Summary

AI application economics break SaaS’s familiar 80%-margin scorecard. Randle argues investors should underwrite terminal 5–7-year margins, gross-profit multiples, and absolute gross profit per customer: an AI product at 50% margin can be superior to SaaS at 75% if it produces $500,000 versus $200,000 of gross profit. His blunt call: “We should not be placing that much emphasis on margins today,” especially because high inference COGS can reflect genuine AI usage. AWS is his analogy—lower margins can coexist with much larger customer spend.
Coding is already a “golden category,” even if Cursor’s market share keeps falling. Randle estimates code generation grew from essentially zero to $6–7 billion of ARR in roughly 2.5 years and could add another $4–5 billion this year; Cursor may have fallen from roughly 80% share to 25–30%, yet still be addressing a vastly larger market. Products with the most usage—Cursor, Claude Code and Codex among them—also improve fastest through deployment and can “leave everybody in the dust.” But growth is real only when an app clears the labs’ baseline: Jasper grew rapidly and then shrank when GPT-4 made its output look too similar to ChatGPT’s $20 offering, recovering only through more differentiated workflow software.
At the latest stated prices, Randle would take OpenAI at $500 billion over Anthropic at $350 billion. Anthropic remains slightly ahead in coding and probably B2B commercialization, while OpenAI has recovered ground with Codex; the decisive asset is ChatGPT, whose growth trajectory Randle finds almost impossible to stop. Having passed on OpenAI at $32 billion over nonprofit structure and dilution concerns, he now predicts it could be a trillion-dollar company next year: “I missed the forest for the trees.”
Benchmark’s small fund is designed to maximize multiples, not win every mega-round. Randle says the five best investments in its last fund, marked at last-round prices, stand at roughly one 60x, two 30xs and two 20xs—returns no post-ChatGPT OpenAI round matches. Benchmark therefore need not buy every lab financing; its two north stars are being the founder’s closest, highest-ROI partner and generating the highest money-on-money return in an LP’s venture portfolio. A customary 20% ownership target is an input rather than the goal: lower ownership in Mercor can still produce exceptional returns if Benchmark remains its most consequential venture partner.
Mega-funds may make immense absolute profits while still failing venture’s return test. Randle’s argument is structural: “You ship your fund size,” so $7–10 billion vehicles must write enormous checks, and those checks inevitably become the main product and organizational priority. He doubts their managers can credibly promise 5x net across the relevant basket of funds; Harry’s pushback is that unprecedented outcomes may still rescue the model, which Randle accepts in dollars but not necessarily in multiples. Tiger may finish far better than its reputation suggests, given positions in Databricks and OpenAI and preferred-stock protections on some losers, but Randle still expects many AI companies to go to zero or fall 90% while rare winners compound for decades.
AI’s moat remains technology, not merely distribution. Distribution earns a company the opportunity to build, but Randle says exceptional AI products require scarce talent, nuanced model pipelines and workflow design—not “bringing in the OpenAI API” beside a text box. The labs establish a $20 or $200-per-month experience baseline, so application companies charging more must create deeply differentiated workflow value that survives the next model release.
Commodity AI infrastructure can overwhelm quality concerns. Randle changed his mind on AI clouds, initially dismissing CoreWeave as a commodity middleman before astronomical inference demand overwhelmed that objection. He cited CoreWeave at roughly $60 billion and Nebius at roughly $30 billion in public-market value, with more than $100 billion across the public sector. He still expects CoreWeave and similar companies could eventually fall 70%, but says demand can justify investing with momentum.
Price matters only against the company’s own upside. Randle ranks people, product, market: people are the upstream engine, product is the strongest evidence of their quality, and market is most fungible because companies can pivot. SpaceX at $150 billion, Rippling at a $250 million Series A and Figma at $400 million on $4 million of ARR taught him to remove intimidating zeros and underwrite TAM, competitive position and upside rather than market convention. Models provide a base-rate yardstick—perhaps the path others underwrite for a 3–5x—but detailed forecasts become false precision; the real test is whether the qualitative view says the company will “absolutely smoke these projections.”
Benchmark’s greatest risk is stasis, not one missed cycle. Randle calls stasis the biggest threat over the next two decades: Benchmark must evolve with the asset class while preserving its two north stars and continuing to reach the very best founders. His long-run optimism rests on AI lifting GDP per capita as population growth slows—“continuing growing the pie” as the foundation for a functional, less zero-sum society.

Deep dive

1. Great investors turn process into conviction

Mary Meeker’s quantitative reputation obscures what Randle considers her real gift: she is “the most qualitative investor” he has worked with. Reading a company’s historical and projected numbers “like she’s reading the Matrix,” she sees an 8–10-year narrative—for DoorDash, not an abstract growth rate but perhaps 20% of households ordering monthly.
The lesson was to use numbers to drive an investment story rather than become trapped in the model. Meeker’s sequential data gives her a way to visualize what a company becomes, joining quantitative discipline to a qualitative judgment about adoption, behavior and market position.
Peter Thiel’s genius, in Randle’s telling, appears as much in institutional design as stock-picking. Founders Fund employees could invest personally beside the fund, creating a concealed conviction test: if an investor would rather keep their money in the S&P, “why would we give our LPs this allocation?” Younger staff sometimes used unsecured debt lines to participate.
Founders Fund’s famously intense investment committees worked because relationships were secure enough to permit “no holds barred, complete truth-seeking.” Randle could tee off on Keith Rabois without hierarchy intervening; disagreement felt like fighting with a sibling, not navigating a political bureaucracy.

2. Mamoon Hamid taught taste by putting excellence within reach

Mamoon Hamid’s core lesson was that young investors must see excellence up close. Without early exposure to exceptional founders, management teams and boardrooms, they cannot reliably recognize that standard “in the wild” or hold weaker portfolio teams to it.
Randle sees a common line through Figma, Glean and Rippling: B2B software with consumer-like products, unusually strong user love and engagement, and teams capable of meeting a correspondingly high product bar.
Hamid’s advantage is “impeccable taste” across people, product and market, sharpened inside a deliberately narrow zone of strength. His mentorship encouraged Randle to develop an equally specific taste rather than imitate a generic venture playbook.
Harry supplied his own exaggerated endorsement: when Hamid offered to bring him into one of his deals, he told his team the diligence was already finished—“It’s B2B. It’s kind of PLG. It’s Mamoon”—capturing how much informational value an investor’s accumulated taste can carry.

3. OpenAI at $32 billion became the miss that changed Randle’s instincts

Randle loved ChatGPT immediately, but passed on OpenAI’s $32 billion round because the nonprofit structure, employee units and likely dilution looked “really gnarly.” Those risks were valid—the structure nearly destabilized the company, and talent hiring caused heavy dilution—but they did not matter beside its unprecedented growth and utility.
His diagnosis is unforgiving: “I got spooked and I missed the forest for the trees.” The private-equity training that gave him analytical discipline also encouraged him to over-weight structural complexity when the product itself was supplying extraordinary evidence.
Josh Kushner’s instinctive reactions to Spotify and Instagram now provide Randle’s counter-model: when a product feels inevitable, trust that intuition enough to avoid letting secondary defects dominate the decision. OpenAI remains his biggest miss and “hurts to this day.”
Randle predicts OpenAI could become a trillion-dollar company next year and raise at that level “no problem.” Harry’s related rule, learned from Kushner: if an investor is willing to accept less allocation merely to accommodate someone else, the weakened appetite itself says, “Don’t do that deal.”

4. ChatGPT gives OpenAI the edge while coding remains contested

Asked to choose between OpenAI at $500 billion and Anthropic at $350 billion, Randle chose OpenAI, though he called both potentially good investments. His downside analysis begins with ChatGPT: he cannot see what knocks it off its growth trajectory or prevents it becoming the most important consumer destination and app of the next five years.
Anthropic probably retains a bit of an edge in B2B commercialization after committing more time and resources to enterprise selling. It also remains slightly ahead in coding through Claude Code, Sonnet and its broader model suite.
OpenAI has nevertheless “made up a bunch of progress” with Codex, turning coding into hand-to-hand combat rather than a settled Anthropic advantage. ChatGPT’s consumer position, rather than a claim that OpenAI wins every workload, determines Randle’s valuation preference.
Cursor’s fate produced Randle’s honest non-answer: “I don’t know.” What he rejects is the inference that declining share means declining opportunity; Cursor can lose relative ground to Claude Code, Codex and Cognition while growing into a far larger absolute market.

5. Code generation is expanding faster than share is fragmenting

Cursor may have moved from roughly 80% of its initial market to 25–30% as competitors arrived. Randle’s estimate, however, is that code generation expanded from essentially zero to $6–7 billion of ARR in approximately 2.5 years.
His old “golden category” test identified markets adding $1 billion of net-new ARR in one year—large enough that a multistage fund effectively needed a position. Coding could add $4–5 billion across products and services this year alone.
Harry asked whether AI makes every category golden and whether the threshold should rise from $1 billion to $10 billion. Randle conceded AI enlarges many markets, particularly where software absorbs labor, but not all: a niche such as AI for veterinarians might still lack enough customers and budget.
Usage is also strategic. Randle expects products such as Claude Code, Codex and Cursor to improve fastest because AI products get better through usage; richly funded companies that have not put products into developers’ hands may face “a rude awakening.”

6. Labor budgets require a new taxonomy for AI companies

Randle’s concrete example was a Kleiner Perkins-backed home-services voice-AI business offering a 24/7 receptionist. A customer spending roughly $250,000 on seven ServiceTitan products spent the same on this single, newly launched AI product.
The economics worked because the customer could reduce three receptionists to two while answering calls and booking appointments around the clock rather than only from nine to four. The AI product both reduced labor expense and captured otherwise lost revenue.
That relationship cannot be understood by forcing the company into a SaaS template. Robert F. Smith, CEO of the first firm Randle worked at, Vista Equity Partners, used to say SaaS “tastes like chicken”: the businesses were similar enough for a repeatable operating playbook.
Investors are accustomed to roughly 80% gross margins, high-80s gross retention, 120%+ net retention and little capex. AI applications put inference directly into COGS. Lower margins may therefore indicate genuine usage; unusually high AI-app margins can mean customers barely touch the AI functionality. Randle wants investors reasoning from terminal economics, not rewarding an attractive but potentially empty percentage.

7. Absolute gross profit can matter more than margin percentage

Randle’s preferred comparison: if ServiceTitan produces $200,000 of gross profit per customer at 75% margin while an AI company produces $500,000 at 50%, “I don’t care” that the second percentage is lower. It has captured a broader relationship and more economic value.
The analytical units should become gross-profit multiples and absolute gross-profit dollars per customer, alongside a first-principles view of margins five to seven years out. Training costs and companies developing their own models add further differences from conventional SaaS.
AWS is his analogy. He estimated its gross margin at perhaps 50–60% and operating margin around 30%, yet it is commonly a software company’s largest line item—far above Salesforce, Workday or Adobe—because infrastructure spend is so expansive.
In the early 2010s, a $150 million-revenue software company could show a startling $30 million AWS COGS line. AWS’s percentage margin was less important than the multiples more customers spent on it; Randle suggested it might be a trillion-dollar standalone business if separated from Amazon.

8. Growth is real only when an app clears the labs’ baseline

AI companies can go from 0 to 100 in less than a year, but Randle keeps “easy come, easy go” in view. Jasper grew rapidly and then shrank because its early revenue lacked enough scaffolding and durable customer value.
GPT-4 exposed the weakness: customers judged Jasper’s output similar to what ChatGPT offered for $20 per month. Randle believes Jasper subsequently recovered by embedding LLMs throughout differentiated marketing workflows rather than selling lightly packaged model access.
The labs now set the minimum customer experience. An application charging materially more than ChatGPT’s $20 or $200 per month must outperform the lab product enough to support enterprise distribution, retention and a sustainable business equation.
Randle rejects the claim that moats moved wholesale from technology to distribution. Distribution provides “the right to build differentiated technology,” but exceptional AI products require scarce talent, careful model pipelines and tasteful workflow integration—not an API placed inside a text box.

9. AI talent and product craft remain the scarce technology

The classic “seven powers” have not disappeared; the stakes have risen because applications scale faster while the labs improve and distribute their own products faster. Sustainable growth still depends on differentiation that persists through successive model upgrades.
Randle reframes the technology moat as partly a talent moat. Few people can decide where LLMs belong in a workflow, how outputs should be improved, and how the full product should behave sufficiently well to outshine the labs’ applications.
That scarcity explains researchers receiving billion-dollar contracts and “LeBron money.” Distribution without such builders may produce access, but not the exceptional product required to defend the customer relationship.
Harry’s distribution-and-data thesis therefore received a direct rebuttal: Randle said the moat remains fundamentally technological. Product quality is also evidence about the people capable of building it.

10. Commodity AI infrastructure can overwhelm quality concerns

Randle changed his mind most sharply on AI clouds. He initially dismissed CoreWeave as a middleman reselling commodity compute—a broker with structurally weak margins—but astronomical inference demand overwhelmed that business-quality objection.
He recalled CoreWeave raising privately around $3 billion; at the episode’s stated point it was roughly a $60 billion public company, while Nebius was around $30 billion and the sector exceeded $100 billion of public market capitalization, before counting rapidly growing private players.
Randle still thinks CoreWeave and similar companies “probably go down like 70%” at some point. Yet early investors already gained liquidity after a 20x move from the level where he rejected the company, making his original commodity critique economically beside the point.
The lesson is deliberately uncomfortable: when demand resembles the first hyperscaler wave—and AI inference may show an even steeper cohort curve—“sometimes you just got to shut your mind up and invest with the momentum.”

11. Benchmark invests its fund size instead of chasing every laboratory

Randle applies Conway’s law to venture: firms “ship” their fund size, team and structure. A $7 billion fund with 50 investors must participate in mega-rounds because billion-dollar checks are among the few ways to deploy its capital productively.
Benchmark’s smaller fund can play another game. Its five leading investments from the last fund, marked at last-round prices, were approximately one 60x, two 30xs and two 20xs; Randle said no OpenAI round since ChatGPT’s launch reaches those multiples.
Harry calculated OpenAI at $32 billion as perhaps 12–15x headline appreciation, but closer to 6–8x after dilution. He contrasted that with Benchmark’s Lovable, LangChain, Sierra, Mercor and Fireworks positions: for a small fund, cash-on-cash return—not prestige—is the product.
Randle acknowledged that skipping labs could threaten relevance and access. His present counterevidence is Benchmark’s relationships with people such as Bret Taylor, whom he called the godfather of AI applications, and Brendan at Mercor, whom he cited on the AI-infrastructure side.

12. Ownership is an input; partnership and return are the outputs

Benchmark’s two north stars are to become each founder’s closest, highest-ROI partner and to generate the highest money-on-money return among an LP’s venture holdings. A customary 20% ownership target is not itself one of those goals.
Harry cited an article putting Benchmark’s Mercor ownership at roughly 10%, below its historical norm. Randle’s response was that larger potential outcomes create more ways to deliver exceptional returns while remaining the company’s most consequential venture partner: critics “confuse the inputs for the outputs.”
Harry challenged the “best partner” claim with Delian Asparouhov’s criticism that Benchmark fires founders. Randle noted that replacing founders was once routine—Google’s investors immediately sought a professional CEO—but said governance and board-founder relationships have changed substantially for the better.
Founder loyalty cannot erase legal, ethical or fiduciary duties. Harry described boards sacrificing the cap table to protect founder NPS; Randle agreed, adding that great founders do not want sycophants or “GPT-4o in the boardroom,” but adults willing to spar and improve the company.

13. Price matters only against the company’s own upside

Benchmark’s four general partners each represent 25% of the partnership and retain distinct styles: Eric Vishria gravitates toward inception, while Randle expects initially to invest more around Series A, B and beyond. The shared constraint is not stage but exceptional founder partnership and return potential.
Randle admitted insecurity about arriving as a growth investor. Vishria answered that Bill Gurley had been a public-markets analyst before Benchmark; Randle also cites Pat Grady as evidence that the best investors increasingly transcend stage rather than live inside an organizational category.
His ranking is people, product, market. People are the upstream engine; product is the strongest evidence of their quality; market comes third because it is most fungible. Teams can pivot, as Slack did, while a non-exceptional founder or product organization is much harder to transform.
SpaceX at $150 billion taught Randle to remove the intimidating zeros and compare TAM, competitive position and upside. Rippling’s $250 million Series A and Figma’s $400 million valuation on $4 million ARR looked absurd relative to market convention, yet relative pricing would have screened out their excellence.

14. Models test conviction, but mega-fund structures reshape behavior

Randle uses models to lay out the base-rate future—perhaps the path other growth investors underwrite for a 3–5x—and then asks whether his qualitative view says the company will “absolutely smoke these projections.” Beyond that yardstick, detailed forecasts become false precision.
Figma illustrates the trap: any accurate-looking forecast of its eventual duration, growth and profitability would have seemed engineered to win investment committee approval. Simple market sizing also failed because counting designers missed adoption by product and other roles.
In his 2021 essay “Playing Different Games,” Randle predicted bifurcation between Tiger’s high-velocity, founder-friendly, low-touch model and Benchmark’s concentrated, high-touch craft. The undifferentiated middle became his “J.C. Penney funds”; after Tiger faltered, “we got six or seven more Tigers.”
Harry disputed applying Tiger’s north star to Thrive, Lightspeed or General Catalyst. Randle softened the Thrive characterization but proposed an organizational test: ask the principals, junior partners and associates whether deployment affects promotion. When billion-dollar checks generate perhaps 95% of profits, he argues, they inevitably become the main product.

15. Mega-funds can win in dollars while losing the venture-return test

Randle accepts Harry’s central pushback: OpenAI, Anthropic and Cursor may become far larger than anyone once imagined, allowing their investors to make immense absolute sums. The disagreement is whether that translates into the multiples LPs expect from venture.
He doubts leaders of the largest firms can tell LPs “with a straight face” that a pari passu basket of their funds will deliver 5x net. Returning even 4x net on an $8–10 billion vehicle, he argued, approaches a scale that “defies the laws of physics.”
LP demand for private technology may postpone any reckoning because institutions currently accept lower returns for access. But Randle notes that LPs already have private equity for lower-return exposure, often with better liquidity; venture’s distinctive role is unusually high money-on-money performance.
Inside a 50-investor platform, the 23rd partner may inherit 30 merely good company relationships, need a couple of deals for promotion, and hope one creates tenure. Randle says that resembles investment banking or a large buyout shop more than “meeting really interesting founders” and making only the best investments.

16. Tiger may be vindicated, but the next crash still demands survival

Randle agreed that Tiger’s 2021 portfolio may finish far better than its reputation suggests. Large positions in Databricks and OpenAI, plus preferred-stock liquidation protection on many of the things that do not work, could make “#JusticeForJohnCurtius” more than a joke.
His conditional scenario was explicit: if Databricks becomes a $400–500 billion company and OpenAI becomes multitrillion-dollar, Tiger’s fund may be “pretty okay”—not the best portfolio an LP ever owned, but nowhere near a money incinerator.
The excesses remain vivid. At a December 2021 or January 2022 Miami party, possibly featuring Vanilla Ice, Randle watched public technology fall 30–40% and thought of The Dark Knight Rises: “Gotham is burning,” while the industry attended one last decadent celebration.
Today may rhyme with the dot-com cycle: many AI companies will go to zero or fall 90%, while the rare enduring winner compounds for 20–30 years. Benchmark’s answer is constrained fund size and careful capital use, preserving the ability to weather a crash rather than forcing LPs to flee.

17. Benchmark’s greatest risk is stasis, not one missed cycle

A partner advised Randle that a failed first Benchmark investment might be liberating: once it fails and “LPs still love us” and nobody is fired, the second decision carries less psychological weight. Early apparent success might instead compound the pressure.
Asked where he would put money among his former firms for the highest cash-on-cash return, Randle chose Founders Fund because incubation produces equity when buying it becomes too competitive. Anduril demonstrates the possible fund-level result; every few funds, or every five to ten years, the firm has incubated an unbelievable company.
Benchmark itself must remain dynamic while preserving its two north stars. Randle calls “stasis” the largest threat over the next two decades: involvement with the very best founders is the asset class’s currency, and tradition cannot be allowed to make “the tail wag the dog.”
His ten-year optimism is macroeconomic. Social media showed capitalism optimizing human attention toward glued screens, but AI can instead raise GDP per capita as birth rates slow. Drawing on Peter Thiel’s framing, Randle sees economic growth—“continuing growing the pie”—as the essential condition for prosperity and a harmonious, non-zero-sum society.

⚡ Inside Google Labs: Building The Gemini Coding Agent — Jed Borovik, Jules + AIE CODE Preview 140: Deep Potential’s Zhang Linfeng and Sun Weijie: AI for Science, from the Beginning to Now