Patel reports his newsletter has two hundred thousand subscribers with five thousand paid, generating under five percent of revenue.
“So the newsletter itself is, like, 200,000. Paid newsletter is, like, I think, like, 200,000 people. Paid newsletter is, like, 5,000. Nice. But then, like, the main”
Patel says his business model shifted so that 95% of revenue now comes from selling datasets.
“the business transformed into, you know, selling data, right, or build building datasets and selling them. Right? And when the business truly flipped to that, we're now 95% of the revenue is that data”
Patel says neo clouds must scale massively, add software, accept real estate returns, or go bankrupt.
“you either have to go, like, really, really big, or you need to move into the software layer, or you just make commercial real estate, or you go bankrupt.”
Patel argues neo clouds with long-term enterprise and hyperscaler deals will succeed over on-demand providers.
“And so those are probably less likely to be able to succeed versus those who are locking in these massive deals with or long term deals that may not be massive but with enterprises or with AI labs, or with the hyperscalers who have no capacity because the demand is just so incredible.”
Patel says inference providers sell public endpoints at flat or negative margins, compensating through private deployments.
“Most of the inference providers are selling at flat margins or even negative for their public endpoints. And they then make it up when people do private deployments.”
Patel describes public inference as a loss leader to generate private sovereign and enterprise engagements.
“And so I think that's what a lot of this is, is you end up with the public business as like a loss leader just to generate private engagements, whether it be sovereigns or enterprises.”
Patel questions whether neo clouds targeting developers on short-term rentals can achieve ROI.
“Those that are just targeting developers on short term rentals, they may not be able to get their ROI back.”
Patel left Substack when SemiAnalysis reached multi-million dollar revenue with increasing share from non-Substack sources.
“And then I left when we hit a few million of revenue. More and more of the revenue was coming from not Substack.”
Patel left Substack initially after reaching a few million in revenue to avoid the 10% fee.
“And then I left when we hit a few million of revenue. More and more of the revenue was coming from not Substack. The reason was like, Hey, Substack takes 10%.”
Patel explains SemiAnalysis publishes public research on GPU benchmarking to build scale beyond institutional client work.
“Let's figure out how to figure it out and just post it publicly, and that begets more and more scale. So we've done this with a lot of GPU benchmarking and testing”
“Base Ten, like Tuhin's probably made more money off of, you know, Minimax or GLM or DeepSeq than those companies have because he's serving those models at scale. And same with, Linett Fireworks.”
SemiAnalysis expanding beyond research and newsletters into direct moneymaking was predictable given their accumulated expertise.
“This was always kind of inevitable, right? Was SemiAnalysis really going to stick to research, and newsletters, and models for financial companies forever? Or are they gonna branch out into making money themselves from all of this research that they're doing?”
The speaker suggests SemiAnalysis was unlikely to remain solely a research and newsletter business and would eventually monetize their analysis through direct investment.
“This was always kind of inevitable, right? Was SemiAnalysis really going to stick to research, and newsletters, and models for financial companies forever? Or are they gonna branch out into making money themselves from all of this research that they're doing?”
Open models staying competitive at lower cost raises fears that the model layer will become commoditized.
“However, major FUD has also emerged as a result of open model success : if open models stay capable enough relative to the closed frontier at a fraction of the cost, won't the model layer become commoditized?”