David Claydon @dcbath

Partner and Founder, Kaya Partners. AI infrastructure, energy and growth. kaya.eco London/ Washington/ Copenhagen Joined March 2010

Tweets

3K
Followers

368
Following

822
Likes

634

David Claydon @dcbath

a day ago

@BenZaranko It already is, Ben.

0 0 0 52 0

View Details

David Claydon @dcbath

a week ago

@ZavCorin @amitisinvesting @ZavCorin is about as shrewd a thinker on credit markets/ instruments as there is… duly noted Zav.

1 0 1 21 0

View Details

David Claydon @dcbath

a week ago

@TorstenBell Yup. We have to do more because the UK has the biggest gap between the stock of debt + cost of funding vs the underlying deficit and the rate of nominal growth.

0 0 0 25 0

View Details

@coatuemgmt Is this not a nonsense comment? Agents work for humans. The future is proliferating agents doing a myriad of background tasks, serving users in a different way, migrating away from chatbot interface. Will we call that no human traffic? Odd.

0 0 0 486 0

View Details

David Claydon @dcbath

2 weeks ago

@SanderTordoir Better late than never, eh

1 0 7 2K 0

View Details

David Claydon @dcbath

2 weeks ago

@kellyjgreer What throughput efficiency gains at what volumes would that require?

0 0 0 76 0

View Details

David Claydon @dcbath

3 weeks ago

@ericlonners Tell us more Eric..

0 0 0 26 0

View Details

David Claydon @dcbath

3 weeks ago

@rowlsmanthorpe Yeah… but don’t people use agents…?!

0 0 0 27 0

View Details

David Claydon @dcbath

3 weeks ago

@ZavCorin No. It’s routine. No plan to get hooked, China simply has a compute shortage.

0 0 1 13 0

View Details

David Claydon @dcbath

3 weeks ago

@trevornoren Thx, reality to me is impressive model capability has been developed before the applications layer has been developed to fully leverage model progress. That’s a matter of time. So I don’t doubt quantum of ultimate spend, or ability to deliver value, just wonder when it arrives

0 0 0 138 0

View Details

David Claydon @dcbath

3 weeks ago

@b_judah Bout as you can see, the politics is about product prices, not crude prices. US Diesel crack spread was $20 in Feb. It’s $82.25 today. So getting crude flowing is good, but it doesn’t address the fundamental ‘pump price’ problem.

1 0 29 11K 0

View Details

David Claydon @dcbath

3 weeks ago

@MosesSternstein Not really. As the footnotes say, the data in the right panel are from a paper released in March probably (at best) including February revenues. The chart on the left is as of end July.

0 0 1 40 0

View Details

David Claydon @dcbath

3 weeks ago

@lugaricano @SVNieuwerburgh @HannoLustig Seems reasonable to think that if this much capacity were to be built (and it is not locked in at this level), it is unlikely that nominal GDP would grow at around 4% as the paper implies. What’s wrong here is not the projected spend per se, but the GDP effect of that spend.

0 0 0 300 0

View Details

David Claydon @dcbath

3 weeks ago

@RMladek @rhodium_group good addition/ detail to layer your upon recent report on China/ AI infra.

0 0 0 0 0

View Details

David Claydon @dcbath

4 weeks ago

@ChrisRMcGuire Yes. But to your penultimate point, is it not the case that given quantum compute ‘printed’ plus leap in VR capabilities plus models that the gap is growing significantly anyway?

0 0 0 33 0

View Details

David Claydon @dcbath

4 weeks ago

@GregEmersonian @TimmerFidelity I think that’s right. The question is whether inference margins can remain positive if compute cost (rental) rises and token prices fall. Does expected task efficiency make that realistic? Will token volume rise?

1 0 0 19 0

View Details

David Claydon @dcbath

4 weeks ago

@darrenpjones Without China agreeing, very few in the US system want to ‘pace the frontier’. That seems to include labs and government. Accessing US-China agreement is upstream of this perfectly reasonable letter. Need to agree that first, simply restating that it is profound is insufficient.