Jerry Murdock

1. Key Themes

The Neocloud Massacre Is Coming

Jerry believes at least half of all neoclouds will disappear within 36 months, and a financial dislocation would accelerate that timeline dramatically. The differentiator isn't scale — it's capital efficiency and willingness to generate actual profit. He uses Fireworks vs. Base Ten as the canonical example.

"Neoclouds right now, there's a whole bunch of them. I think at least half of them go away within 36 months. And if there's an economic disruption, a lot of them will go away right away." 00:00:00

"I think Fireworks is making a lot more money than Base Ten... I bet on Fireworks over Base Ten, 10 times better business, in my opinion, because they're more capital efficient." 00:11:56

Credit Market Complacency Is the Real Systemic Risk

Jerry's macro concern is not that AI is bad — it's that the unprecedented debt levels funding AI infrastructure, combined with complacent credit markets (narrow spreads, Japan's treasury holdings, post-Leopold leverage risks), could trigger a financial dislocation that cascades into the tech sector.

"There's many opportunities for there to be problems... The spreads are too narrow between real risk and not so much risk. And they're not really accounting for that." 00:07:59

"Japan holds a trillion dollars in treasuries. And if they have to unwind that, if they sold 300 billion worth of treasuries, in order to be able to buy dollars to support the yen, we would have a real problem on our hands immediately." 00:08:30

Open Source + ASIC Chips = Underappreciated Tsunami

Jerry has been consistent that open source models combined with ASIC chips (not GPUs) represent a coming wave of adoption, driven by the massive cost differential with frontier models and enterprise security concerns about data going to the cloud.

"If you're looking at a frontier model with double-digit cost per token, and you're looking at an open source model that's 10, 11 cents per token... it's still enough of a difference that there's going to be a massive adoption." 00:13:58

"I said it last February. Look, ASIC chips are really ideal if you're thinking about model customization... you don't need a GPU for that. Too expensive." 00:28:05

The Token Is NOT a Commodity — Customization Changes Its Value

Jerry directly pushes back on Jensen Huang's and Gavin Baker's "a token is a token" argument. As enterprises customize models with providers like Fireworks, the token's economic value shifts dramatically — both in what it produces and how verbosely it does so.

"I disagree with a token is a token... The more you customize the model, the more the token changes its value. Right? Because the more you customize what's being produced. And some models, they talk a lot more than other models. And so they produce a hell of a lot more tokens." 00:17:09

Continuous Learning Models Will Obsolete Every Model That Exists Today

This is Jerry's most structurally important thesis: within roughly 2–10 years, continuous learning models will require fundamentally new architectures and training, rendering all current frontier and open source models obsolete — including any Chinese backdoors embedded in them.

"Within 10 years, I believe we... continuous learning models will come into existence. That means that every generation of every model we have today dies, goes away." 00:48:16

"I think you're going to see... lifelong learning, which is truly more how human intelligence works. But those models, in my humble opinion, will replace every model that exists today." 00:13:35

Sandboxes Are the Most Underestimated Security Layer in AI

Everyone is running models in containers thinking they're safe. Jerry argues containers are fundamentally insecure for agentic workloads, and the sandbox layer — companies like E2B and Docker — is where the critical security value will concentrate.

"I don't know how many developers are running their models in YOLO mode, but probably a lot... Well, containers aren't safe. You need sandboxes. This is why the big container company, Docker, said themselves containers aren't safe. You better put in a sandbox." 00:22:50

"If you look at E2B and you look at Docker, they're probably the two best at understanding all that stuff. So you start there because if you don't get the sandbox right, forget everything else." 00:25:12

Blockchain Will Find Its True Utility in Agent Payments

Jerry's contrarian five-year prediction: blockchain's real-world killer app isn't Bitcoin speculation — it's serving as the payment and inference-exchange rails for AI agents. He names specific companies already building this infrastructure.

"Blockchain for agent payments. Blockchain is in the valley of disillusion right now... those innovations, whether you're tokenizing stocks or you're going to do payment rails, or you're going to use blockchain for inference, like on Akinaki or Gonka, those things are going to be real innovations." 01:01:51

Frontier Models Are the Most Extraordinary Innovation in Human History — But the Application Layer Doesn't Deserve Frontier Valuations

Jerry separates the deserved mega-valuations at the model and infrastructure layer from what he sees as unjustified pricing in the application and neocloud layers. Infrastructure that "rides the shoulders" of frontier models earns its economics. Apps do not — yet.

"The frontier model companies have done something extraordinary in the history of the world... It's definitely on the level of inventing fire... The companies that can follow right on top of them and not get killed by them, those guys, they deserve those economics. Other categories? No way." 00:34:42

The SaaS Existential Threat Is Just Beginning

The shift from co-pilots to co-workers (autonomous agents) is just starting to threaten SaaS companies. Companies without a genuine AI strategy embedded in their product — not just a bolted-on co-pilot — face existential risk within two years.

"The co-work era, it begins the threat. So the threat to the SaaS world is just starting right now... If you're not doing that now, good luck." 00:43:21

PE Is Highly Vulnerable to Any Financial Dislocation

Levered PE portfolios in SaaS-heavy vintages face a double threat: AI-driven EBITDA compression and debt structures that can trigger margin calls quickly. Jerry draws a direct parallel to how Forstmann Little didn't survive the telecom bust.

"The problem with the PE business is the leverage on the businesses. And if EBITDA drops, churn increases. And if it happens rapidly through a financial dislocation and there's a margin call effectively on the debt, yeah, it's going to be tough." 00:44:38

2. Contrarian Perspectives

A Token Is NOT a Token

While Jensen Huang, Gavin Baker, and most of the market treat compute tokens as a fungible commodity, Jerry argues customization fundamentally changes what a token is worth economically. A verbose customized model producing 10x the tokens for the same task changes the entire unit economics equation.

"I disagree with a token is a token. That may be true at the moment with pretty much frontier models. But I disagree with it because companies like Fireworks and others are helping companies to customize. The more you customize the model, the more the token changes its value." 00:17:09

Chinese Open Source Backdoors Are a Near-Term Non-Issue Because Those Models Won't Exist in 10 Years

The mainstream worry is that Qwen, DeepSeek, and others may have backdoors. Jerry's response is almost dismissive: these models will be completely replaced by continuous learning architectures within a decade, so the backdoor threat is self-limiting. This directly contradicts the urgency of most national security framing around Chinese AI.

"All these models are not going to exist in 10 years... if there's backdoors today, they better do what they're going to do now because they're not going to exist in 10 years." 00:47:56

OpenRouter's 5% Markup Is Already Dead Money — It Just Doesn't Know It Yet

The market treats OpenRouter as a durable routing and aggregation layer. Jerry thinks it's a transitional convenience product that blockchain-based inference exchanges (Akinaki, Venice.io) will eliminate in three to four months.

"OpenRouter charges 5% on top of that, which is a crazy amount of money. That's not going to last... I think you're going to see a big disruption in that model in the next three, four, five months." 00:35:38

"Shame on the enterprises for letting them burn all that money. That's a huge amount of money." 00:36:43

Owning the Chip Layer Is the Wrong Long-Term Move for Model Companies

As Anthropic builds Jalapeno chips and DeepSeek builds its own silicon, the instinct is that vertical integration = competitive moat. Jerry believes this is short-term optimization masquerading as strategy. The real opportunity lies in the complexity between model, agent, and human — not the chip layer.

"Actually owning the chip is going the wrong way long term. Short term, it makes sense for larger companies because they want to optimize chipsets for models... The opportunity is that level, which is from the loops to customization, multiple things, security." 00:28:59

Meta Is the Most Likely Mag 7 to Become "Boring" — Like an AT&T

Consensus treats Meta as an AI winner due to its ad business, Instagram, and WhatsApp scale. Jerry would short it among the Mag 7 — not because it fails, but because it devolves into a utility: stable, dividend-generating, but strategically stagnant.

"It would be meta because that's the one that may become boring. It may become like a telephone company. I mean, it'll just be this malaise, like owning an AT&T or something." 00:58:11

3. Companies Identified

Fireworks AI

Inference provider and model customization platform founded by ex-PyTorch team members from Meta. Jerry is an investor and cites it repeatedly as the gold standard for capital efficiency among inference providers.

"That team was the geniuses that understood PyTorch in particular... They're going to move up the stack. They're going to move up and do a lot more fine-tuning and refinement and customization for people. And they're going to be the best ones at it. That's why they're going to succeed." 00:23:48

"It's three and a half years to a billion. It'll be four years to two billion if they hit end of year targets this year." 00:33:07 (Harry attributing Fireworks' growth trajectory)

E2B

Cloud sandbox provider for AI agents. Named as one of only two companies that truly understand how models interact with tools — the foundational layer for agentic security.

"If you look at E2B and you look at Docker, they're probably the two best at understanding all that stuff. So you start there because if you don't get the sandbox right, forget everything else." 00:25:12

Docker

Container platform that has pivoted to sandboxes for AI agent security. Cited as one of two foundational sandbox companies and noted for publicly acknowledging that containers themselves are not safe.

"Docker said, themselves said containers aren't safe. You better put in a sandbox. And that's why they've had a huge success with Docker sandboxes." 00:22:50

Akinaki (Gonka)

Blockchain infrastructure company that has launched a decentralized inference exchange (Dodex) on mainnet. Jerry views this as the beginning of the end for OpenRouter-style routing with markup fees.

"There's a blockchain company called Akinaki that has just launched Dodex on their mainnet. And this thing is an exchange to go out and buy inference. And as part of that, all the model routing is done for you." 00:35:38

Venice.io

Inference exchange platform building direct access to inference without the OpenRouter 5% markup. Named alongside Akinaki as part of an imminent disruption wave.

"Venice.io is doing that. And there's two or three other guys that are now in the process of building exchanges that you can go directly, get the inference you need without paying the 5% markup." 00:37:11

Vignan (Vignana)

AI company founded by Saad Khan, former Meta employee. Jerry and Vinod Khosla both identify the CEO as exceptional.

"The founder of Vignan, he's absolutely going to do it. There's no question in my mind. Avin, this guy came out of Meta as well. His name is Saad Khan. But Vinod Khosla said this is one of the best CEOs he's ever seen." 00:38:41

Insight Partners

Jerry's own firm. Singled out as best-positioned among PE/growth firms because of a deliberately small PE portfolio — minimizing exposure to the leverage risks plaguing other firms.

"My company Insight, they have a very small PE portfolio. Very, very small. It almost doesn't even matter. So I think I feel the best about them long term because they've been really intelligent about how they deployed the capital there." 01:00:16

Menlo Ventures

Early Anthropic investor. Cited as one of the top three venture firms navigating the AI transition.

"The guys like Menlo that did Anthropic, like how do you say early, but early enough. You know, they're going to do great." 01:00:56

Benchmark

Named alongside Menlo and Khosla as a top firm, despite missing OpenAI and Anthropic — credited for Factory and other portfolio wins.

"Benchmark, well, they missed Anthropic and OpenAI. They've done amazing with Factory and a bunch of other great ones that we've talked about." 01:00:56

Khosla Ventures

Named as one of three firms that will "crush it" in AI, credited largely to Vinod Khosla's early identification of exceptional founders like Saad Khan.

"I think Khosla is also phenomenal. I mean, Vinod with Avin and, you know, his companies, they did amazing. I think Khosla is in that group. I mean, they're just going to crush it." 01:01:24

Robinhood

Cited for launching Robinhood Chain and achieving approximately $1 billion in revenue — a validation point for blockchain finding real utility beyond speculation.

"What Robinhood did with Robinhood Chain, great. They got a billion in revenue probably." 01:02:21

OpenRouter

Model routing and aggregation platform charging 5% markup on inference. Identified as a transitional product whose business model will be disrupted within months by decentralized inference exchanges.

"OpenRouter has massive amounts of transactions because people are basically lazy... OpenRouter charges 5% on top of that, which is a crazy amount of money. That's not going to last." 00:35:38

McCaw (Likely Mecor or similar data company Harry is invested in)

Data provisioning company Harry is an investor in. Discussed in context of the thesis that specialized enterprise data companies will expand TAM from frontier model training to enterprise and mid-market fine-tuning.

"You'll see the likes of McCaw go from purely selling to frontier models to selling to enterprises and even mid-market who need specialized data that they might not have. And that massively opens the TAM. That's the $200 billion opportunity." 00:51:25 (Harry speaking)

Lagora

Legal AI company Harry is an investor in, competing directly with Harvey. Used as a case study for the risks of two well-funded competitors watching each other instead of innovating — creating vulnerability to a niche disruptor.

SSI (Safe Superintelligence — Ilya Sutskever's company)

Mentioned as supposedly releasing the first continuous learning model end of August. Jerry declined to evaluate it without a personal relationship with the team.

"Would you have done SSI at 30 billion, Ilya's company, which is supposedly coming out with the first version of their continuous learning model end of August?" 00:50:10 (Harry asking)

Bending Spoons

PE/acquirer cited as a buyer of distressed SaaS assets like Airtable — with the observation that there aren't many buyers like them in the market.

Factory

Coding agent / AI dev tools company. Cited as a strong Benchmark portfolio win alongside other AI companies, compensating for Benchmark missing frontier models.

4. People Identified

Co-founder and Managing Director of Insight Partners (managing $90B+). Veteran of 25+ years of technology cycles. Investor in Fireworks AI, Lagora, and many others. On the board of the Santa Fe Institute.

"Jerry has seen pretty much every technology cycle of the last 25 years. He's invested in some of the biggest companies across those 25 years." 00:00:21 (Harry introducing him)

Lynn (Founder of Fireworks AI)

Founder and CEO of Fireworks AI. Thesis: specialized intelligence and custom models will dominate over general frontier models for enterprise use cases. Jerry is an investor and strong advocate.

"Lynn has margins in the 35% range, she said on the show publicly." 00:25:23 (Harry citing Lynn's public statements)

Saad Khan

Founder/CEO of Vignan (AI company). Former Meta employee. Both Jerry Murdock and Vinod Khosla have independently identified him as exceptional.

"His name is Saad Khan. But Vinod Khosla said this is one of the best CEOs he's ever seen. And Vinod was one of the best CEOs building Sun. So when someone says that, you take it seriously." 00:39:10

Vinod Khosla

Founder of Khosla Ventures, former co-founder of Sun Microsystems. Cited for exceptional founder identification ability.

"Vinod with Avin and, you know, his companies, they did amazing. I think Khosla is in that group. I mean, they're just going to crush it." 01:01:24

Bill Gurley

General Partner at Benchmark. On the board of the Santa Fe Institute with Jerry. Referenced for consistent, disciplined venture capital thinking.

"Bill's on the board of the Santa Fe Institute with me. I mean, his guidance is pretty smart. He's pretty much on point with a lot of things with venture capital." 00:37:57

Gavin Baker

Investor at Atreides Management. Referenced for his "a token is a token" thesis, which Jerry directly and publicly disagrees with.

"A token is a token is actually what Gavin Baker said the other day. And Jensen doesn't give a shit whether you put it on a frontier or an open source." 00:16:51 (Harry quoting Baker's position)

Alex Karp

CEO of Palantir. Referenced for his public argument that large enterprises fear frontier model providers eating their lunch — a position Jerry partially agrees with while noting Karp's customer base is self-selectingly paranoid.

"Alex Karp has said the biggest enterprise customers in the world don't want to work with frontier model providers. They're scared that they're going to eat their lunch." 00:20:26 (Harry quoting Karp)

Tom Lee

Market strategist. Referenced for calling a 10%+ S&P drawdown in the fall, and separately for warning Bitcoin could be hacked within two years.

"It was Tom Lee that called for a 10% decline or drawdown in the S&P this fall." 00:45:33

Satya Nadella

CEO of Microsoft. Referenced for publicly acknowledging Microsoft has deep data about how communication works inside enterprises.

"Microsoft, Satya Nadella himself came out and said, look, we've got a lot of the data around the communication, how communication works inside an organization." 00:21:14

Orlando (Toma Bravo)

Cited as a well-regarded PE leader facing significant portfolio headwinds from AI disruption of SaaS holdings.

"I like PE today but I'm looking at your Toma Bravos of the world and I'm just like, ouch. I really like Orlando and he was great on the show." 00:43:44 (Harry speaking)

5. Operating Insights

Build a Culture That Targets Margin Even Before Achieving It

Low-margin land-grabs are a legitimate strategy, but only if the team has an explicit plan and cultural commitment to monetization. Jerry will not invest in companies that normalize low margins as a permanent state — the innovation roadmap must show a path to margin.

"I believe that you need to build a culture that works... if you're not thinking about it, I won't invest in you." 00:27:02

For Sandbox/Security Products: Price the Intelligence, Not the Compute

The current dominant business model for sandbox companies is charging on compute. Jerry explicitly calls this "a dumb idea." The smarter model is "bring your own compute" — charge for the proprietary knowledge of how to run, network, and trace sandboxes efficiently. This principle generalizes: charge for the knowledge layer, not the commodity layer.

"Most sandbox people make money on compute. Frankly, that's a dumb idea, in my opinion. You want to have a model that says bring your own compute. And we'll make money because we understand how to run sandboxes better. We know how to network them better. We know how to provide traces better." 00:27:02

Niche Domination Before Ocean Expansion — Peter Thiel's Advice Applied to AI

When evaluating competitive two-player legal AI markets (Harvey vs. Lagora), Jerry's instinct is to skip both incumbents and find the third entrant who dominates a hyper-specific sub-niche (e.g., trust and estates). The competitive intensity between the two frontrunners creates blind spots they will exploit together.

"I want to go for the next innovative young company who's maybe not trying to do all things for all lawyers and be more highly specialized... start with a niche, dominate the niche and then grow it out." 00:30:48

Know When to Hit the Bid — Don't Let "The Coach" Talk You Out of a Good Exit

The Flipboard story is an operating lesson: when two strategic acquirers are bidding near a strong number, advisors who say "don't sell" are often wrong. Boards should evaluate the company's decade-long durability, not just current momentum.

"Flipboard had taken the billion-dollar exit, but they didn't... The board bought into the coach's advice, and they stayed with it. And now Flipboard, you don't care about it, right? They missed the opportunity." 00:40:00

Invest in Founders Who Have No Other Option

The single clearest filter Jerry articulates for backing a founder: they aren't building a company because they want to — they're building it because they have to. This compulsion signals the level of commitment that survives financial dislocations and competitive onslaughts.

"It's not that they want to build this business. It's that they have to build this business. And if you find that in a person and you recognize that they have the commitment to it, because the commitment to it is all in. And there's no other option." 00:37:57

6. Overlooked Insights

The Santa Fe Institute Is a Stealth Early Warning System for AI's Most Consequential Breakthroughs

Jerry mentions almost in passing that he is on the board of the Santa Fe Institute, which convened chief scientists from all major AI companies. That meeting produced two findings that haven't entered mainstream investment discourse: (1) no one can measure AGI even if it arrives, and (2) the most important design space is the complexity interface between model, agent, and human. This is not published research — it's closed-room consensus among the people actually building these systems. Jerry's investment thesis flows directly from this meeting.

"When I was at the Santa Fe Institute, I'm on the board there. We had a meeting with a bunch of chief scientists from all the major AI companies. And what we came back with out of that meeting, two points, was one, we don't know how to measure AGI even if it shows up. But the second point... we should focus on this complexity between the model and the agent and therefore the human being to the degree that they're in the loop. That's where the opportunity is." 00:28:59

The investment implication is direct and specific: Jerry says the most compelling place to invest is everything from "the loops to customization, multiple things, security" — the model-to-human interface layer. This is a non-obvious, highly informed conviction that most investors are not building around.

Cursor's Acquisition Was Not a Scaling Win — It Was a Strategic Pivot Out of a Dead End

The market narrative around Cursor is that it was acquired at $60 billion because it was the fastest-growing dev tool ever built. Jerry's read is fundamentally different and almost no one is saying it: Cursor pivoted out of the IDE space, convinced Elon Musk they could build models (without having proved it), and hit a bid because Elon was "pretty desperate" to solve XAI's problems. The acquisition was an exit from a position, not a graduation to a bigger one.

"I think what Cursor did, because the team is really smart, is they pivoted out of the IDE space. And in that pivot, they convinced Elon that they could build models. They hadn't proved it yet, but they convinced them that they knew enough to do it. And Elon was pretty desperate to solve his problem with XAI. And so it was a great fit. And they got the 60 billion." 00:39:30

The implication: the $60B price was driven by Elon's desperation and Cursor's persuasion, not purely by Cursor's demonstrated model-building capability. This reframes what investors should learn from the deal — and raises real questions about whether XAI got what it paid for.