Dylan Patel's SemiAnalysis Empire: $500M Fund, Citrini & NVIDIA?
- Patel sees tighter chip controls and dependence on Jensen Huang as NVIDIA’s main risks from the allegations surrounding its CEO. If Huang leaves, Patel sees no clear successor beneath roughly 50 direct reports; he expects limited direct customer backlash because buyers still need GPUs. Investigations could take months, and he allows that no illegal act may ultimately be established, but his export-policy call is categorical: “there’s definitely no way new very high-end chips will be approved.”
- Permits are pushing meaningful data-center capacity into later years. SemiAnalysis has moved multiple gigawatts from 2027 into 2028, with additional 2028 capacity slipping because “permits are just not available.” Space offers little relief over the next three or four years: Patel takes the over on Musk’s 2028 ambitions and guesses a gigawatt might arrive this decade, “maybe 2030.”
- The product layer is already redirecting substantial AI spending without owning the best model. Persistent agents have risen from zero to 45% of SemiAnalysis’s $50 million annual AI spend, predominantly through Perplexity Computer using Claude and OpenAI underneath. Patel calls it “just a better harness,” while arguing that Muse and Instinct deliver better products than OpenAI Dots despite weaker underlying models.
- Application adoption does not settle the questions of valuation, margins or customer retention. Patel sees real ROI at companies such as Sierra and Harvey, but acknowledges potentially high acquisition costs and businesses that are “pure pass through.” Developer tools may be easy to switch; enterprise discounts tied to allocating at least 70% of AI spending to Anthropic create a different retention mechanism.
- Patel’s $500 million solo-GP fund concentrates on seed and Series A infrastructure where he believes his research expertise transfers. Its scope includes robotics, silicon, optics, manufacturing, networking and infrastructure software. “I don’t know shit about applications” is his boundary: hardware investing, he argues, requires both a strong founder and a strong idea.
- The Citrini acquisition preserves a bearish AI voice inside Patel’s broader business. He remains bullish over the long term, while Citrini retains editorial control: “Dude, do whatever you want.” His stated commercial bargain is that research credibility supports data and consulting revenue; he points to publicly criticizing Fluidstack’s delays despite holding a large investment.
- Patel expects AI to accelerate hardware innovation, while acknowledging that current models still struggle with demanding technical work. SemiAnalysis’s GDPX benchmark spans cyber, chip and hardware design, materials science, biology, quant and RSI. “The models suck at most of these benchmarks,” he says, but predicts the number of companies building hardware successfully will “skyrocket” over the next year.
1. NVIDIA’s exposure runs through leadership and chip access
O’Shea steers discussion of the allegations surrounding Jensen Huang toward market consequences. Patel’s worst case is conditional: if Huang leaves, “NVIDIA is like fucked in my opinion.” Roughly 50 direct reports span different functions without an obvious successor. He expects limited customer backlash: Google still needs GPUs, and NVIDIA has prioritized AI over some high-end gaming production.
The export distinction in Patel’s account is where the hardware ends up. He says Megaspeed bought GPUs deployed in Southeast Asia primarily, or “almost entirely,” for Alibaba. Renting that offshore capacity to China could involve no legal violation; chips subsequently entering China would change the analysis. He does not establish that Huang personally committed an illegal act.
His supply-chain estimates put GPUs reaching China on the order of “hundreds of thousands,” with many more hundreds of thousands rented to Chinese companies abroad. He says his team has observed chips going to China, but still allows a scenario in which the conduct under discussion was legal. Investigations, congressional questioning and any finding of wrongdoing would take months.
Patel contrasts the H20 ban with subsequent H200 approval, noting that China did not want the H200, and recalls expectations that Blackwell could become available to China once Rubin launched. His new call is categorical: “there’s definitely no way new very high-end chips will be approved.” He thinks eventual restrictions probably extend beyond direct sales to companies renting GPUs to Chinese customers.
2. Rental access could preserve revenue while shifting who captures it
Patel’s tentative preference is to limit China’s chip ownership while permitting some access through rentals. Buying and deploying hardware is cheaper than renting it; routing consumption through Oracle, Microsoft or Google also leaves more profit dollars in the US economy. He bets some version of an approved-provider framework, with security requirements, is more likely after earlier diffusion rules proved too complicated.
Competitive consequences could be uneven. Patel thinks AMD could sell to a Singapore-based company renting capacity to Chinese clouds with much less scrutiny than NVIDIA would now receive. That could redirect some deals serving Alibaba, ByteDance or similar ultimate customers toward AMD or another vendor, without requiring the data center to be inside China.
The less direct risk is distraction from NVIDIA’s support of the ecosystem. Patel cites favorable neocloud allocations, startup investments, software help and financial backstops. He asks whether scrutiny leaves Huang less room or attention for that work precisely when credit markets are “closing down” and others are reluctant to provide support. He raises the possibility without claiming disruption is inevitable.
3. Permitting is delaying capacity that could otherwise be built
SemiAnalysis has pushed multiple gigawatts from 2027 into 2028, and quite a few gigawatts scheduled for 2028 are slipping too. Patel’s distinction is explicit: these delays are not necessarily construction, labor or equipment shortages. Capacity is moving because “permits are just not available,” making local administrative decisions a central forecasting variable.
Discussing Texas, Patel views electricity-price protection and responsible water use as solvable requirements. Union mandates look harder because the state has little union labor, while electricians and industrial plumbers can already earn $200,000 or more on data-center projects. He questions how compulsory membership would work for those workers and whether it would prove politically popular.
Governments retain several ways to halt projects: water, wastewater, stormwater and air permits. An off-grid facility using little water can still be stopped through its air permit. Patel expects investment and employment incentives to encourage accommodation, but his team has already adjusted project timelines rather than assuming permits arrive because the physical resources exist.
Space does not solve the next three or four years in meaningful volume, he argues. Musk is trying for substantial deployment in 2028; Patel takes “the over on that.” He guesses a gigawatt might arrive this decade, “maybe 2030,” while retaining a larger uncertainty: engineering is difficult, and space could be “way more than a few years away.”
4. Community economics matter more than generalized opposition
A survey conducted with David Shore found that 49% of residents in counties with a 100-megawatt data center did not know it existed. Only 40% knew about similarly sized capacity coming within three years. Patel says most jurisdictions explicitly banning data centers lack meaningful existing capacity; broad hostility does not necessarily reflect experience of a nearby operating facility.
Patel’s provocative water comparison is that Musk’s data centers use less than three In-N-Outs. O’Shea objects: “You don’t get an asterisk.” His qualification matters: the calculation includes cattle’s supply-chain water, and also the data centers’ embedded materials, chip fabrication and power generation. He accepts that poor designs can waste water or create noise, while arguing those outcomes are avoidable.
A Louisiana county paid teachers roughly $20,000 after data-center tax receipts, Patel says. But he accepts the opposite arrangement can fail residents: property-tax breaks, higher grid prices and few permanent jobs after construction. His practical answer is to share the upside, including subsidizing electricity within a 30-mile radius, which he argues would cost developers very little.
5. Better models do not automatically produce better products
Patel says Anthropic “crushed it in Q2,” OpenAI led in Q3, and Anthropic is rebounding with Opus 4.5 and GPT-5.2 while OpenAI delays releases. He revises his initial description of Anthropic’s Q3 from poor to “okay.” His warning is that online observers “over-index on any individual data point,” while valuations move more smoothly than the narrative.
In his assessment, Instinct and Muse have worse models than OpenAI but are “way better products” than Dots. Instinct primarily uses open models. Patel blames some of OpenAI’s product limitations on Guardian, which intervenes in Codex and Dots, often overzealously in his experience. Muse’s clearer interface reinforces his broader criticism of the frontier labs’ product execution.
O’Shea’s pushback is concrete: what if an agent cancels something important or says something harmful? Patel initially answers, “Yeah, let me roll the dice,” then concedes, “Obviously, that’s incorrect.” His rebuttal is comparative: he is not hearing about Muse and Instinct producing those failures constantly, so he thinks OpenAI has placed intervention too far toward restriction.
His own spending illustrates the importance of packaging. Persistent agents went from zero to 45% of a $50 million annual AI budget within a few months. Most of that portion runs through Perplexity Computer, with Claude and OpenAI underneath: “it’s just a better harness.” He does not know whether the relationship will persist; his organization can change providers quickly.
6. Revenue growth leaves retention and margins unresolved
Patel expects consumer products such as Muse and Instinct to become relatively sticky around their use cases, while developers can switch Claude Code and Codex easily. Enterprise purchasing adds contractual friction: he has seen Anthropic offer discounts conditional on at least 70% of a customer’s AI spending going to Anthropic. That concentration requirement can sustain usage beyond a model’s temporary lead.
“Most startups will fail” remains his baseline, alongside confidence that some will be highly successful. Instinct appears to have staying power, although he does not know whether its valuation is justified. Sierra and Harvey show rapid revenue growth and real ROI, in his view, even if customer-acquisition costs may be high.
The unresolved question is who retains the economics. Harvey’s reported margins have “seesawed,” while Patel suspects Perplexity’s margin on his spending is not high: he probably would not switch if it were a more expensive way to consume Claude. Inference-provider multiples might be unjustified—or warranted by sufficiently fast revenue growth and good margins. He leaves that judgment company-specific.
7. The $500 million fund follows Patel’s claimed circle of competence
Patel connects the expansion to wanting to “know everything,” have fun, make money, influence consequential decisions and spend time with loved ones. Data sales and consulting remain the bulk of the research business’s revenue. Investing is another way to act on seeing something promising early: he has raised $500 million as a solo GP, targeting seed and Series A.
The fund spans robotics, silicon, optics, manufacturing, networking and software connected to hardware or infrastructure. Patel points to colleagues with experience at ASML, Applied Materials, Lam Research, TSMC, Intel and NVIDIA, plus cloud infrastructure, compilers, model training, data centers and power plants. Robotics is still aspirational: the firm is trying to become expert in its supply chain.
“I don’t know shit about applications” explains his boundary. He might help an application company reduce infrastructure costs without knowing how to scale its product to a billion users or judge its eventual success. Hardware, he argues, is less forgiving of backing an excellent founder who will simply find a direction: it requires both a great founder and a great idea.
He declines to disclose check sizes because doing so “removes optionality for me to do something different.” Sourcing comes through friends, colleagues and the broader network; employees can refer interesting companies. His description of the pitch environment is blunt—most pitches are “boring and wrong”—but he believes there are enough strong innovators to find worthwhile opportunities.
8. Citrini keeps its voice and its disagreement with Patel
The acquisition is intended to improve how Citrini operates while preserving its identity. Patel praises James and the team; research director Kyle retains the most say over publication. When they asked whether they could remain bearish, his response was, “Dude, do whatever you want.” Citrini remains a separate company rather than being folded into SemiAnalysis.
The disagreement is substantive: Citrini turned strongly negative on AI months earlier, while Patel remains “super bullish AI in the long term.” O’Shea recalls its market-moving paper, probably from February. Patel stresses that the team described that piece as a “thought exercise,” not necessarily a prediction of what it believed would happen.
Their methods differ too. SemiAnalysis tracks infrastructure and supply-chain quantities, tests chips and models, and follows details such as cables and memory prices. Citrini’s approach can involve trying an activity firsthand: to investigate AI drug design, the team developed scorpion antivenom. Patel’s qualification survives the anecdote—it works “okay,” rather than constituting an unqualified breakthrough.
9. Research credibility is the commercial asset Patel says he must protect
Patel sees information as more valuable than ever while criticizing established information-service businesses. He calls Gartner’s monetization “like a racket” and rejects paid opinions as a model for his own firm. SemiAnalysis has never run ads or accepted payment for an opinion, he says; selling that credibility would undermine the data and consulting business built around it.
His biggest customers are hyperscalers, semiconductor companies and AI labs, including companies the firm criticizes sharply. Some have tried to cancel contracts after unfavorable commentary, only for their teams to insist they still need the data. Patel acknowledges he has been wrong; the commitment is that he and the team “will always say what we believe.”
Iren illustrates why technical criticism and asset value can coexist. Patel says its cloud technology and cluster reliability can be poor while its powered land is valuable. Those are separate assessments of the same company, not a requirement to adopt one uniformly positive or negative position. Research can recognize an attractive resource without endorsing every part of the operation.
He also cites publicly reporting Fluidstack delays despite holding a large investment. The fund is separate from SemiAnalysis and Citrini, he says, with consulting information subject to compliance controls and no public promotion of portfolio companies. His objection to investor self-censorship is memorable: ownership need not make someone “the tail wagged by the dog.”
10. Running the operation remains hard as technical ambitions expand
The business now has about 115 people across 15 countries, with offices in New York, Chicago, San Francisco, Oregon, Singapore and Korea. Patel describes the costs of operating at that scale plainly: lost sleep, missed personal commitments and constant choices about which work receives attention.
His hardest examples are firing people, being sued by someone fired, resolving team disputes and negotiating with customers or suppliers. Those experiences differ from simply deciding whether to invest. Asked how to hire, motivate and manage through them, he admits, “I don’t know the answer to these things”—even after some repetitions, “You just kind of fly blind.”
Upcoming work includes inference testing of Huawei chips, Amazon’s chips and AI startups’ chips. The new GDPX benchmark tests cyber, chip design, hardware design, materials science, biology, quant and RSI. Patel is interested in what progress on those difficult tasks enables: “the models suck at most of these benchmarks.”
His forward call is that hardware development will “accelerate like crazy.” Within a year, he predicts both the number of hardware companies and the number executing well will “skyrocket.” Robotics and niche semiconductor technologies remain areas of enthusiasm, especially packaging, networking and memory; improving model performance is the mechanism he expects to widen participation.
Full transcript
You’re building out an empire here.
We’ve got people in 15 countries. There are about 115 people, with offices in New York, Chicago, San Francisco, Oregon, Singapore, and Korea. We track every data center in the world, and as a firm, we know more about data centers than anyone else. Forty-nine percent of people in counties with a 100-megawatt data center didn’t even know there was a data center in their county.
Oh my God.
My biggest customers are hyperscalers, semiconductor companies, and AI labs.
Let’s start with the aftershocks that will come with the NVIDIA and Jensen scandal. How are you feeling this one out?
Yeah, well, good to be here. It was wizards-themed. Is it sorcery? I get that it’s like source code or whatever, but—
It’s not source code at all.
It’s not based on source code?
It was for sourcing deals. I come from the VC world.
Oh. I was thinking source code, or that we’re magical.
Magic would be cool. That makes a lot of sense. There was a play on it.
But your aura has gone down a lot. It was very—
Damn.
It was like source code. Wizards, source code—
VC.
—and then VC.
I know.
I think this is a—
This is where I really need to go. That’s why I live in LA. I’m closer to the witchery—or whatever they call it, all the wizards.
There are wizards in LA?
There’s lots of stuff in LA. There are crystals, reiki—
There are a lot of crystal people, crystal girls. But what do you mean? What’s reiki? What are wizard people?
Oh, you need to get into this.
Please explain.
It’s all energy healing and the Earth and kumbaya stuff. I think some of that has migrated up north because you guys have taken a liking to mushrooms, psychedelics, and ayahuasca.
No, I haven’t. I’ve never done a drug in my life.
Really?
Yeah.
No alcohol?
No, completely.
Holy crap.
Obviously, I drink.
That’s crazy. This is kind of on theme, though, because we were calling this the Hogwarts library over here. You have a deep, secret library in the back of this office that’s actually quite nice. Congrats.
Thank you. Thank you.
For today, I was trying to come up with all the different things we could talk about, because we could talk about everything with you. There’s a lot going on, whether it’s IPOs, scandals, models, data centers, new regulation, all the research you’re putting out, or even how you’ve been growing SemiAnalysis.
Let’s start with the aftershocks that will come with the NVIDIA and Jensen scandal. How are you feeling this one out?
Bad thing he did, right? Very bad. There’s a lot to read in the tea leaves, and a lot going on there. Diving too much into the personal life is probably a disservice and rude.
Yeah. By the way, that’s not the angle I’m going after at all.
Yeah, yeah, yeah.
I’m talking about the market.
Yeah, yeah, yeah, definitely.
Let’s talk through the supply chain. We can shelve the personal stuff.
Yes, yes.
That’s personal, but you run an AI research firm.
Yeah.
You focus a lot on semiconductors. Let’s break this down.
Sorry, you didn’t come to me to diagnose a relationship?
No. Please tell me. What is your—
I don’t know. I think it’s important to say that it’s a very bad thing, and it seems bad. I feel bad for the parties involved.
Supply chain-wise, East Asia has very different standards, so I think a lot of the East Asian CEOs don’t even give a shit. They’re like, “Oh yeah, duh, who cares?” A lot of the people in the supply chain actually don’t care, which is kind of crazy.
As far as the U.S. and what’s going to happen there, it’s much more meaningful. Let’s start with the worst-case scenario: this all blows up, and Jensen is no longer the CEO. NVIDIA is fucked, in my opinion, if that ever happens, because there’s no clear number two. There are many people who report directly to Jensen. I won’t say names, but a handful of them are super integral, and they work on different things. There are different people on the hardware side and the product side, and obviously some of the other executives in the company, but there’s no one clear successor. Jensen has 50 reports or whatever, so it’s kind of hard.
As far as what happens in AI and AI infrastructure, I think the U.S. government is definitely going to put the hammer down on restrictions. NVIDIA has been very strong at getting its way. Obviously, it’s not allowed to sell everything to China, but the restrictions have moved quite a bit in the direction of relaxation ever since the October 2022 restrictions that the Biden administration put down. They’ve relaxed on the chip side.
It’s one thing to say, “We want to sell, and we think it’s best for American dominance over the technology industry,” which could be a valid argument in some ways. Selling your technology, rather than restricting it and propping up their competitive industry, could make sense. But now those arguments are sort of screwed. From some perspectives, or in some people’s view, you’re automatically in the group of people who are treasonous.
I think it’s a bit of a stretch to say she’s a spy directly. I actually think she’s just the Megaspeed lady. She’s part of the Alibaba clan, and she’s an executive with secure capacity. She definitely had a relationship with Jensen, but I don’t think she’s—I mean, she’s a well-connected individual and, therefore, she was obviously talking to the CCP, just like any well-connected individual in China is going to talk to the CCP. So she’s an asset, and we don’t know what Jensen told her or whatnot.
It’s 100% clear that they bought a fuckload of GPUs and deployed them in Southeast Asia, primarily for Alibaba—or almost entirely for Alibaba. It would have been against export restrictions to sell directly to Alibaba for deployment in China. Now, you could argue no laws were broken if those chips were shipped and deployed in Southeast Asia and rented to China. But if some of them made their way to China, then laws were broken.
The intelligence community has been very anti-sending chips to China, whereas economic interests won out in some cases with the H20. Eventually that got banned, but then the H200 was allowed. China didn’t want the H200, but basically, there were a lot of dealings there.
Now there’s definitely no way new, very high-end chips will be approved. The thought was that when Rubin came out, NVIDIA would be allowed to sell Blackwell to China. A lot of people believed that was going to happen. There were murmurings in government and within a contingent. Another contingent obviously didn’t want any of that to happen, because the right wing is very split: there are the super hawks, and then there are the people who say, “Let’s just make money.”
I think the intelligence community is mostly on the super-hawkish side. They learned about this, and it probably leaked from them to the journalists who reported on it, if I had to guess, but I don’t know.
Formal investigations and things like that take time. The PR aspect of this is obviously bad, but it’s not like NVIDIA sells to that many consumers. They even shut down manufacturing of some of their high-end gaming chips so they could focus on AI chips. They don’t really care about consumer sales that much, so I don’t think there’s too much of a backlash. It’s not like Google is going to stop buying GPUs because Jensen did this. I don’t think there’s a direct backlash to NVIDIA from that perspective.
At some point, Jensen is going to be called in front of Congress, and they’re going to question the fuck out of him. Up until now, he hasn’t had to go to Congress and provide all his data on where the chips are going.
We track supply chains. We’ve seen chips go to China. In fact, I almost bought a GPU server from China and shipped it to the US, but technically I’d be committing export violations if I shipped it from China to Hong Kong and then to the US. So I didn’t do it because I couldn’t get anyone in the US government to say, “It’s okay.” I just wanted to do it for a media bit.
Mm.
Right? Ultimately, it was against the law for me to ship a GPU server that shouldn’t be in China under US law to Hong Kong and then to the US, because the China-to-Hong Kong transfer is illegal. Shipping it straight out of China would’ve been too risky.
Anyway, I think it’s pretty clear there are chips going to China. We have estimates on how many. It’s on the order of hundreds of thousands. It’s not a small number of GPUs that have made it to China. This is in the supply chain information that we sell. In addition, many more hundreds of thousands are rented to Chinese companies. Probably this can be used as a weapon to crack down on some of these things.
Then the question is: should they crack down on this? I’m sort of of the belief that you don’t want China to have too many AI chips, but you should still give them some and still make some money off of them. Maybe it’s better not to ship chips to China and have them rented from the US, or from companies outside of China. Maybe you make a rule that only US companies can rent chips to them, like hyperscalers, or some other way of doing it.
There was a law at one point that was going to pass related to AI diffusion rules. It would have meant that only approved companies with certain security practices could rent chips to Chinese companies, or to companies headquartered in China and other risky countries. That rule got shut down because it was so complicated. I bet something like that is probably more likely, and it’s probably good from a national security standpoint: we don’t really want to ship chips there, but we’re okay with letting you have some chips.
It’s a cost thing, right? If you buy a chip and deploy it, it’s much cheaper than renting a chip. If you rent a chip from, say, Oracle, Microsoft, or Google—all of whom sell lots of GPUs to ByteDance, for example—it’s perfectly fine. There are more profit dollars that the US economy is making versus China. There are a lot of thought processes around what exactly shakes out.
I don’t think anything will happen fast, though. I think it’s going to take many months for any investigations to happen, for Congress to call him up, and for any wrongdoing to be found, if there is any. There could be a world where he actually did nothing illegal. Yes, he slept with a woman while he’s married, which is very morally wrong, and he sold GPUs that were clearly meant for Alibaba, but they were all installed outside of China. So it wasn’t like an illegal act happened.
That could be the case, but I think he’ll still get roasted in front of Congress endlessly, and that’s when probably worse things happen. Then there will definitely be regulations that come down. It seems like the left wing is stricter on diffusion of AI chips than the right wing is, which is interesting because the right wing came in saying, “Yo, we’re going to be hawkish on China.”
So we’ll see what ends up happening. At least the way the campaign promises are shaking out, the left wing seems to be more hawkish on regulations. When this new administration comes in, even if they can’t pass any laws, they might try to pass something. Then Trump has to veto it, which is obviously going to look really bad, and then eventually there are elections.
For now, I think the ramifications are mostly that NVIDIA probably eventually has more restrictions on selling its chips to China, or to companies that rent them to China, even though that’s probably fine within some limit. The other aspect that would be interesting is that Jensen has done a lot to prop up the ecosystem, whether through favorable allocations to neoclouds, investing in a variety of startups, helping them with software, buying Hugging Face and Poolside and all these other small deals, or backstopping neoclouds.
Does he have less room to do some of this stuff? Is he, or is NVIDIA, distracted in any way because they can’t focus entirely on the business? Does any amount of detraction from the actual work they’re doing slow down how much the ecosystem is propped up, all at a time when no one else in the market wants to prop it up? Credit markets are closing down and all these things. So I think it’s a quite interesting path.
I remember listening to your interview with Dwarkesh, and you guys were talking about how you’d just learned about the credit markets. You were like, “We just spent a month learning about debt and debt cycles,” and that kind of thing. But—
Yeah, people clowned us, but then we did the pod. I think I’d spent a little bit more time, but we were still very noobish compared to people who live and breathe credit. People clowned us, and then a month later, interest rates started skyrocketing, and we were like, “Yeah, clown us if you want. We were right.”
We’ll get into that a little bit later, but what are the biggest questions that you’re getting hit with right now? They could be about Jensen, the new regulation, or the change of the guards in the US government.
Obviously, a lot of people are curious about what I think about this. People in the media have been trying to chase down this lead for a while. I think there have been reporters who asked me even a year ago, and I said, “No comment. I’m not wading into this storm.” I didn’t know that much, but I’d heard some things. I just wasn’t going to wade into that shitstorm.
A lot of people are asking what happens to NVIDIA if Jensen leaves. If Jensen has to leave NVIDIA, what happens to the company? Then what happens with the restrictions and regulations? Does this prop up AMD, Huawei, or whoever? How beneficial is this to competitors?
What do you think?
Oh, I mean, yeah, I think those are fair questions.
Fair?
I think it’s very fair that AMD has free rein to sell to companies in Singapore that are renting to Chinese clouds. If the data center is outside China, AMD could happily sell to them, and no one’s really going to give a shit. But they’re going to give a lot of fucks if NVIDIA does it again. There are probably some deals that AMD, or any other vendor, could win where the ultimate customer is Alibaba, ByteDance, or whoever.
It’s probably most relevant to get into the shift in DC, but also in the US, and more regulations, as you say. A lot of that does come down to data centers, whether that’s pushing them out of states that used to applaud them and now don’t. How are you seeing the stress or pressure from that, and do you see data centers and space as a solution?
I think space is a solution that’s still a few years away, so there’s a lot to say about what happens over the next few years in terms of regulations. It’s at least a few years away in terms of real, high volumes. I know Elon’s trying to do a shitload in 2028, but I’ll take the over on that. At some point this decade, there could be a gigawatt in space. I bet it’s this decade, maybe 2030, if I had to guess.
As far as the space question, that doesn’t solve anything in the next three or four years, or at least not a meaningful amount of capacity. Potentially, space is way more than a few years away, too. Engineering is very hard there.
As far as domestic regulations, Governor Hinojosa in Texas has been pretty positive and constructive in her discussions, I think. We’ve tried to talk to people in her administration and her close circle, and about what she’s said publicly. She says she’s not going to stop data centers. She’s just going to make sure they don’t raise electricity prices and are reasonable stewards of water, which I think most data centers can do pretty easily.
So those two are solvable. The union labor thing is the other one that’s kind of sketchy, because there’s very little union labor in Texas. How do you mandate union labor? Do you tell all these electricians making $200K or more working on a data center, or industrial plumbers making $200K or more at a data center, “Hey, you have to join a union, and you can’t just work for…”? That seems like a very unpopular thing. And that applies to all the other construction crews, right?
That seems like the biggest issue, at least based on what she’s said publicly, what her administration says, and what we see building data centers in Texas. She has a lot of levers to fight back against any data center, right? There are water permits, wastewater permits, stormwater permits, water consumption rules, and air permits. Those are things she can control and stop very easily. If she does, a lot of data centers stop. There are data centers that aren’t even connected to the grid and use very little water, but she can still stop them through the air permit. So it depends on how much of a stickler she is on the union labor point.
I imagine it won’t be that crazy, because there’s a lot of money coming to Texas, a lot of workers coming to Texas, and a lot of industry. It’s not the eye-watering numbers you hear about data centers, because most of that comes from outside Texas, like the chips, networking, and a lot of the materials. But it’s still a shitload of people working in the data center industry: tens of thousands of people, and hundreds of thousands in the broader sense.
We’ll see what happens there, but we’ve definitely adjusted our timelines. We track every data center in the world, and we’ve pushed back multiple gigawatts from 2027 into 2028. Quite a few gigawatts in 2028 are getting pushed out, too. Not because they can’t build them, because the labor isn’t there, or because of some equipment shortage, but because permits just aren’t available. I think that’s the biggest risk.
There are a lot of jurisdictions all around the U.S. The vast majority that have banned data centers have no data centers. Most of the jurisdictions explicitly banning data centers or running moratoriums don’t have any anyway, or don’t have any meaningful capacity coming online. We ran a poll in all the counties that have a 100-megawatt-plus data center already, under construction, or planned for the next 3 years. Funnily enough, 49% of people in counties with a 100-megawatt data center didn’t even know there was a data center in their county.
Oh my God.
It’s pretty fucking hilarious, because how do you not know this? We did this poll with David Shore, who collaborated with our team. Another finding was that only 40% of people knew a data center was even coming when there was a 100-megawatt data center planned in the next 3 years.
I think it’s a function of the fact that, even in the exit polls, data centers weren’t among the top issues in that election. The issues were inflation, war in Iran, and things like that, which are really pertinent to people’s lives. Most people don’t realize there are data centers coming to their county or region, or that there are already data centers there. They just have a nebulous “data centers are bad” view because of things on the internet, both reasonable things and stuff from bad sites, as well as completely unreasonable, fake, astroturfed stuff.
Climate people need something to focus on, and there’s also some level of external actors wanting to push anti-data-center narratives. Most people don’t know about the data centers that are coming. The places that have them haven’t banned them yet. Texas is an example. There are some counties in Ohio that have banned them, but not the state as a whole.
I think it’ll become more and more pertinent, but these companies can do things to make it work, right? We put out a post saying you can just subsidize the power for everyone in a 30-mile radius, and it costs almost nothing. You can subsidize things and just give people money, and it’ll be fine.
So where does the negative narrative come from? I’ve heard amazing stories, too. I know Dina Powell was speaking about this on stage at the All-In Summit. She brought, I think, a city council leader or someone like that for a particular area, and they can now fund all the education in their area because of the benefits and revenue it brings to their county.
Yeah. There’s a county in Louisiana that got a massive amount. They paid out all their teachers, like $20K, because there’s a rule that they have to credit all the property tax they collect back to government employees. They were like, “Oh, shit. Okay.”
If done improperly, a data center can be loud if your house is right next to it. If done improperly, data centers can use a lot of water. But if done properly, they use very little water. All of Elon Musk’s data centers use less water than 3 In-N-Outs.
Really?
Yeah. You have to count the supply chain of cows, but yes. If you count the supply chain of cows—
Oh my God.
—3 In-N-Outs is—
You don’t get an asterisk.
No, I think it’s fair, right? What’s the embedded water usage? We did the same for the data center: all the embedded water usage of the materials in there, including the fabs building the chips, and all the embedded water usage of the power generated from turbines, including getting the oil and gas. Even if you compare the embedded water usage of the data center with that of In-N-Out, 3 In-N-Outs are equal to all the data center capacity Elon Musk has.
That’s just because agriculture wastes a lot of water, and no one cares about that. But if a data center takes in that water and it evaporates for cooling, people freak the fuck out. There are ways to use water in data centers that are very inefficient, but no one does that.
Yeah.
Likewise, I think there’s a concern that data centers don’t return much to the community if they’re done incorrectly. If you give them a property tax break and they’re grid-connected, it’s possible they could cause grid prices to increase. They might not give much property tax back, and almost no one works at a data center. So it’s great while they’re being constructed, but afterward, who cares? Then there’s a big boom and a bust.
Most people aren’t paying attention to that. I think most people just don’t like them because they’re a little loud. If you’re close to them, sure. But if you’re not close to them, there’s no noise. That’s the whole point of that survey: there are 100-megawatt data centers, and 49% of people don’t know there’s a data center in their county. It’s not like it’s some small one.
Using it also kind of leans into the huge negative aura and narrative around AI. Data centers are the physical thing you can actually protest—
Yeah.
—versus software.
I agree. I think there are also a lot of incorrect numbers. One of the books on the New York Times Best Sellers list overestimated the water usage of data centers by about 1,000 times.
They revised it, but no one pays attention. They all just cite this.
No one—
Yeah.
They just cite this, and it’s like, okay. I think people dislike AI more than they dislike data centers, but data centers are an easy thing to point at.
So let’s talk about hating AI. We have some awesome, huge IPOs coming out. SpaceX had a big one earlier this year, and they’re now an AI company. We have Anthropic coming up. If this doesn’t land around that time, it’ll be before or after—who knows? And then we have OpenAI. We had Cerebras earlier this year. There’s a bunch of stuff happening in the AI world.
I recently had Apoorva Agarwal on the pod, and we talked about his four seasons of AI. He has this whole thing where Google will be in vogue, then Anthropic, then OpenAI, and then—what was the fourth one? SpaceX. How do you deal with the juggling of who’s the top dog at the moment?
The thing is, the market doesn’t actually shift valuations that much. People talk about how this company is doing well, then this company, then this company, but the market doesn’t go all over the place. It smooths the curve.
Anthropic obviously crushed it in Q2, and then OpenAI crushed it in Q3, while Anthropic did pretty poorly. They did okay in Q3. Now they’re back to crushing it because of Opus 4.5 and GPT-5.2. OpenAI keeps delaying model releases, and then Google will get back in. These are competitive markets. People like us, who watch this stuff and are terminally online, over-index on any individual data point. That’s just human nature.
Are we still watching model releases, or are we watching applications now? It seems like Meta is winning with the release of—it’s not called Jolly. What is it called?
Muse?
Muse.
Did you name your Muse Jolly?
No, that’s what it’s called. I’m just obsessed with the little—
The bear? The thing is called Jolly?
Character.
Really?
Yes.
The white thing?
Yeah.
Oh, I didn’t realize that.
I don’t know if it’s a bear, either. I don’t know what it is.
It’s a teddy bear. It looks like a teddy bear, right?
Okay.
I don’t know. It’s not a bear. It’s a teddy something.
It’s something.
Yeah.
It’s something fluffy.
Yeah.
But—
The fluffy thing. Got it.
The fluffy thing—that’s Muse. How do you handle Muse coming onto the scene, with AI applications starting to pick up some of the momentum?
Applications are definitely important to watch. It’s pretty clear that Instinct and Muse, despite being worse models than OpenAI’s, are way better products than OpenAI’s. I think some of OpenAI’s safety focus restricts them. They have this thing called Guardian, and it stops the model from doing certain things in Codex and Dots, often overzealously. So Dots is just a shitty product compared to Muse and Instinct. Manus and Muse will run with what you want them to do.
There’s something to be said about being safe, but good products need to be good products first. If it texts someone incorrectly or books an appointment incorrectly, fine, whatever.
What if it cancels something important or says something bad?
Yeah, let me roll the dice.
Oh boy. Okay.
I don’t know. Obviously, that’s incorrect. But you’re not hearing about Muse and Instinct doing this all the time, so I think OpenAI is just on the overzealous end.
Products are very important. Throughout the year, our spend has seesawed all over the place in terms of what we’re using, with the majority being Claude Code for most of the year. But over the last few months, we’ve gone from zero persistent agents to 45% of our $50 million a year AI spend going to persistent agents. The vast majority of that isn’t through Anthropic’s Claude Tags or OpenAI’s Slack agent or Dots. It’s through Perplexity Computer, which is still using Claude and OpenAI on the back end; it’s just a better harness.
I don’t know if there’s much of a sticking point for us long term. We’re a small, nimble organization that moves fast. But it shows that products are really important. Instinct doesn’t have the best model. They’re primarily using open models, and yet OpenAI is terrible compared to them. Product is very important.
I think products will be stickier in general. Muse will probably be a lot stickier, and Instinct will be a lot stickier. People will index to those use cases. I don’t think Claude Code and Codex will be that sticky, because developers can switch easily. Enterprise contracts will be sticky, though. I’ve seen Anthropic offer enterprise customers a discount if 70% of their AI spend is with them, and no discount if it isn’t. Something like that is very sticky.
We’ll see what sort of stickiness happens. I do wonder how much of a problem it is that Anthropic and OpenAI are bad at product. Have you seen the OpenAI chat app? It’s so dumb. You open ChatGPT, and there’s a chat section and a work section. In each one, you can slide between models. You can choose an instant response, or you can choose extra high, pro, or medium reasoning effort on the chat side. On the work side, it selects the models, but there’s no pro option. It’s like, what the fuck is this?
I think most people just log in and start using it. They have no idea you can toggle between models. Genuinely, I think probably 90% of people—
I agree, but why do you have all these options, then?
I don’t know.
It’s bad product. This is a billion-user app.
It’s a fair point, because Tesla will at least take things off the production line. Why don’t the model companies do that?
Meta is good at product, and Muse is intuitive. I was able to beta test Muse. Alex texted me, “Hey, try this new product out.” It was the week Instinct was popping out. He said, “We’ve been working on this for a while, but try it out.” So I tried it, and I was like, “Fuck, this is pretty good.” The UI was clear and good. That’s not the case for OpenAI and Anthropic, so I don’t know what they’re doing.
It’s this cycle of hiring a Meta VP who wasn’t actually that important at Meta, creating a bunch of garbage, then hiring a new Meta VP. I don’t know what’s going on. But yeah.
As the market shifts into applications, do you remain bullish? I know your sentiment changed a little bit. What’s your temperature check?
Sentiment on what?
The ROI. Is it there, given how much these companies have to spend to grow, apart from the big leader?
Most startups will fail. That’s just a fact of life. But some of them will be very successful. It seems like Manus has some staying power. They’ve gotten a lot of users. I don’t know if it’s worth the valuation they’ve got, but they’re kicking ass. They have a legitimate shot at staying power.
Companies like Sierra and Harvey are growing revenue really fast. Maybe their customer acquisition cost is really high, but they’re solid, and they’re in there. I think those companies have ROI. Other companies may not. Are inference providers worth the valuation multiples they have? Potentially not. Or maybe they are, because they’re growing revenue so fast and their margins are good.
So it’s just a question of—I think most startups will fail. I think some will succeed. I wouldn’t want to rain on anyone’s parade. I have my own thoughts on each company. Most of these companies are doing badass shit, right? The market dynamics are such that you have to pivot and move really fast. Some application companies will work. Some of them are just pure pass-through, though, right? I don’t know if you saw Harvey’s blogs. They showed their margins have just seesawed all over the place. Or if you’ve seen, “Yes, 45% of my company’s AI spend now flows through Perplexity,” but I don’t know what Perplexity’s margin is on that.
Yeah.
I don’t think it’s that high, because if it were a more expensive way of consuming Claude than Claude itself, then I probably wouldn’t switch off of it, right?
This is great. You’re really getting into the investor seat, which will lead us into our next topic: your starting of a fund. You started one.
Yeah. Already started it.
So why extend out? You’re building out an empire here. You acquired Citrini. You started a VC fund. What’s the goal here? How are you building out something analytical to extend?
1. Why Start a Hardware Fund
I think there are many ways to—I guess, for me, what are the base goals in life? Maybe “base goals” is a crazy thing to say. What are the things I want to do professionally and in life? I want to have fun. I want to make money. I want to know everything, which is obviously a crazy goal and impossible. I want to spend time with those that I love, like my family and stuff. And I want to influence the people and things that matter.
That makes sense if you’re an analyst in a research firm. You’re influencing people based on your research and opinions and the facts you provide. And you’re the closest to knowing everything, although obviously it’s impossible to know everything. There are many ways to monetize that, right? There’s selling research, there’s doing consulting, and that’s the bulk of our revenue. It’s either these data sets that we sell to people or consulting projects we work on with hyperscalers or infrastructure funds, or what have you. We work on these one-on-one consulting projects, and that stuff is sort of firewalled away from the research.
Another way is: if you see something awesome super early, we started an early-stage venture fund. We raised $500 million, solo GP. We’re focusing on seed and Series A. That’s about all I’ll say. It’s in the infrastructure space, all the way from robotics to silicon and optics and manufacturing, and all these aspects of hardware and networking. There are many different places we’re investing, but all early stage.
We’ll never use the platform I’ve built to pump the companies I’ve invested in. But there are many ways to help them, right? I have 115 people who work for me across SemiAnalysis and Citrini. And now the third company is the fund, which is separate and firewalled away. But there are many ways to help people. If a company needs help, maybe we sell them research, maybe we do consulting. If they’re super early stage, we can invest and give them some help pro bono. There are a variety of different ways to help people. The key thing is: I want to know everything, I want to have fun, I want to make money, I want to influence, and obviously spend a lot of time with those I love.
So as you think about that, there are obviously so many advantages to having you as an investor, and there are obviously some conflicts with that too. Most investors do have their own thing or their own company. You mention that you have a lot of influence and knowledge, and it’s clear you’re compiling that together for the fund. Why did you choose the hardware side of it?
There’s some software stuff too, right? Software for hardware, software for infrastructure. I think I don’t know shit about applications, so why would I invest in applications? I don’t know how to scale an app to a billion active users. Why would I be useful? I could be useful to an Instinct, right? Help them think about infrastructure and cost or whatever. I met the founder recently, and we talked about some of the stuff, but at the end of the day, I don’t know shit about the product, and I don’t know how to scale users. So how can I? I don’t think I could even make the determination that he’s going to succeed or not. Seems smart. They seem like good people, but I don’t know. How am I supposed to invest in that?
Whereas in my company, SemiAnalysis, I have people who have worked at ASML, Applied Materials, Lam Research, all these companies making equipment for semiconductors. I have people who did R&D at TSMC and Intel. I have people who’ve designed chips and worked on kernels at NVIDIA, people who’ve done infrastructure at Google, Microsoft, and Amazon, and people who’ve worked on training models. People who’ve worked on compilers. People who’ve built data centers. People who’ve built power plants. I have people who’ve done all of these things in my company, and I can tap into that expertise.
I think we as a firm know more about data centers than anyone else. I think we as a firm know more about where and how power is being generated for these data centers behind the meter. I think we know more about chips and all the aspects of chips and software, and evals, right? We’ve started testing eval models with GDPX. There are all these different things that I think I have expertise in as a company, or that we’re trying to be experts in, and we’re trying to sell data and do consulting. But it’s not application stuff. It’s models, infrastructure, hardware, and places where we’re aspirational, like robotics. No one really knows the whole robotics supply chain, but we’re trying to be experts in the supply chain. Hopefully we’ll be the experts in it, and then it would make sense, right?
Anyways, I don’t understand investing in shit you don’t know anything about. I think one of the most toxic traits about investors, whether in San Francisco or New York, is that they think they know everything about something they don’t know shit about.
They don’t know anything about.
They don’t know shit about. Yeah. And I don’t know shit about a lot of things.
I think I know some about the things I’m investing in, and I think I know more than the vast majority of people. But even in places I’m investing, I’m learning stuff all the time. If I didn’t know anything, I’d be so lost. How do you invest like that? Obviously, you invest in founders, and then you invest in ideas. A great company has both of those things. I think in hardware it’s harder to just have a great founder and have them figure things out. You kind of have to have a great founder and a great idea. Whereas in a lot of the application software world, I think you can just invest in a great founder, and they’ll figure it out.
So what does early stage mean to you, though? What’s the check size you’re going at? It’s such a sliding scale.
I’ll keep that confidential.
Confidential?
Yeah. Why would I tell people what I’m doing? That gives— that removes optionality for me to do something different.
Maybe someone wants your money.
If they’re awesome, they’ll find me, or I’ll find them, right? Hopefully.
I guess it’s a fair point. So are you doing a lot of outbound sourcing, or is it kind of coming to you? Do you just get it through your cabal?
“Cabal” is a strong word.
I don’t really know if I’m even using that word right.
I think a cabal is a group of people who are conspiring and, you know, evil. It has some evil connotation to me.
What’s the non-evil version of that?
I don’t know.
Your friends.
Yeah, my friends—
Okay.
—and team.
Besides your friends.
Friends, team. I mean, I don’t know.
There are 115 people who work for me. If they see something interesting, they’ll refer it.
Oh, you’re going to get inundated with pitches now.
Yeah, but most people’s pitches are boring and wrong. That’s the broad view I have.
Yeah. VC life.
VC life. But anyway, I think there are a lot of good ideas out there, a lot of interesting work, and a lot of innovators. We’ll find them.
Fair.
The check sizes—I don’t want to talk about the specific details.
Okay. We’ll leave that one to the imagination with a $500 million fund. Congratulations on that. I know you don’t like to talk about it, so I appreciate you sharing. In terms of Citrini, can we talk about that? What was the thought process there? It seems like you guys sometimes conflict in how you view the world.
2. Citrini’s Independent Voice
Yes. They’re bearish on AI. That’s okay. If they want to be bearish on AI, that’s fine. James is awesome, and the whole Citrini team is awesome. They’ve captured the market’s imagination. They’ve been right a lot of times, and their way of doing research is very different from how SemiAnalysis does research.
Let’s operationalize it. Let’s help them be better and improve that. I think there are a lot of things they can do better, from how I see it, while still holding on to their ideal of who they are in their research. So it just made sense to join forces with them and help them with the things we can help them with.
It’s a separate company. It remains a separate company. Kyle, their director of research, has the most say over what gets published. At one point, when we acquired the company, they asked, “Are we allowed to be bearish?” I said, “Dude, do whatever you want. I’m not here to change your voice. I’m here to help you run this as a better company. Make it a better company.”
I think the team has been pretty positive about that. They’ve been nailing a lot of calls. They got bearish on AI months ago, and I thought that was crazy.
They’ve always been a little bit bearish on something, but I remember when the Citrini paper broke out. That was probably February, right? And it shook the markets. That became a whole topic, and that was just one thing, but it had ripple effects.
I think some of the things they said have come true. They said it was a thought exercise, right? They said it wasn’t necessarily what they believed was going to happen. Anyway, they’ve been super bullish on AI, but then they flipped and were super negative on AI a few months ago.
I’m still super bullish on AI in the long term, but our jobs are very different. At SemiAnalysis, we track the exact infrastructure: every data center, every chip. We test the chips and the models. We test the models on biology, security, cyber, chip design, and all these different aspects. We test clusters, and we track every bit of the supply chain: how many cables are shipping, how many servers are shipping, how many GPUs, how many tools are shipping for the fabs, and what the price of memory is. SemiAnalysis is trying to answer all these questions on a very technical and numerical basis.
Citrini tries to be technical. They wanted to know what was going on with AI drug design, so they went and developed a scorpion antivenom with AI. It works okay. It’s pretty surprising. They wanted to know if oil was going to stop being shipped, so they sent someone to the Strait of Hormuz. They’re going to keep doing crazy stuff like this.
It made a lot of sense. I’m not going to make them part of SemiAnalysis. It’s a separate company.
Mm-hmm.
They’re awesome. There’s a lot they can do better, and we want to help them do better.
I’m really curious: as you build out these multiple businesses, who’s a business leader you look up to? Are you going off someone else’s playbook? How have you learned?
I don’t think anyone’s ever done what we’re doing. There are tons of business leaders you can envy and idolize. Calling back to the earlier part of the conversation, I used to say Jensen was one, but I don’t know if I should say that anymore. I don’t know if I believe that anymore.
Aw.
It’s very sad. You idolize the man, and then you’re like, “Fuck.” The same could be said for Elon, right? He’s a tremendous businessman, but he’s done some stuff in his personal life, and I’m like, “What the fuck?”
At least he owns it.
Yeah, he owns it.
It’s not like he’s pretending to be a family guy and then blatantly not, you know?
Yeah, yeah, yeah. There are a lot of very interesting businesses across history and humanity, but ultimately no one has really done the mix of things we’re doing, so it’s uncharted waters.
Media is obviously being completely rewritten. Information services businesses are being rewritten. Everyone recognizes that Gartner and IDC and those kinds of companies suck and are useless, and yet information is more valuable than it’s ever been. What does a new kind of information firm look like?
Media and attention: the old guard is, in many ways, bad. There are still awesome reporters at places like The Wall Street Journal, Bloomberg, or The New York Times, but those brands have definitely taken a huge hit and fallen off in quality a lot. There’s the creator economy, and there are people who are big on YouTube or another platform like Twitter, X, Instagram, or Substack. Attention is important. How do you monetize attention? How do you not just attention-hack, but also provide useful things?
Through the SemiAnalysis newsletter, the Citrini newsletter, and their social media, there are a lot of valuable things you can put out there. What’s the right way to monetize attention? If you look at Gartner, the way they monetize it is like a racket. They’ll say, “Pay us, or we’re going to rank you poorly.” It’s not that explicit, but you could look at their AI software agent index. They put IBM’s coding agent—whatever the fuck it’s called—above Cursor or Claude Code. I don’t remember which one.
I don’t know.
I could be sleeping on IBM. They have the Magic Quadrant, right? How the fuck is IBM’s coding agent better? Whatever. They’ve done this for a lot of companies. The right way is obviously not to do that kind of bullshit, because then you lose your credibility and no one cares what you say.
One thing we will never do is take a paid partnership or take payment for an opinion. We’ve never done ads. We never do any of that shit. Our biggest customers are the ones we criticize the most. My biggest customers are hyperscalers, major semiconductor companies, and AI labs, and I don’t think there’s one that we only say positive things about.
We fall in and out of love with Anthropic, and I’ll say that clearly. We’ll say positive and negative things about them, and the same goes for any of these companies. For people to care what you say, you have to be objective and say what you truly believe. That’s sometimes antithetical to the business. We’ve had people try to pull their contract because they were mad at what we said, but then their team said, “Well, no, actually, we still want their data.” It ends up being an interesting thing: you have this attention, and you have this business where you’re selling data and consulting services.
People only care about your data and consulting services because you’re an objective, truth-seeking player. You could monetize the attention by selling a paid opinion, but if you do that, you lose the whole game. You always want to give your own opinion, and you can be wrong sometimes. I think I’ve totally been wrong sometimes. But my whole team and I will always say what we believe, even if we’re wrong. I think that’s a little different. We’re a lot more cavalier.
We’ve said things like, “We think some of Nvidia’s practices are anticompetitive,” with regard to how they treat certain neoclouds. We’re not going to shy away from that, even if they tell us it’s false. We’ve seen plenty of data points that suggest otherwise. We’ve also had opinions like, “OpenAI’s product sucks in many cases,” or, “Anthropic’s product teams suck in many cases.” What’s going on? Why is Perplexity my primary persistent agent supplier? They’re doing a lot of things right, but OpenAI and Anthropic are messing up. And at the same time, we can say, “Wow, these are the best companies in the world, too.”
As long as people believe our opinions are true to our word, there will always be people who cast shadows and doubts. They’ll say, “You’re just saying that because someone paid you to.” Sometimes we’ll talk about something publicly, and then our research will say something completely different. For example, we’ve said Iris Energy, or Iren, has terrible technology on the cloud services side. Their clusters can be unreliable, and they don’t have all these other things in ClusterMAX. At the same time, we’ve said, “They have powered land, and that’s worth a lot.” We’ve said both of these things.
Or take Fluidstack: I have a large investment in the company, and we’ve still said publicly that its data centers are delayed. You don’t have to compromise your morals or beliefs. I think a lot of VCs believe they have to be the tail wagged by the dog that is their companies, but it doesn’t matter. Good founders recognize that I’m going to give them honest feedback and tell them what I think is true. If they disagree, that’s fine. There’s no animosity. As analysts, I think we criticize the things we love the most. Just because you’re invested in something or have a financial interest doesn’t mean you have to be the tail wagged by the dog.
So is that the biggest difference you’re going to make as an investor?
Oh, yeah. I think I won’t shy away from it. For one, we’ll never publicly promote the companies we’ve invested in, because those are separate companies.
Mm-hmm.
The fund is a completely separate entity and company from SemiAnalysis and Citrini. They’ll never say anything to pump a company on the venture side. At the same time, we have all these connections through those companies. I just happen to know people in every layer of the supply chain, and I can help you get introduced to them, hire people, or whatever.
There’s a difference there. If there’s information I have under a consulting contract, there are compliance rules. We spend a lot of money on compliance. You can’t just willy-nilly do stuff. Even if I’ve invested in a company through my fund, or that company is a big customer of SemiAnalysis’s data or consulting, that doesn’t mean we won’t say what we think when we’re testing something empirically or tracking a supply chain number. If we think the technical direction is wrong, we’re going to say that. I think that’s the key point.
One thing I think might need to be added to benchmarks is spy benchmarking.
Yeah.
I think we need to know which companies are most vulnerable, which might have the most spies in them, and from which countries. Just an idea.
Yeah, I don’t know how to check that out.
I don’t know. You’re the expert here.
Not on spies.
Well, you’re in San Francisco.
Yeah, and?
Are your customers not coming to you asking for spy protection?
No.
No?
Why would they ask me for spy protection? They want to know what data centers Amazon is building, how good a model is, or what a networking optical cable costs. That’s the kind of stuff they want to know.
They don’t want to know how you protect your data center or your chips from spies?
No.
No?
No.
Okay. Maybe we’ll check back on that one.
Maybe not yet. I’m sure it’s a valuable service. Who would do that? I mean, the OpenAI board has this CIA or NSA guy. That seems pretty cool.
They have to. That makes sense.
He’s just on the board, which is pretty cool.
What does your board consist of?
Me, and Doug sometimes.
Okay.
Yeah.
Before I shift into spy benchmarking, I wanted to ask you—
Because there’s an LPAC. That’s like a board—
The what?
There’s an LPAC. It’s a board for a fund, I guess.
Yeah, there is.
Not really, not really, but—
Well, kind of. I mean, that does count. Before I went into that, I was going to ask you: I do think you’ve built a really unique business. You’ve built a phenomenal business in quite a short amount of time, and you’ve ridden an incredible market. It seems like you’re creating some durability with the fund and other things, and with how you’re planning research and pricing things. Along this journey, I can’t imagine it’s all been great. What have been the hardest lessons you’ve had to learn, or the most challenging moments?
Running a company is very hard. You lose sleep, so that’s fun. But it’s a lot of fun. You miss a lot of things. You have to choose what you’re going to work on, so you’re going to miss a lot of things, whether that’s parts of some people’s lives, your own life, sleep, or health. Obviously, that stuff gets sacrificed sometimes.
It’s also really tough when you have to make tough decisions. When you’re running a business that does a lot of revenue, you have tons of tough decisions, whether it’s negotiating with a customer or team members, or managing conflict between team members.
Kittens.
Yeah, we had a kitten in the office. Something that was really cute. But there are tough decisions to be made all the time. I think that’s another difference: a lot of investors have never made tough decisions. “Do I invest in this company or not?” That’s one thing. Have you fired people? That fucking sucks. Have you been sued by someone you fired? That fucking sucks.
These things are annoying, and you have to learn how to deal with them and know the playbook. How do you negotiate with a customer or a supplier? How do you hire someone? How do you convince someone to work for you? How do you keep everyone motivated? I don’t know the answer to these things. You just kind of fly blind.
Maybe I’ve done it a few times, but I still haven’t done it much.
But you’ve built it out to how many offices and how many employees?
We’ve got people in 15 countries. There are about 115 people. We have offices in New York, Chicago, San Francisco, Oregon, Singapore, and Korea, and people in a bunch of other countries.
You speak all those languages?
Fuck no. I speak English, Gujarati, some Hindi, and Spanish. I’d say I can get by in two of those languages. I’m decent; I can converse, but I’m not good. No, I don’t know that many. I wish I did. I know enough Chinese to make people laugh.
Really?
Yeah.
Is it because your Chinese is so bad?
Well, that's exactly it.
Okay.
It's bad, and I only know phrases that are insane.
Really?
Yeah.
Should I dare to ask which ones you know?
No, I mean, it's up to people to ask me. There are some old Chinese proverbs, and some things you can say based on the body language of what's happening. Even if you don't understand what the fuck is going on, you just say this phrase, or you say the word, like… [unintelligible]
An easy example is, you hear some people and there's a certain intonation when they're speaking, and you say [unintelligible], which means kind of like, “What? Why?” People will be like, “What the fuck?” It kind of makes sense, and they'll think you know Chinese. You don't know Chinese. But there's obviously much more complicated stuff.
Okay, so if I wanted to—
You can talk about cows playing piano—
Oh.
—and how you can't share an umbrella with a fat person. There's all sorts of things like this that you can say that are really insane.
So if I wanted to show that I can hang, what should I say?
Don't ask me. I don't know shit.
But you just said—
I lived in Taiwan for a while, and I don't know. It's got context. If you learn it, then I'm giving you the alpha, and the alpha's decayed. It's leaked.
All right.
It's over.
Whatever, man.
It's over. You gotta come up with your own wizardly spells. Your sorcery.
You're right. It all comes back to sorcery. It's a good circle-around moment. Okay, so I think we're just about at time, if not more. Dylan Patel, what are you most looking forward to in the next 12 months?
I think there's a ton. There's just a lot of stuff that we're launching and doing. There's testing of various chips, including Huawei's chips, Amazon's chips, and various AI startups' chips, on inference performance.
The new model releases are sick. We launched a new benchmark called GDPX. It tests across many domains: cyber, chip design, hardware design, materials science, bio, quant, and RSI. I'm excited to see the models get good at them, because the models suck at most of these benchmarks. They're very complicated tasks. As the models get better at them, what does that unlock for society?
I think we're gonna see tremendous progress in technology, especially hardware. I think hardware's gonna accelerate like crazy. In a year, the number of companies making things in the hardware realm is gonna skyrocket, and the number of companies that can do it well is gonna skyrocket. It's gonna be really fun to see all the innovation in society from that. Robotics, obviously. There are tons of niche semiconductor technologies that I'm really excited for, especially on the packaging, networking, and memory sides. So, yeah, I think there's a lot of things I'm excited about.
Cool. Is your favorite chip still Takis? Is that what you said?
Takis.
Takis? Takis?
Takis. Takis.
You know what that is?
T-A-K-I-S.
I don't.
Takis. The blue ones, or purple ones. I don't remember what color.
They're purple?
They're blue or purple. I'm partially colorblind.
Really?
Yeah. They say it's red-green colorblind, but I can tell the difference between red and green, and I don't fucking understand that. It's a really interesting one, right? It lives on the X chromosome, the type that I have. I can see most colors right. It's fine.
So this has been clinically confirmed?
Yeah. It lives on the X chromosome. It can only express itself in men. It goes from the maternal grandfather to his daughter. His daughter always has it. If you draw a Punnett square, that daughter will pass it to 50% of her sons, because she has two X chromosomes.
Wow.
For some reason, if it goes from the woman to her daughter, it can skip a generation too. But it lives on the X chromosome, and it generally comes from your maternal grandfather. It's an interesting thing: I see color a little bit worse, but I see way better in the dark. I think it's rods versus cones, the light detectors in your eyes. Your rods are better at figuring out whether it's dark or light, and cones are better at seeing a spectrum of color. I can see every color around us, but my color skills are a little worse.
What color is this?
It's red.
What color is this?
It's blue.
Well, then you can't see purple? Is that what it is?
No, I see purple too. I just see purple and blue as similar.
Oh, okay.
But I'm also just kind of dumb.
No, I don't think so.
No, I think so. I think stylistically, right?
You're stylistically dumb?
Maybe.
Maybe less aware.
Less aware. There we go. Stylistically less aware.
There you go.
That sounds much more sophisticated. Anyway, I think it's a fun thing to be able to see better in the dark.
How did we start talking about this? I don't know. All right, Dylan Patel, thank you so much for your market update.
Yeah, for sure.
It's a lot.
Thanks for having me on the Wizards of Waverly Place.
I know. I'm gonna steal this room from now on. I think it's very fitting.
Awesome. Awesome.