How should AI companies trade off gross margin against growth?

Aidan Gomez · Aug 19, 2024

Canva's choice to keep prices flat and absorb AI costs is a good bet

They want to grow and expand their user base by giving the most useful product possible, and the cost of AI is falling so quickly that margins are not worth worrying about right now

Scope: for now, given falling AI costs

34:39 20VC: Chips, Models or Applications; Where is the Value in AI | Is Compute the Answer to All Model Performance Questions | Why Open AI Shelved AGI & Is There Any Value in Models with OpenAI Price Dumping with Aidan, Gomez, Co-Founder @ Cohere

Daniel Khachab · Oct 28, 2024

At Choco's stage of maturity, model choice should be 100% about utility and functionality rather than price

Build value first and worry about cost later; their costs fell anyway even as volume rose, without any investment in cost optimization

Scope: specific to their stage of maturity

17:12 20VC: Why SaaS is Dead | Why AI First Companies Will Win | We are in the Middle of a Cold War for AI Talent | Why Europe is F******* and We Need to Stop Whining with Daniel Khachab, Co-Founder @ Choco

Anton Osika · Aug 18, 2025 · hedged

At Lovable's current stage the right priority is maximizing mindshare and the number of users who love the brand, deferring margin optimization until later

Mindshare and brand love are what matter most right now; the payback/margin discipline can come afterwards

Scope: he holds both perspectives and only "indexes a bit more" on this one; acknowledges margin changes do affect growth speed

15:12 20VC: Lovable CEO Anton Osika on $120M in ARR in 7 Months | The Honest Truth About Defensibility and Unit Economics for AI Startups | The State of Foundation Models: Long Grok, Short OpenAI, Why | Replit vs Lovable vs Bolt: What Happens

Maor Shlomo · Nov 24, 2025

Cursor's strategy of growing fast on expensive frontier models and then migrating users onto its own trained model (Composer) is a strong playbook, because margins can be fixed overnight once distribution is won

Distribution is the hard part; model cost structure can be changed drastically and quickly once you have your own efficient model

Scope: framed as taking the pessimistic view that Cursor's margins were bad

38:57 20VC: Base44's Maor Shlomo on How Vibe Coding Will Kill SaaS and Salesforce | Why it is BS that Vibe Coding Platforms Do Not Have Defensibility and Bad Margins | Why He Worries About Google, Not Replit and Lovable | Why Long Anthropic, Not OpenAI?

Amjad Masad · Apr 25, 2026

Cost is secondary to performance; optimizing for cost only makes sense once you hit the asymptotic plateau of the S curve in your domain

Capital is flush, and in every tech era — cloud, mobile, SaaS — the goal is growth and being at the edge until things rationalize, at which point you focus on gross margins

Scope: especially in a capital-flush environment

12:10 20Product: Replit CEO on Why Coding Models Are Plateauing | Why the SaaS Apocalypse is Justified: Will Incumbents Be Replaced? | Why IDEs Are Dead and Do PMs Survive the Next 3-5 Years with Amjad Masad

Amjad Masad · Apr 25, 2026

AI companies should build the best product first and only enter an optimization period afterward; optimizing margins early is premature optimization

As in programming, thinking about performance and cost too early leads to wrong decisions; margins move around because product quality is the priority

13:25 20Product: Replit CEO on Why Coding Models Are Plateauing | Why the SaaS Apocalypse is Justified: Will Incumbents Be Replaced? | Why IDEs Are Dead and Do PMs Survive the Next 3-5 Years with Amjad Masad

Shiv Rao · May 16, 2026

Cost is not something to optimize for right now, though a future version of the company will care more

Putting the user first has led them to build at a layer of the stack that already gives them agency over their P&L, which is a bonus rather than the objective

Scope: specific to their product and its place in workflow

22:27 20VC: Lessons from Jensen Huang on "Founder Mode" | How to Know if OpenAI or Anthropic Will Kill your Company | How USV Liking Music Made Them $1BN on an Investment | The Five Year Desert to Product Market Fit & a $5.3BN Valuation with Shiv Rao @ Abridge

Avishai Abrahami · Jul 13, 2026

Vibe-coding companies should prioritize quality over cost optimization right now — he would take 20% more quality over 20% less cost.

The market is brand new and just starting, so the job is to make the product better.

17:56 20VC: Wix's Founder on What Wall St Gets Wrong About AI and Wix | Will Base44 Win the Vibe Coding Wars | The Truth About the Economics of Vibe-Coding | The Buyback Disaster: Lessons Learned with Avishai Abrahami

Jason Lemkin · Jan 4, 2024

OpenAI's revenue growth is generational and epic even if its gross margins may not be good

Going from roughly $100M to $1.4B run rate in a single year is growth at rates that justify the escalating valuations

Scope: gross margins may be poor

4:03 20VC: Predictions for 2024: What Happens to Early Stage VC Funding, Do a Load of Venture Funds Die, What do LPs Do in 2024, Does Figma Kill the M&A Market, Will IPOs Comeback & What Does a Trump Administration do for Startups with Jason Lemkin @ SaaStr

Martin Casado · Jul 28, 2025

Thin margins in AI application companies are not endemic to the business model — they're a board-level choice to prioritize distribution during land grab

With access to cheap private capital during a land grab, the rational trade is margin for distribution, because an incremental user acquired now can be monetized forever and a user not acquired now is lost forever; tech markets have always done this — the early web wasn't monetized at all

Scope: contrary to his own experience that low margins are structural

22:29 20VC: a16z's Martin Casado on Anthropic vs OpenAI: Where Value Accrues | Cursor vs Replit vs Lovable: Who Wins and Who Loses | The One Sin in AI Investing | Why Open Source is a National Security Risk with China

Byron Deeter · Aug 25, 2025

What matters when investing is the future margin profile, not current margins; very poor early gross and net margins are acceptable in capital-intensive transformative businesses.

Snowflake had negative gross margins late into its life cycle and Stripe, Twilio and Shopify went through similar capital-intensive journeys before leverage kicked in; venture investments are not cash-flow or gross-margin based in the short term.

Scope: requires conviction that margin leverage arrives over the horizon

7:35 20VC: Do Margins Matter in AI? | Is Defensibility Gone For Good? | Is Vertical SaaS Dead in a World of AI | What SaaS Rules Are BS and No Longer Apply in a World of AI | The Future of Venture: Why Chanel vs Walmart is BS with Byron Deeter

Jonathan Ross · Sep 29, 2025

Groq should keep its margins as low as possible so long as the business stays non-volatile, rather than pursuing high margins

The only reason to hold a high margin is to be able to pull in cash when needed; because demand for compute is so high that Groq could raise prices at will if it had to, it retains that option and can afford to run low margins now

Scope: conditional on the business remaining non-volatile; depends on retaining the ability to price higher when needed

63:25 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Lin Qiao · Jul 20, 2026

30–40% gross margins are not the new normal for the inference layer; they reflect a deliberate choice during hypergrowth rather than a structural ceiling

Margin optimization is a constraint problem, and imposing margin constraints slows innovation and growth — during high-experimentation phases you shouldn't overbuild or optimize prematurely

Scope: speaking for Fireworks, not other companies

47:44 20VC: Are OpenAI and Anthropic Overvalued? The Open-Source AI Reality | How Token Costs Will Fall 10x And Usage Will Explode 100x | The Future Is Not One AGI; It's Millions of Specialised Models with Lin Qiao, Founder and CEO @ Fireworks

Matt Murphy · Jul 27, 2026

Margins still matter a lot, but investors are in a tricky period where many great companies currently run at 20-30% margins and probably have a path to 60-70%

Credible paths involve optimizations, decoupling cost structure from inference, and building your own complementary model on proprietary data to lift gross margin

Scope: Lovable assessment is Matt's own judgment

19:17 20VC: Leading Anthropic's First Ever Round | Will Open Source Threaten Anthropic's Business | Do Margins Matter in a World of AI | Why Triple, Triple, Double, Double is Not Good Enough Today | Why Series A is Hard Today with Matt Murphy @ Menlo

Jerry Murdock · Aug 22, 2026

The depressed or zero margins across AI companies are a deliberate land-grab strategy, not a permanent condition — capital-rich players take the customer real estate now and build margin in later

Early in a cycle with new technology it's always a real estate game, like the Oklahoma land rush where people staked ground first and sorted ownership later

Scope: only works if you have capital; early in the technology cycle

25:46 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

Jake Saper · Mar 10, 2025

Investors should not be very concerned about closed-model providers squeezing the margins of application-layer companies

Competition among closed models has driven pricing down so application companies are seeing gross margin increase over time; and open source LLMs are good and improving, giving app providers a credible threat to spin up their own model at near-total margin control

Scope: framed as possibly not a commonly held belief; applies to application-layer companies built on top of models

40:32 20VC: Lessons from Investing $2BN and Returning $8BN in Cash | Why Most Venture Partnerships are Broken | We Sold Salesforce Early and Lost Out on Billions | Are The Best Deals Always Expensive and Competitive with Jake Saper @ Emergence Capital

Yamini Rangan · Mar 18, 2025

AI integration will not degrade HubSpot's margins, because model and inference costs have already fallen significantly since 2023 and the cost curve will keep declining.

Inference costs are falling; and high enough usage to create a margin problem would itself be a good sign — they are still early in getting features into daily use.

Scope: still early in feature adoption

34:51 20VC: HubSpot CEO on Where Value Accrues in SaaS AI | How HubSpot Competes Against Salesforce | Why B2B Is Not a Winner Take All Market | How to Go From SMB to Enterprise an Win | How SEO Dying Changes HubSpot's Business with Yamini Rangan

Harry Stebbings · Sep 29, 2025

Asking whether a company like Canva implementing AI would hurt its margins was a naive question, because the cost of implementation has since come down by about 98%

He cringes rewatching old episodes where he asked whether AI implementation would damage Canva's margins

49:59 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Harry Stebbings · Oct 17, 2025

App-layer companies argue they can expand margin over time via model selection — routing simpler requests to older, cheaper models.

41:38 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

Maor Shlomo · Nov 24, 2025

Current vibe coding margins are not bad and will improve drastically as cheaper open-source models absorb a larger share of prompts via routing

Cheap open-source models are getting much better, and many prompts (e.g. changing a button colour) don't need a frontier model; routing them to smaller models also gives users a faster experience

Scope: depends on building the switch/proxy layer; frontier models still used for hard prompts

29:25 20VC: Base44's Maor Shlomo on How Vibe Coding Will Kill SaaS and Salesforce | Why it is BS that Vibe Coding Platforms Do Not Have Defensibility and Bad Margins | Why He Worries About Google, Not Replit and Lovable | Why Long Anthropic, Not OpenAI?

Lucas Swisher · Feb 23, 2026

AI application companies' early margins will improve substantially because token costs are falling fast and companies will optimize across their own models, frontier models and small cheap models per workload

The cost curve is coming down so quickly that a margin that was deeply negative two quarters ago is positive now, and workload-appropriate model selection lets companies optimize further

Scope: Presented as the bull case

33:38 20VC: Inside Coatue's $70BN Machine: Why Price Matters Least | Why Mega Markets are the Most Important | How to Assess Durability of Revenue and Margins in AI with Lucas Swisher

Elias Torres · Mar 21, 2025

Companies must pursue profitability and growth simultaneously rather than trading one off, and the fundamentals that determine success — brand and product differentiation — have not changed.

He took a long time to learn from Cancel that focus doesn't mean doing only one thing; what changes is how you scale and grow, not the underlying determinants of success, and there is no single company model that applies to everyone.

Scope: different products allow very different growth styles, e.g. Cursor

42:11 20VC: Selling Drift for $1.2BN is the Biggest Failure: What No One Tells You About Selling Your Company | Why Incumbents Are Slower & Worse Than Ever | Why the Most Valuable Companies in a World of AI Will Not Have More Than 100 People with Elias Torres

Zach Lloyd · Oct 17, 2025

The right response to thin AI-coding margins is to make model usage more efficient and re-price so revenue rises with usage rather than falls — without harming growth

A large share of every dollar of revenue goes to the model providers, so pricing must be aligned with customer value; the constraint is doing this without damaging fast growth

Scope: must not harm growth; the alternative of just growing and raising more capital may be smarter

34:23 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

Maor Shlomo · Nov 24, 2025

Margins should still be constantly watched, because running a lean, healthy business is what allows sustainable growth without hitting the brakes later

At his previous, capital-heavy company he learned that unhealthy margins force a reckoning later; healthy margins, retention and spend improve every other aspect of the business

Scope: while planning for model costs, which are 90-99% of spend, to change aggressively every few months

37:03 20VC: Base44's Maor Shlomo on How Vibe Coding Will Kill SaaS and Salesforce | Why it is BS that Vibe Coding Platforms Do Not Have Defensibility and Bad Margins | Why He Worries About Google, Not Replit and Lovable | Why Long Anthropic, Not OpenAI?

Lin Qiao · Jul 20, 2026

Companies should optimize gross margin continuously but never over-optimize it, because optimizing for margin at the expense of growth is a disaster outcome

Optimizing gross margin is really optimizing for product differentiation, which should be continuous; but you can climb to a high margin number by not growing at all, which is an absolute disaster

Scope: assumes product-market fit exists

54:55 20VC: Are OpenAI and Anthropic Overvalued? The Open-Source AI Reality | How Token Costs Will Fall 10x And Usage Will Explode 100x | The Future Is Not One AGI; It's Millions of Specialised Models with Lin Qiao, Founder and CEO @ Fireworks

Zach Lloyd · Oct 17, 2025

It's crazy to claim margins don't matter; a fast-growing company with negative unit economics will eventually run out of money, so growth needs to be sustainable

At some point you run out of money if you're growing really fast on negative margins, even if you could probably raise more

Scope: accepts that some of the cost can be accounted for as marketing spend

24:12 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

Oren Zeev · Feb 2, 2026

The belief that only growth matters is dangerous

He has seen this pattern repeat many times before; humans are self-validation machines rather than truth seekers

0:00 20VC: 50% of Funds Will Go Out of Business | Why Growth Expectations Today are BS and Will Not Last | Why Oren Zeev Takes $0 Management Fees But 30% Carry | Why GPs Should Not Tell LPs Their Strategy

Oren Zeev · Feb 2, 2026

A company growing 2x with very healthy economics is a better investment than one growing 3x with unhealthy economics

As long as the market is large enough to sustain the growth for the next few years, the healthier compounder wins

Scope: assumes no same-scale competitor is growing much faster; assumes a large enough market

14:40 20VC: 50% of Funds Will Go Out of Business | Why Growth Expectations Today are BS and Will Not Last | Why Oren Zeev Takes $0 Management Fees But 30% Carry | Why GPs Should Not Tell LPs Their Strategy

Oren Zeev · Feb 2, 2026

Founders should grow healthily rather than optimize purely for top line; growing crazily while ignoring economics is a disaster waiting to happen

When you have the choice between fast sustainable growth and pure top-line optimization, the latter blows up

Scope: exception in rare land-grab markets with fast-growing competitors, where you have no choice but to play the game; growth is still super important

17:06 20VC: 50% of Funds Will Go Out of Business | Why Growth Expectations Today are BS and Will Not Last | Why Oren Zeev Takes $0 Management Fees But 30% Carry | Why GPs Should Not Tell LPs Their Strategy

Des Traynor · Nov 15, 2023

Margin compression will be a second major cause of AI startup mortality, because the B2B SaaS model assumes 70-90% gross margins that expensive third-party API calls destroy.

LTV/CAC dynamics in SaaS presume very high margins; if core workflows run on costly API calls you don't own, revenue coming in the front door goes straight out the back door.

Scope: applies to startups whose core workflows depend on third-party model APIs

22:06 20VC: How to Survive and Thrive in a World of OpenAI, Are LLMs Being Commoditised, Where Does the Value Lie; Infrastructure or Application Layer, How Apple Could Win in a World of AI, How Amazon Could Threaten OpenAI and Why Google Struggle with Des Trayn

Harry Stebbings · Nov 17, 2025

AI application layer companies have terrible margins and cost more, not less, to build and run

They require large engineering teams, and companies like Replit and Lovable pass roughly 80% of revenue through to Anthropic

31:14 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Harry Stebbings · Aug 3, 2026

Investors are not over-rotating on margins; margins genuinely matter, and mid-30s-margin AI businesses are materially different from traditional 65-80% software margins

Businesses like Fireworks operate at mid-thirties margins, far below classic software margin profiles

48:48 20VC: 70% of Neolabs Will Die | There Will be a $100BN US Open-Source Model | Data is a Trillion $ Market | Governments Cannot Regulate Models: It is Too Late | The Cyber Attacks to Come Will be Insane with Anastasios Angelopoulos @ Arena

Mike Cannon-Brookes · Oct 13, 2025

It is too early to judge AI companies' margin profiles: monetisation models aren't understood, pricing schemes change every three months, and the time-degradation of model value is unresolved

A model that costs a billion dollars to train may not earn that back in the nine months before the next model supersedes it; these questions will get sorted out but nobody can say how yet

Scope: says the issues will eventually be resolved in some form

39:58 20VC: Atlassian CEO on Why Everything is Overvalued & Are We in an AI Bubble | Do Margins Matter & Does Defensibility Exist in an AI World | Is Per Seat Pricing Dead & The Future of Vibe Coding with Mike Cannon-Brookes

Lin Qiao · Jul 20, 2026

Gross margin calculation has become genuinely hard because model depreciation now vastly outpaces the six-year hardware depreciation schedule

Hardware used to refresh every three years and depreciate over six; now one vendor ships three SKUs a year, new models run best on newest hardware, and models peak in value before the next one arrives — so in three years you'd be running three-year-old models on nine-generation-old hardware

Scope: value of old hardware is 'questionable', not zero — could still matter in a war scenario

53:13 20VC: Are OpenAI and Anthropic Overvalued? The Open-Source AI Reality | How Token Costs Will Fall 10x And Usage Will Explode 100x | The Future Is Not One AGI; It's Millions of Specialised Models with Lin Qiao, Founder and CEO @ Fireworks

Harry Stebbings · Feb 9, 2026

ElevenLabs should subsidize and cut prices to own the market as a land grab rather than risk churn on price

They've raised $500M and could raise far more, so they can afford to buy the market for a year or two

63:02 20VC: Is SaaS Dead in a World of AI | Do Margins Matter Anymore | Is Triple, Triple, Double, Double Dead Today? | Who Wins the Dev Market: Cursor or Claude Code | Why We Are Not in an AI Bubble with Anish Acharya @ a16z

Brendan Foody · Jun 1, 2026 · hedged

Mercor could likely raise prices around 30% without much demand impact but shouldn't, because pricing should be set to win the market over a decade rather than optimize the next six months

High margins invite competition, so leaving no oxygen in the market for competitors matters more than near-term margin capture; the binding constraint is capacity, not demand

Scope: couldn't double prices; capacity, not demand, is the current constraint

34:57 20VC: Mercor CEO on Why Application Layer Companies Have No Defensibility, The Model is the Product | Token Spend Will Exceed Headcount Spend in 5 Years | The True Cost of Hiring AI Researchers in the Valley Today with Brendan Foody

Zach Lloyd · Oct 17, 2025

Reselling model capacity at a discount is not a business worth being in; you must be a premium differentiated product rather than competing on price

If you're reselling Claude for 75 cents on the dollar there's no point to the business; Warp is perceived as a premium tool for serious coding, not as the cheap option

Scope: alternative path: be very strategic and target the super-profitable cohorts

35:54 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

Zach Lloyd · Oct 17, 2025

If forced to prioritize, don't screw up revenue growth — but you must still make real progress on margins, and the bigger long-term risk is being a below-market reseller of intelligence tokens rather than near-term burn

Burn can likely be covered by raising more money, whereas being a below-market reseller of tokens is structurally a bad business; smart investors need to see margin progress

Scope: he does not believe Warp is that kind of reseller; there is nuance beyond podcast-friendly 'growth at all costs'

38:32 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

Brendan Foody · Sep 15, 2025

Running capital-efficiently builds a more durable, sustainable business, even though he wonders whether they should burn hundreds of millions to subsidize supply or demand

Being thoughtful about how markets develop over time is what ensures the business is still around in ten years

Scope: genuinely tempted by the aggressive-burn alternative; says the business does not need to subsidize either side today

48:43 20VC: Mercor: From $1M to $500M in 17 Months: The Fastest Growing Company in the World | How to Think About Margins and Revenue Sustainability in AI | Why Evaluation Benchmarks in AI are BS Today with Brendan Foody

Jerry Murdock · Aug 22, 2026

Fireworks is a roughly ten-times better business than Base10 despite similar valuations, because it is more capital efficient and more willing to make profits

Base10 took contracts like the Cursor deal that delivered revenue and scale but little earnings; putting up a lot of capital without making money puts you at risk

Scope: stated as his opinion; comparison of inference providers

11:27 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

Harry Stebbings · Aug 19, 2024

Canva is suffering margin compression because it has infused AI throughout its product without charging more per seat

Every AI query costs money, so they incur more cost while making the same revenue

33:58 20VC: Chips, Models or Applications; Where is the Value in AI | Is Compute the Answer to All Model Performance Questions | Why Open AI Shelved AGI & Is There Any Value in Models with OpenAI Price Dumping with Aidan, Gomez, Co-Founder @ Cohere

Harry Stebbings · Oct 14, 2024

Adding AI features degrades software margins on every query

The CPO of Canva described exactly this margin denigration per query, and Canva subsequently announced a 300% price increase

64:37 20VC: Investing Lessons from FC Seeding Uber, Airtable and Coupang | Why Pro Rata is the Original Sin in VC | Why Liquidity Has Died in 2024 | Why LPs are Pissed with VCs | The Hard Truth About Seed Fund Economics with David Frankel @ Founder Collective

Maor Shlomo · Nov 24, 2025 · hedged

Highly autonomous coding agents that run unattended for 20-30 minutes risk burning a lot of money when they misunderstand what the user wanted

If the agent goes off and builds something the user didn't actually mean — especially for free users — the compute is spent regardless

30:49 Long running autonomous agents burn margin on misread intent

Andrew Feldman · Oct 6, 2025

Positive margins are what allowed Cerebras to raise more money at a higher valuation from better investors than peers with negative margins

As you prepare to be a credible public company, investors scrutinise margins — it's how you move from being an idea to a real company

42:44 Positive margins attract better investors and higher valuations pre ipo

Anish Acharya · Feb 9, 2026

Investors should unbundle margin analysis for AI companies: treat organic month-one traffic as traffic rather than acquired users, count the negative margin on free trials as CAC, and judge the business on the durable margin profile of converted paying users

Much of the top-of-funnel is organic and free to acquire, so blending it into gross margin misrepresents the underlying business

29:44 Treat subsidized free usage as cac not cogs

Anish Acharya · Feb 9, 2026

The fact that AI products carry real marginal costs is a good thing because it forces business model hygiene early

Founders can no longer do 'field of dreams' investing — build a free product and figure out monetization someday; they have costs today like every pre-software small business did

62:32 Marginal costs are healthy forcing early business model hygiene

Lucas Swisher · Feb 23, 2026 · hedged

AI-native companies may end up with higher terminal operating margins than the last generation despite lower gross margins, because AI makes engineering, sales and legal functions leaner

Lower gross margin is substituted for lower OpEx, and operating margin is ultimately what matters

35:16 Lower gross margin offset by leaner opex so operating margin wins

Anton Osika · Aug 18, 2025

Vibe-coding unit economics will improve over time as revenue shifts from pay-to-build usage to subscriptions from users who love the platform, leaving AI compute as only a small part of cost

Early on users are essentially paying to build, but if the platform creates enough value users stay on subscriptions

12:13 Revenue mix shift to subscriptions improves vibe coding margins over time

Everett Randle · Nov 10, 2025

AWS is the model for AI companies: lower gross margins are fine when the customer spends multiples more, and for the right AI companies in the right markets larger revenue per customer will make them far more valuable than SaaS peers despite worse margins

AWS has lower gross margins than Adobe but is the largest line item for any large software business, which is why it is an enormous business

26:36 Lower margins fine when revenue per customer is multiples higher

Harry Stebbings · Oct 17, 2025

Advice to simultaneously keep growth high, improve margins, move into enterprise and maintain product quality is idyllic — startups are a game of explicit trade-offs and choosing what you give up for a specific gain

You cannot sustain growth, fix margins, enter enterprise and perfect product all at once; the useful decision is being willing to trade a little growth for margin improvement and naming the consequences

37:09 Growth margin enterprise and quality cannot all be optimized at once tradeoffs are mandatory

Shiv Rao · May 16, 2026

The trigger for cost optimization is when costs start getting in the way of creating more impact or raising more money — not going public

The mission determines the financing path; IPO is a means, not a milestone that forces discipline

23:04 Impact or fundraising friction is the trigger not an ipo

Brendan Foody · Sep 15, 2025

Margins do matter, but whether negative margins are acceptable is contextual: aggressive subsidies can make sense if distillation makes serving an order of magnitude cheaper and the customers are sticky enough to justify the LTV

If you can distill models to be 10x more efficient within twelve months, running aggressive serving margins today can pay off — it comes down to stickiness and whether subsidies buy large long-term LTVs

36:55 Negative margins justified when distillation cuts serving cost and customers are sticky

Harry Stebbings · Sep 15, 2025

Burning cash as a loss leader is right only when you face real competitive pressure and can undercut rivals asymmetrically; absent that, stay efficient because large cash reserves become a problem you have to make work by buying growth

With asymmetric undercutting ability you can strangle competitors out of the market, but at scale idle cash forces you to buy growth, a stage you don't want to reach

50:09 Burn only with asymmetric competitive advantage otherwise stay efficient

Harry Stebbings · Sep 15, 2025

Foundation model labs are destroying much of their economics to win the current race and can only sustain that for a limited time

55:43 Labs burn margin unsustainably to win market share short term only

Severin Hacker · May 19, 2025

AI content generation improves margins while interactive AI features carry a real per-call unit cost, which is why they sit in the top-tier subscription

Content generation is a one-off cost, whereas live LLM features require paying for every API call, so they must be made economically viable

21:29 Content generation is one off cost while interactive ai features carry recurring per call cost

Oren Zeev · Feb 2, 2026

Focusing on margins too early can indeed hinder a company, and growth remains the more important variable; the right moment to shift to economics depends on the business and competitive environment

In land-grab markets you must win share first and assume margins can be fixed later, but most businesses are not like that

18:19 Timing of the margin shift depends on the market and competition

Andrew Ng · Nov 17, 2025

Margins ultimately matter, but AI companies should build against forecast future costs rather than today's costs

Token prices are falling rapidly and, in his experience, teams can bend their own cost curve down even faster than market token prices fall; the first priority is building a product users love

36:48 Underwrite to forecast future token costs not todays

Cliff Weitzman · May 9, 2026

Companies, like bodybuilders, should run distinct bulking (growth) and cutting (margin) cycles rather than pursuing both at once

You can't oscillate weekly — you have to commit for roughly six months; trying to grow and cut simultaneously is like pressing the gas and the brake at the same time

13:27 Alternate distinct growth and margin cycles

Sam Altman · Apr 15, 2024

The marginal-cost-versus-marginal-revenue question for AI is a non-issue.

You only need to believe compute prices keep falling while the value of better models rises, and the equation resolves itself easily.

16:32 Falling compute cost and rising model value resolve margin economics

Jonathan Ross · Sep 29, 2025

You must eventually reach profitability or at least breakeven, and the real function of high margins is absorbing volatility — but a high margin also invites competitors in, so margin is a trade of stability against competitive moat

With razor-thin margins a market move can leave you unable to raise money or get a loan; at the same time 'your margin is my opportunity' for competitors

47:31 Eventual profitability required margin tradeoff between stability and competitive exposure

Jonathan Ross · Sep 29, 2025

A company should keep margins as low as it can while remaining stable and make its cash flow on volume instead, because charging a high margin puts you at odds with your customer

Margin restraint builds equity and trust with customers the way brand equity does — trust pays interest — and you want maximum alignment with the customer rather than extracting from them

48:45 Low margin strategy builds customer trust and alignment cash flow from volume

Jonathan Ross · Sep 29, 2025

Hardware companies are in a better position than other AI companies because their units carry positive margin at the point of sale

Unlike software-only AI businesses, you know the margin the moment you sell the hardware

62:06 Hardware margin known at point of sale gives structural advantage over software ai

Bucky Moore · May 5, 2025

In three to five years the margin profile, growth rates and revenue scale of the foundation model companies will look materially better than today

They are only beginning to layer consumer and enterprise applications on top of the capital-intensive model asset, which reshapes the business

7:50 Model companies margins improve materially as consumer enterprise apps layer in

Lin Qiao · Jul 20, 2026

Unlike the SaaS era, product-market fit and a durable business are now two separate things — companies can find PMF and still scale into bankruptcy

In SaaS, compute was a commodity so COGS didn't matter once you found PMF; with AI, inference cost means scaling usage can destroy the economics

15:51 Inference cogs means pmf no longer guarantees a business

Mati Staniszewski · Sep 8, 2025

ElevenLabs' unit economics are much healthier than those of comparable AI application companies because it controls its own research, product and distribution.

Owning the full stack means they aren't dependent on someone else's model costs, and they can trade cost optimization for speed when releasing a new model.

44:42 Owning the full stack improves unit economics vs app layer peers

Andrew Feldman · Mar 24, 2025

Gross margin is a measure of technical differentiation — a negative gross margin business is selling a commodity whose value creation isn't recognized by the market — and Cerebras' technology is what lets it hold margins where others can't

Traditionally gross margins reflected how differentiated your technology was; if you're running negative gross margins it speaks for itself

40:16 Gross margin level signals genuine technical differentiation negative margins signal commodity

Alexander Embiricos · Feb 21, 2026

Companies should accept a hit to margin this year to win the race to deploy agents and connect them into customer systems, because costs will come down significantly and those integrations will be very sticky

If this is the year agents get deployed broadly at work, it's also the year they get wired into all the various systems, which creates stickiness worth racing for

53:53 Accept a margin hit now to win sticky agent integrations