How compute and AI will create next-gen superapps

Chips, NVIDIA, and Intel with Dylan Patel

The launch of OpenAI’s GPT-5 has ushered in a panoply of views on the future of AGI and end-user applications. Does the platform’s aggressive router presage a future of lobotomized AI responses driven more by compute efficiency than quality? Will new chip models be able to make up the difference? And how will OpenAI, which recently hired Fidji Simo from Instacart to become CEO of Applications, expand its revenue beyond API calls and consumer subscriptions?

These are huge questions which will ricochet throughout the tech economy. Thankfully, we have a veteran hand this week to go over it with us in the form of Dylan Patel, founder, CEO, and chief analyst of Semianalysis. He’s the guru everyone reads (and listens to), covering the intricacies of chips and compute for a readership of hundreds of thousands of subscribers.

Joining me and Lux’s Shahin Farshchi and Michelle Fang, we discuss the questions above plus how Mark Zuckerberg is transitioning Meta Reality Labs, the hopes and dreams of new chip startups, the future of AI workloads, and finally, Intel after the U.S. government’s purchase of 10% of its shares.

This interview has been edited for length and clarity. To hear more of the conversation, please subscribe to our Riskgaming podcast.

---

Danny Crichton:

Your beat has to be one of the most exciting in the world right now. There’s so much going on, most prominently the U.S. government’s investment in Intel. But that’s not all, right? Everyone from the hyperscalers, to materials companies, to U.S.–China, everything is coming to a head. What are the top priorities you are seeing right now? What are people not paying attention to?

Dylan Patel:

As revenue ramps for these labs and they continue to complete these massive rounds — whether OpenAI’s $30 billion round or Anthropic’s most recent round, which is well north of $10 billion — it tells you what their spend is going to be for compute. By the end of 2026, OpenAI could be spending anywhere from $50–70 billion on compute. Anthropic could be spending $30 billion, which is a humongous step-up.

Each of those companies will have more infrastructure deployed through their cloud partners than the clouds themselves had deployed as a whole when ChatGPT was released. So the level of spend here, the level of infrastructure build out, it really is mind-warping.

You get a lot of folks saying, “Oh, there’s no ROI here.” But actually, the ROI is there in terms of the revenue curve. Whatever dollars they’re making from services and inference are actually being spent on compute, of course, but a big portion of that is being round tripped to training.

Danny Crichton:

A few weeks ago we got GPT-5. And one of the most important facets of GPT-5 is it’s really not just one model. It’s a router across multiple models, ones that are very compute-efficient, some that are sort of in the middle of the performance curve, and then Deep Research and others that are extremely compute intensive.

To what degree do you think chip efficiency drives the cost curve down for OpenAI, Anthropic, and the other LLM models versus algorithmic improvements, not just to the inference model, but to these routers?

Shahin Farshchi:

I can take a stab at that. Ultimately, it’s hardware and software that go together to build these kinds of advantages. You made a comment, Danny, on chip efficiency. I would take that a step further and focus on system-level efficiency. You can make the chips super efficient, but if they’re not being utilized, then you’re not getting a return on your investment.

A lot of the more recent innovations are not only around chip performance but also memory performance, system-level performance and ultimately, chip utilization or investment utilization. I think there’s still a lot of opportunity there. Nvidia has done a great job with Blackwell, but I think there are greater improvements to be made. And there are a lot of great startups, a few of which that we’ve backed.

Dylan Patel:

The other aspect of this is that it’s really hard for these companies to balance their infrastructure today because of how bursty the traffic is, whether it be to Deep Research, whether it be Claude code usage being so high during certain hours. Volumes of all things on the internet undulate a lot throughout the day, but AI is even more spiky. And then you layer the constraints of service-style plans — say $200-a-month or $20-a-month type plans — or even free plans, and you have a double whammy there.

How do you deal with that? For OpenAI, it was adding the router. For Anthropic, if you’re using Claude code at night, they upgrade you to Opus, but if you’re using it during the day, they make you use Sonnet. And they have these hourly rate limits for that specific reason. Another way of doing this is what OpenAI does with “batch query,” where you can get half price off the API call if you batch it.

The nice thing about AI is that a lot of it is not latency sensitive, right? It’s not like search. If I use a thinking model, or if I’m using these sorts of applications that aren’t immediately responsive, there’s actually some time to ping a server on the other side of the nation or all the way to another country. So that’s another aspect of this infrastructure build out that’s really noteworthy.

Danny Crichton:

I don’t know if it’s apocryphal or not, but the argument for Amazon Web Services was essentially that Amazon had this massive surge of traffic during the winter holidays, when they had to double or triple their infrastructure. And then what do you do with all this hardware that you’re not using for the other 11 months of the year? The argument was, well, we can make that a platform, we can sell it off.

But how do you balance between the kinds of AI algorithms that you can batch, you can do overnight — “Hey, go through my data, I want a report the next morning” — and those that need an immediate response — “I’m on a call with a customer support agent, I need an answer in the next five to 10 seconds”?

Dylan Patel:

That’s a critical question, and it is hard to answer. OpenAI had to do so much work to get their voice mode to work. If you look at other voice modes out there, one way they solved it is they make the beginning section of the query go to a dumber model. They can respond faster, so you immediately get a response.

A lot of times, even humans throw “ums” and “yeahs” in the beginning of a conversation so you get time to think about the answer. If it’s a voice agent, it’s the same thing.

If it’s a robot, maybe the high intelligence is doing the route planning, but the on-device model for the robot is actually doing the fine motor feedback.

Danny Crichton:

Let me pivot the conversation, because I think GPT-5 is opening a new era of development for these companies. OpenAI is almost 10 years old. They have a product that’s used by hundreds of millions of people, arguably a billion if you include a lot of very soft users. That’s the fastest growth of any consumer app in history.

And so there’s a huge question of, as these companies are optimizing their models for performance and efficiency, how do they start to adapt to consumer use? Do you see an evolution of the business models?

Dylan Patel:

You can already see certain things changing. With Anthropic, it initially it seemed like $200 a month was easy. Then people used it way, way, way more. There were people on Reddit showing $20,000 worth of usage in a single month versus the $200 for the plan. And so they had to introduce rate limits, and they’ve had multiple iterations.

Pay-per-usage is really, really the hope. You pay per usage, you get some margin on that. But at the same time, the subscription model is the one that customers understand more.

The monetization method is where there’s a lot of work to be done, especially in the value-capture side. So I think it’s truly an open question as far as $20 a month, two hours a month, they’re going to release a $2,000 a month thing — and I’m going to buy it for every single person who works for me. I just know that’s going to happen. But it’s tough to say exactly where it goes from here.

Michelle Fang:

To double-click on the business model, Dylan, do you see a world where AI interfaces like GPT-5 become superapps that end up both serving ads but also becoming the platform that facilitates purchases?

Dylan Patel:

A week and a half ago, two weeks ago, we had a post that said GPT-5 sets the stage for ad monetization and superapps. So many people saw that title and angrily DM’d me like, “They’re not going to do ads, we’re not going to do ads.” It’s like, no, but that’s how you clickbait people into reading it. And then you explain the superapp. Their current CEO of Applications built AI monetization via purchasing on Shopify.

And what is a superapp if not something that can do everything for you? So you open up WeChat and it actually lets you book an Uber and it lets you book a nail salon appointment and it lets you get a haircut and it lets you pay people and it lets you buy things and it lets you chat with people. Obviously, it lets you do everything, and you’re going to do everything through AI. That’s the hope. That’s what Meta wants to do. That’s what OpenAI wants to do.

I think if you gave an AI app the option of delivering something to your home — like, “oh, hey, I bought you a pizza” — the user would love it, and then you take a “take rate” on it. I think the take-rate model has to be the easiest way to monetize the user. You give them your credit card information, you authorize them to take X percent of any purchase, and then that’s it. Just like Uber does, just like Airbnb does.

Danny Crichton:

We’ve placed a huge amount of focus on the application layer, and we’re talking about the infrastructure layer. But there’s also an innovation layer that goes across all of them, which is all the new chip companies. So we’re in a world in which Nvidia is one of the most valuable companies that’s ever been created in the history of humanity. But at the same time, you see this massive expansion of investments into a whole slew of additional companies trying to specialize in other parts of the stack, trying to compete directly within Nvidia, or just find their little niche.

When you look at the startup landscape right now, what are you seeing?

Dylan Patel:

People want to do everything and capture the entire value of, say, an Nvidia, and compete directly with them. But at the same time, if the infrastructure spend as a whole is skyrocketing, right, the majority of that goes to chips. Over time, the hope is that a lot of this moves away from Nvidia to other vendors, whether it be Google’s own chips, Amazon’s own chips, Meta’s own chips, et cetera. You do start to get more proliferation, OpenAI has a chip team. And then you could be a vendor for those firms as well.

Danny Crichton:

So you do a Shopify to Amazon. Amazon is a one-stop shop, and then you create a platform that empowers a whole ecosystem of new shops, or something like that. You take a different model, a different approach.

Dylan Patel:

Correct. And Shopify was very successful, so why don’t you do this with chips, right? There are many customers. Nvidia can be your customer, AMD can be your customer, as can all these other hyperscalers.

Danny Crichton:

Shahin, when you think about the startup ecosystem, how are customers picking from a wider ecosystem of companies?

Shahin Farshchi:

Customers are spreading the net very, very wide.

But one point to keep in mind is that a lot of startups like to talk about unseating Nvidia, but you don’t necessarily have to unseat Nvidia to build a massive company. Go back 10 years ago, AMD went from a single-digit billion-dollar market cap company to a double-digit billion-dollar market cap company just by taking single-digit percentage market share away from Intel.

You can very much build a massive business today just by taking some market share away from Nvidia. And the way you do that is by understanding where there could be gaps or becoming a viable second source to Nvidia for customers that don’t want to be locked in.

But to the points Dylan made, there are many ways to play this game. You can be a vendor into Nvidia or into any of these companies that are building accelerators and still be an interesting business. In fact, packaging and testing are areas people haven’t been paying attention to that are now becoming extremely important. In fact, if you look at the AI chip scarcity, it’s driven by packaging and integration more so than the semiconductor itself.

Danny Crichton:

Dylan, we overlap on a lot of the technical and business subjects, but you and I have a massive disagreement, which is the U.S. government’s purchase of 10% of Intel. I think it’s a catastrophe. I have never seen state capitalism in the United States prove successful. You are more optimistic. Tell us why.

Dylan Patel:

I don’t like state capitalism either. I think it’s a terrible idea. We win by being an economy that’s competitive and capitalistic. And there are certain ways to do state investment into industries that actually improves those, right? Look no further than Chinese auto. It is the most cutthroat competitive market in the world. Yes, there are some SOEs, yes, there were a lot of subsidies, but actually it’s extremely competitive.

And so when you look at fabs, obviously I would love for that to be the same case, but we’ve got some different realities. The first is the CHIPS Act, it was a... I don’t know. What’s your opinion on the CHIPS Act generally? Subsidize fabs a little bit and get them to be built here? Subsidize them enough to make it a little bit better than building fabs elsewhere?

Danny Crichton:

Chris Miller, Jordan Schneider and I wrote a paper pre-Biden administration, which in some ways became the CHIPS Act. Parts of that legislation were directly from that paper. For example, the regional innovation hubs idea was pulled from our paper. So I mean, I’m generally in favor. Generally the idea was moving the economics around, but it was not about ownership of companies.

Dylan Patel:

When you take a look at the CHIPS Act and how it was administered, there were certain things that had to be done. Intel had to ramp to a certain level of volume on their fabs to get the money. Samsung, same. Likewise TSMC. There’s no question TSMC is going to ramp to that volume in their fab as much as they kick, moan and scream about it being too expensive.

Now with Intel, it’s slightly different. Economically, it did not make sense for Intel to build an Ohio fab. But they lobbied for it, got the grant — or guaranteed loans is what I should have said. But Intel will not ramp to that volume because they didn’t end up getting a customer for their 18A technology.

Now, 18A is not a complete failure. They’re actually manufacturing some chips right now. New laptop chips will launch with it in the next six months. From everything the Dells, the HPs, and the Lenovos are saying, the chip actually looks okay. It looks competitive with AMD, which is on TSMC process technologies. It’s not bad, it’s not completely behind.

So Intel is at number two, but this industry is very, very capital intensive and very, very difficult. And the switching cost is absurdly high. If you’re in this race to build whatever chip in whatever market, do you just dedicate engineering resources to switching foundries to use a slightly worse node with higher risk on yield or do you invest in improving the architecture?

You only get X percent improvement from node, but then you get a larger improvement from architecture and then you get an even larger improvement from software. There’s these stacking curves, so you might as well focus on whatever’s further up the curve — which is as much investment in architecture as possible.

And so Intel has this really hard problem. They must leapfrog TSMC to get any customer business, otherwise it makes no sense. So then you’re stuck. Intel is the only American fab. It’s the only company in America that’s doing process R&D. They are behind — not absurdly so — but they are behind. And it is extremely strategically important that there are fabs in the United States. A Taiwan invasion is a true risk and a true possibility. There are many parts of the U.S. government that believe it will happen this decade, but many folks think it’ll happen in the next five years.

By 2028, if Intel does not get more money, I think they’re bankrupt. And so you have this really challenging problem where, effectively, you must subsidize them for geopolitical reasons. But how do you do that if the method we created via the CHIPS Act won’t work — they’re not going to meet the obligations? They’re not going to ramp up Ohio in time. In fact, they can’t even afford to ramp up in Ohio because their business has lost share faster than expected.

The fab side needs to be subsidized somehow, and the fastest mechanism was converting the failed obligation on the CHIPS Act to an ownership stake.

Danny Crichton:

Shahin, Michelle, I know you have comments, so what are your thoughts?

Shahin Farshchi:

Listen, this is all symbolic. This is all political. Ultimately, Intel doesn’t need to be owned by the U.S. government. There are many smart investors out there with tons of access to capital.

If you look at TSMC or these other great institutions, a lot of them have been supported by governments. A lot of these things just take time and a lot of capital. In the case of manufacturing chips in America, it makes absolutely no economic sense to do this. It’s all for geopolitical purposes, and it’s a cost-benefit question. We’re basically purchasing a very, very expensive insurance policy here.

And the question becomes how much we are willing to spend on this. So far, the amount that we’ve spent seems to justify the cause, but this is not going to be limitless.

Danny Crichton:

Michelle, you were on Capitol Hill for a period, as AI advisor in the Senate. What are your thoughts?

Michelle Fang:

I would echo what Dylan has said and throw a question back: Do you think that there is enough political will for a CHIPS and Science 2.0? What’s happening on the R&D side? A lot of the funding is actually supposed to be allocated for packaging and whatnot, but do you think that there is going to be an evolution of this? And if so, what is the role of the state in the next couple of years?

Dylan Patel:

Yeah, it’s an expensive insurance policy. But I guess again, how do you think through this? If there is a 10% chance that Taiwan is now completely influenced by China, and they can start to do to us what we did to them — cut off Nvidia like we cut off Huawei — what is the impact to the American economy?

Holy shit, it is way, way more than 10% times Nvidia’s revenue, which is $200-240 billion. 10% of that number is more than we gave Intel already. And then you add on Google and all these other companies, you add on the knock-on effect of all the AI spend, right? There’s also the data centers and the capex for all these other things. Laying fiber, all this. Oh, and the AI impact on the economy. The GDP is $400–$500 billion higher this year because of all the capital investment. Okay, 10% of that? Well then, in one year it’s $40 billion, $50 billion.

Our expensive insurance policy is actually quite cheap relative to the 10% number I just made up. Government people know way better, but I just know that people treat this very seriously.

Danny Crichton:

Well, we’ve gone around the world and across the stack, but Dylan, founder, CEO, and Chief analyst at SemiAnalysis, the must-read guide for everything going on in semiconductors and chips, thank you so much for joining us.

---

Don’t miss

Uncle Sam is a terrible board member

America’s degrowth lawyers need to learn from China

Stargate’s $500B bet for AI dominance