflation persists due to the Iran conflict and the subsequent rise in gas prices, but I don't believe the negative supply shocks from it cannot be solved by the federal reserve reversing course and having to hike rates. Hiking rates at a time when Americans already struggling with affordability and private debt would be a terrible move.
On a side note, happy belated birthday