Cofounder @3F_Research. Mainly trying to post cool charts. PM $FCTE $RAA https://xitter.cf/t.co/H7PivYBy3Z Try Caliban https://xitter.cf/t.co/ugvbuqiGcT

San Francisco, CA
Joined December 2012
It is funny the massive amount of compute needed to overcome the lack of a public API for so many products and services out there. Imagine how many flops of compute were needed across models here to just pull up a list of unbooked appointment slots.
My muse called Volvo for me to see when the next available appointment is. You can listen to the phone call here. I have it do multiple calls like this every day (I'm still trying to get better and better at leveraging muse). My muse called, waited on the line for 15 minutes, and then got the information I needed. - Saved me a ton of time - Muse was able to talk to the other person, with no confusion or interruptions on Volvo's side, and got the info we needed (This is the real transcript from the call but the audio has been recreated for this video)
3
16
2,361
Yes. Token expenditure indexes get pushed down by efficiency gains and better hardware, yet that’s often misread as margin compression. They can go up as better models are used for harder tasks. Hard to read as bullish or bearish. Rental rates are a better measure of AI demand.
Just intuitively, GPU rental rates seem a more useful signal that token indices. Any reason to disagree?
2
7
47
10,093
Fernando Vidal retweeted
Yesterday, @NickTimiraos asked Kevin Warsh what had changed over the past year to reverse his view of the economy. Yes the Iran War has created a supply shock. But, the real change is the AI buildout. Never before has the Fed cut in the face of a capex boom until the past year.
36
83
15
565
65,191
People realize slowing down the frontier means we keep seeing the rate of model progress we are currently used to right? The labs are saying the frontier is accelerating from an already rapid pace. I keep seeing sarcastic “so much for slowing down” comments when new models drop.
2
1
9
1,874
Fernando Vidal retweeted
S&P 1500 Software revenue per employee has gone parabolic. If you are looking for evidence that AI is starting to impact the real economy, this is exhibit A.
116
313
54
2,719
244,982
Upward estimates revisions for cloud revenues the next few years at hyperscalers roughly matching the rise in GPU rental rates this year makes sense. Will be interesting to see what Vera Rubin rental rates look like in the coming months.
5
1
37
6,713
Fernando Vidal retweeted
Looking back, Warren's insight here was among the best of the past 5 years. So much change. #fintwit #econtwit @cnbc @WarrenPies
25 yrs of increased globalization has delivered two things: durable goods deflation & a negative stock-bond correlation Will COVID mark a change to these dynamics? I can't think of one chart that frames the secular issues facing asset allocators better than this. @3F_Research
2
10
56
13,312
A good chunk of the stock market is trading like a bet against semis. Across all stocks, the 10th percentile correlation with semis is the most negative in 35 years. I used a new alerting feature we added to Caliban to chart it and let me know if the relationship changes.
2
2
42
6,791
This is what the agentic AI capabilities unlock did to semiconductor earnings estimates in 2026. Is that the last capability breakthrough that will pull forward adoption and surprise analysts this much?
1
11
1
109
20,042
Fernando Vidal retweeted
OIL -Brent futures >$100 (new post-Iran high). -Dated Brent >$108 (DFL >$8/bbl...i.e. physical tightening). Going back to the end of August, the paper market was starting to flash buy signals. Below is a clip from the @3F_Research 8/27 Model Update.
11
20
2
220
23,253
So little slack in the system that one AI service being down just causes a surge in demand for the others and the outage rolls through all the providers. Easy to believe there really is no spare capacity.
2
16
1,699
Fernando Vidal retweeted
Will the Fed hike in September?
33%Yes
67%No
6,152 votes • Final results
67
23
15
158
79,277
Fixing this took a lot of prompt engineering and feedback from an annoyed @WarrenPies for us. We had to think through and lay out much of our internal charting principles and bake them into Caliban.
It's crazy how you can even tell when LLMs write something as small as the label on a graph. It's always slightly miscalibrated to the reader Literary theory speaks of this
2
4
2,240
$NVDA comments on weak H200 demand align with rising availability in our data…Working theory is that demand for pre-Blackwell GPUs plateauing at current rental rates. Ultimately Blackwell remains the key rental market and is still very strong.
Over the past week, we are seeing some early cracks in our GPU availability data...not going to overreact but have to call balls and strikes...On net, bad combo w/ the Anthropic ARR miss.
4
7
75
16,850
Open model capabilities are improving quickly. But empirically every time we review which model to use for Caliban, we end up back at either ant or oai. Recent frontier model price cuts make it an even easier choice, and let us bump up usage limits for our users significantly.
3
14
1,755
Fernando Vidal retweeted
Yesterday, I spoke with Max about: -Neutralizing our equity risk recently, -Fed dynamics, -Compute demand Enjoyed the conversation.
Macro risk is taking over and it's time to reduce equity risk... @WarrenPies argues on @opmpod that the time to be overweight stocks was before strong earnings, not after, especially with looming macro risk from the Fed, cracks in the AI thesis, and poor seasonality. 00:00 Intro 01:21 Why Macro Takes Over 03:19 Earnings Season Structure 05:43 Upcoming Macro Catalysts 08:24 Fed Hike Odds Mispriced 10:37 Political Pressure on Fed 15:29 Sponsor Break HFGM 17:45 Coin Flip Meeting Dynamics 19:42 Rates Curve and Term Premium 24:54 How to Position in September 26:53 Tech Breadth and AI Cracks 30:17 Lab ARR Reality Check 34:40 Tech Leadership Needed 38:17 Data Center ROI Debate 40:36 Election and Policy Risk 43:44 AI Issuance Scapegoats 47:35 Energy as Diversifier 51:48 Anthropic IPO Impact 55:01 Software Versus Semis 58:28 Pair Trades and Correlations 01:01:19 Fade Macro Risk Playbook 01:02:55 Conclusion Check out the full interview YouTube: bit.ly/4hSUX39 Apple: bit.ly/45Im1uC Spotify: bit.ly/4gsKngO
14
35
4
333
59,072
Fernando Vidal retweeted
Are datacenter delays, or cancellations, impacting the AI buildout? Not yet. Chart below from @fernavid 's most recent AI update plots every DC project's construction progress since groundbreaking. 2026 projects actually tracking ahead of schedule (vs 28-month timeline).
35
87
9
548
50,531
Fernando Vidal retweeted
New report out to 3FR clients. In this edition, @fernavid -Benchmarks the new K3 model against the frontier. -Discusses the shift from tokenomics to "taskenomics." -Checks in on the datacenter buildout using granular data.
2
9
74
13,304
Fernando Vidal retweeted
The last few sessions shows how powerful the index can rally when hypers + semis are rowing in the same direction. "Compute demand is strong. Semis are an obvious beneficiary. But, it is not a zero-sum game. Robust compute demand is positive for datacenter IRRs as well."
14
48
5
421
27,587