Topline · 27 Sep 2026 · From the week of 21 September
If the AI Money Dries Up, Which Companies Burn?
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
A hosts-only episode with Sam Jacobs (CEO, Pavilion), AJ Bruno (CEO, QuotaPath) and Asad Zaman (CEO, STA). They debate whether AI application companies with collapsing gross margins, led by Harvey's fall from about +50% to -50% by June 2026, have a viable path. Sam proposes growth-rate thresholds for tolerating bad margins. Asad argues that moving to cheaper open-weight models degrades the product and leaves application companies dependent on frontier-lab pricing and an exuberant funding market. The hosts also discuss the 2021 valuation hangover, which AJ illustrates with QuotaPath's 100x Series B, whether secondaries have changed founder economics, a surge in tech and AI M&A and what it means for employee equity vesting, and a Bulls and Bears round on an AI-driven catastrophe before 2031.
For founders
- Sam Jacobs's straw man for AI gross margins: above 200% growth, bad margins are tolerable temporarily if there is a credible path to a good business; between 100% and 200% the trajectory must clearly improve; below 50% growth the margins must be good.
- Asad Zaman argues that swapping frontier models for cheaper open-weight models fixes the margin line but can degrade the product, because buyers in high-stakes domains want the best ROI, and application companies stay exposed to model-company pricing.
- AJ Bruno raised QuotaPath's $40M Series B at a 100x valuation in 2021 and says that, with SaaS multiples at 1x or less, he couldn't find a buyer today; Michael Walrath warns that about 90% of late-2020 to 2021 growth rounds ended in broken cap tables.
- Asad Zaman counters that secondaries have changed the math, because founders and employees now take liquidity along the way; Sam Jacobs responds that most companies never get the choice to raise at those valuations.
- Tech M&A rose 41% in 2024 and 68% in 2025 (AI M&A rose 85% in 2025), which Asad argues makes four-year vesting with annual cliffs unfair to employees; the hosts debated monthly vesting, buybacks and performance-based equity.
For revenue leaders
- Asad Zaman's test for AI spend is what you would put $1M behind in the next 12 months, for example an AI that runs a full sales cycle for deals under $20K (maybe $50K); he says middle-of-the-pack models can't do that reliably today.
- AJ Bruno says he would react very negatively if a vendor touching his finance or bookkeeping announced it was moving to a cheaper model, because that is the most sensitive part of his business.
- AJ Bruno is offering QuotaPath's management team 125% of market-rate equity if they hit performance milestones over the next 18 months, rather than tying equity to tenure.
- One host argued that equity today rewards tenure the way per-seat pricing rewards usage, and that AI may eventually allow quantitative performance measurement in functions beyond sales.
What was said 33, most useful first
Scaling AI companies averaged roughly 41% gross margins in 2024 and 45% in 2025, with a projected 52% in 2026; application-layer companies ran lower at 33%, 38% and a projected 45%. Listen
Sam Jacobs cited Iconic data to contrast AI economics with SaaS, where the median company ran gross margins in the mid-70s and the best exceeded 85%. He said token consumption on every prompt and document means AI companies cannot assume the SaaS margin floor.
“So Iconic found that scaling AI companies average roughly 41 % gross margins in 2024, 45 % in 2025, and a projected 52 % in 2026”
Harvey's gross margin fell from about +50% at the start of 2026 to -50% by June, which Sam Jacobs attributed to rapid agentic adoption consuming tokens. Listen
Sam Jacobs said Harvey's cost to deliver $1 of revenue went from 50 cents to $1.50 as agents consumed tokens on behalf of human users. Harvey was growing extremely fast, but every dollar of revenue was destroying gross profit. He said Harvey's path out is training and hosting its own models, including open-weight models, and moving away from frontier models from labs like Anthropic and OpenAI.
“by June, Harvey's gross margin due to rapid, agentic Adoption right agents going out and consuming tokens on behalf of their human users Harvey's gross margin had fallen to negative 50 %”
Sam Jacobs's straw man: tolerate bad gross margins above 200% growth if there is a path to a good business, require a clearly improving trajectory at 100–200%, and require good margins below 50% growth. Listen
Sam Jacobs offered this explicitly as a straw man for his co-hosts to react to. Above 200% growth he will accept terrible margins temporarily, for example to subsidize usage or win market share, but only if the company shows the path to a good business. Between 100% and 200% growth he wants a clear improving margin trajectory, and below 50% growth he needs margins to be good. He did not specify a rule for 50–100% growth.
“Above 200% growth, I can tolerate bad margins temporarily. If you can show me, you have a path to a good business.”
Asad Zaman considers it a rational choice for young founders to ignore gross margins and grow as fast as possible in this market, because the downside now includes significant personal liquidity. Listen
Asad Zaman pointed to Instinct, which he said went from not existing to a $10 billion valuation in a handful of months with 100,000 users, mainly in San Francisco, as an example of investors treating 'madness' as a feature. He laid out the young founder's calculus: drive growth as hard as possible while capital, liquidity and demand are available. The worst case is the bubble bursts and they start again at 34 with about $30M in the bank. He said he would do the same.
“I'm 29 years old like the worst -case scenario is the bubble burst what I the fuel I need is no longer available and I start again as somebody who's about 30 million in the bank at 34 years old”
About 90% of growth rounds from late 2020 to late 2021 left founders with broken cap tables, and that today's founders should learn from them. Listen
Asad Zaman read Walrath's post on the show. It advises founders, especially those who have just raised a mega round, to seek out founders who raised huge rounds at big prices in the 2020–2021 cycle and now live with broken cap tables and disinterested investors. Walrath says becoming 'the next Airtable or Mural' is the likeliest outcome even for today's hottest companies, and that VCs have no incentive to get founders thinking about the downside.
“The above describes 90 % of growth rounds done between late 2020 and late 2021.”
AJ Bruno raised QuotaPath's $40M Series B at a 100x valuation in 2021 and says that, with SaaS multiples now at 1x or below, he couldn't find a buyer today. Listen
AJ Bruno said his view in 2021 was that QuotaPath would grow into the valuation, and he still holds it. What he didn't foresee was SaaS multiples falling to 1x or less. He added that a CEO he met at Unbound, running a $50–100M business that is 'doing super well', also couldn't sell even with investors who wanted out.
“So QuotaPath, when we raised a $40 million round Series B at a 100x valuation in 21, my perspective at that time was we're going to grow into it.”
Tech M&A deal counts fell 58% from 2022 to 2023, then rose 41% in 2024 and 68% in 2025, with AI acquisitions up 85% in 2025. Listen
Asad Zaman presented M&A data covering tech companies broadly. AI acquisitions rose 16% from 2023 to 2024 and 85% from 2024 to 2025. Q2 2026 acquisitions were up 49% on Q2 2025. He said companies are buying other companies at a much higher rate than people realize.
“You had a 41 % increase in 24 versus 23 and a 68 % increase in 25 versus 24.”
Asad Zaman proposes balancing pro-rata vesting with a first-year buyback right for underperformers and full vesting on acquisition. Listen
Asad Zaman said simple equity structures over-protect the company, hurt some people and make it harder to attract great talent in a tight market. His balance is pro-rata vesting plus protections. If someone is fired for poor performance within the first year, the company can buy the equity back at a multiple of the grant price or require exercise within a set period. If an acquisition happens, everything vests.
“And so instead, if you just say, hey, you're not performing, we fire you, then you don't get to hold on to that equity forever.”
AJ Bruno is offering QuotaPath's management team 125% of market-rate equity, contingent on hitting performance milestones within 18 months. Listen
AJ Bruno said he wanted to galvanize QuotaPath's management team around a performance structure. He benchmarked market equity levels and offered each leader 125% of market if they hit defined milestones over the next 18 months. A co-host said they love performance-based equity in theory but find it hard to apply to certain roles.
“I'm going to give them each 125 % of what the market is. if we can hit these milestones from a performance standpoint in the next 18 months.”
Listen to the episode Sales team, hiring & comp Link to this
Moving from frontier models to self-trained open-weight models is a real downgrade in quality, however the move is marketed. Listen
Asad Zaman responded to Harvey's plan to train and host its own open-weight models by saying companies use fancy language about training to disguise the downgrade. He likened it to a firm that only hired from Harvard switching to a mid-tier university and claiming nothing changed. He added that law firms are 'trying to do deterministic work with probabilistic technology' and questioned how happy they would be to lose access to the best models.
“it's like going from hiring people only from Harvard to then going and hiring people from the worst university you can find or some mid -tier university and saying, It's the same thing. It's not the same thing.”
Asad Zaman expects gross margin pressure on AI application companies to persist, because buyers in banking, private equity and law will demand the best possible ROI and so force vendors onto frontier models. Listen
Asad Zaman said the application company's job is to build product on top of the best intelligence so the customer gets ROI. He does not believe customers at investment banks, private equity firms and law firms will accept less than the best potential ROI. Vendors will therefore be forced to keep paying for top models, leaving ongoing gross margin concerns that he called 'very concerning'.
“I don't think these customers that sit in investment banks and private equity firms and law firms are going to be looking to not have the very best potential ROI, which means you're going to find yourself forced to use these models to serve your customer.”
AI application companies are at the mercy of model companies' pricing decisions. Listen
Asad Zaman said that when a model company prices low to capture market share, the application company's business looks good. If the model company decides it can capitalize on price as much as it wants, the application company is 'screwed'. He called it a tough position for the application layer.
“you're also then completely at the behest of these model companies. Like if they decide to provide you a cheap model, like an OpenAI is using pricing to capture market share, you're in a good spot, your business looks good.”
Great AI application-layer businesses will exist, but they will behave like roller coasters rather than the stable SaaS businesses operators are used to. Listen
When AJ Bruno asked whether customers would just evaluate the base models themselves, Asad Zaman said product matters. He said the intelligence may be better in one tool while the product sells better in another. He thinks product matters most in domains where institutional knowledge must be extracted and turned into workflows and integrations. He described the swing at scale: a smarter model drives usage up but pushes gross margins down, and the company then has to rebalance.
“I think there would be great application layer businesses but these are not the same sort of application layer businesses of the sort that we're used to.”
AI application companies running on bad margins and usage-driven fundraising can only survive in an exuberant funding market. Listen
Asad Zaman described the pitch as: margins are bad, but look at the usage; models will get cheaper; give us more money. He said he doesn't disagree with playing this game, since it is 'the game on the field'. But a lot has to go right, and an actual 'crater' rather than a speed bump would bring a lot down.
“But you cannot sustain these businesses on anything other than an exuberant market and the downstream risk that people have been able to take with that exuberance in their mind, fueling them. If that goes away, this shit burns.”
Sam Jacobs raised the possibility that widespread use of cheaper open-weight or small models could be the deflationary event that undermines frontier-lab economics and deflates the AI system. Listen
Sam Jacobs framed this as 'one argument'. OpenAI, Anthropic and others may need to keep selling tokens at expensive rates. If companies move use cases from $100K–$1M per month in frontier-model spend to small models on their own cloud cluster or machines, application gross margins rise. The resulting pressure on the frontier labs, though, could take the air out of the whole system. Asad Zaman noted that NVIDIA wins either way, because open models are also trained and served on its chips.
“it puts such pressure on their economics that that becomes the deflationary event that takes the air out of the entire system.”
Open-source models consistently trail the frontier by about six months, so companies won't move important workflows to them. Listen
Asad Zaman said the most advanced open-source models appear heavily distilled from the best frontier models. They can catch up and compete, but frontier labs then release what they have been holding back, leaving a roughly six-month gap. Some AI uses won't need the smartest model, but he said the exciting promise is 'a data center full of geniuses'. For a valuable workflow, a buyer offered a model 25% less effective will pay for the best.
“if you have Astra and then you have something that's 25 % less effective and you have an important workflow, the one that you allocate value to being solved and you're willing to pay for, you will use Astra.”
Asad Zaman expects frontier model prices to stay high until data center supply increases, and sees US resistance to building data centers as the bottleneck. Listen
Asad Zaman framed model pricing as supply and demand: demand is there, so prices fall only if supply grows. He said resistance in the US to building data centers means the country is not building the infrastructure needed for the inference it requires, so prices will stay high until that is figured out. The conversation then noted that AI companies have been 'horrible' at getting communities to support new data centers; it is unclear whether Asad or the co-host who followed said this.
“The only way for price to come down is if supply starts increasing.”
Asad Zaman proposes judging AI investments by asking what you would spend $1M on in the next 12 months, such as an AI that runs complete sales cycles for smaller deals. Listen
Asad Zaman said the bar is something that really works, that you can trust to have done the job correctly, and that would change the company. His example was believing AI can run a sales cycle start to finish for anything under $20,000, 'maybe 50'. He said middle-of-the-pack models can't do these things effectively today and would add cost and risk. With the current top model, though, 'we're not that far', and the next frontier releases in about six months might do much more.
“think about what you'd be willing to spend a million dollars on in the next 12 months on AI.”
One host predicted that some AI application companies will position themselves as never compromising on model intelligence, funded by raising more money to subsidize it. Listen
The hosts discussed a counter-positioning in which a vendor promises to always use the very best model. One of them said such a promise would get their attention. Another said companies have to say this today. That host predicted the warning sign will be vendors that start hiding behind claims that workloads are spread across their own model, which tops a benchmark they also created.
“Don't you think also companies will take the counter position of basically going into the market and saying we are the application layer company that will not compromise on the intelligence”
Asad Zaman puts the odds at roughly 50-50 that the AI boom produces many corrections over seven to ten years without a bust, provided revenue keeps up with investment. Listen
Asad Zaman described two schools of thought. One says a bust is inevitable. The other says this platform shift could bring many corrections over seven to ten years but no bust, as long as revenue and growth keep pace with investment. He worries because so much of the buildout is debt-fueled and interest rates are rising, so revenue has to keep up through better intelligence unlocking use cases and application companies delivering more customer ROI. He said he hopes it works but called it a coin flip.
“But you can't look at this game on the field and think, for sure, it's not going to hit a wall. It's kind of a 50 -50 right now.”
Asad Zaman frames company-building as a choice between a profitable, steady business and a 'go for it' generational attempt, and says both are valid. Listen
Asad Zaman used Adam Robinson as the example of the steady path, saying his businesses make $25–30M with a 50% EBITDA margin and that he seems happy. The alternative is attempting something generational, accepting you might hit a wall and fall out of favor. He said this is the first choice a founder makes.
“his businesses make 25 to 30 million with 50 % EBITDA margin. I think he's happy.”
Widespread secondaries mean founders and employees no longer walk away with nothing when an over-funded company stalls. Listen
Asad Zaman pushed back on Walrath, saying lessons from the SaaS era are being over-extrapolated. He said founders are making a lot of money along the way, secondaries are 'massively up', and employees are taking liquidity too. He cited a Peter Walker analysis published by Topline showing secondaries are not limited to the top few companies, and noted that Clay, which he called middle of the pack, has run tenders. He believes this rebalances risk among VCs, founders and employees and makes going for a generational company more reasonable.
“Founders are making a lot of money along the way. Secondaries are massively up.”
Most companies never get to choose between hyper-growth at huge valuations and a slower path, because the market only favors a small set of companies. Listen
Sam Jacobs disagreed with presenting the situation as a fork in the road. He sees two kinds of companies: those 'in the flow' that are skyrocketing, and everybody else. He said his own company, like AJ Bruno's QuotaPath, faces multiples that have completely shifted, and most founders simply have to deal with being out of favor relative to companies like Harvey and Legora.
“And I think the reality is, Most people don't get that choice.”
AJ Bruno chose not to raise as much as a better-funded competitor in 2021 and later improved QuotaPath's cap table by helping an investor who wanted out to exit. Listen
AJ Bruno said Max from GTM Fund told him he wasn't thinking big enough and preferred a competitor raising more, which angered him. As a second-time founder, he believed the competitor's 24-year-old founders didn't understand what that much money would do to their cap table, and he still thinks so. Later, an investor told him QuotaPath wasn't their thing and asked to get out. AJ worked it out for them, which put QuotaPath's cap table in a better spot.
“we were fortunate to have an investor that was like, you know what? This isn't my thing and just raised his hand and said like, I want out”
Companies are acquiring both to buy capabilities, even though AI coding tools make building easier, and to acquire talent, sometimes taking only the founders. Listen
Asad Zaman said an acquisition can give a buyer a jumping-off point even when tools like Codex and Claude Code make building product easier. Companies that raise rounds are also hunting for small startups to acqui-hire. He cited Google's deal for the Windsurf founders to lead a division, which he said left the rest of the team out in the cold, as the kind of deal employees should worry about.
“They're buying capabilities, even though building product has become easier because of things like Codex and Claude Code.”
Rising M&A makes annual-cliff vesting unfair, and that equity should vest monthly as 'credit for time served'. Listen
Asad Zaman said that as acquisition odds rise, an employee may leave with nothing if a deal closes, say, ten months into a four-year grant. That happens unless they negotiated single- or double-trigger acceleration, which junior and mid-level people usually lack the leverage to get. He proposed pro-rata monthly vesting instead of vesting at the end of each year. Without it, he said, the imbalance fuels frustration when markets get choppy.
“it just vests on a monthly basis and I think that's a big shift that has to happen”
Sam Jacobs would accept monthly vesting only with smaller equity grants, because non-contributors stay on the cap table forever. Listen
Sam Jacobs said Pavilion has dead weight on its cap table, including ambassador-program holders with small stakes who won't reply to emails. He worries that immediate vesting lets a poor hire who stays three to six months join the cap table permanently. He suggested keeping the 90-day exercise window alongside monthly vesting, since people fired quickly often don't have the cash to exercise.
“I'm okay with what you're saying, Asad, provided that they get smaller equity grants.”
Sam Jacobs plans to spend the next 12 months asking shareholders who no longer contribute to give their shares back. Listen
Sam Jacobs said that as CEO he plans to approach people on Pavilion's cap table who aren't helping and ask them to return their shares. He acknowledged they can say no.
“Now they don't have to. They can say no, but I plan on being like, listen, you're clearly not helping.”
AJ Bruno once gave up some of his own founder shares to boost equity for three early hires, and two of them stayed six years. Listen
Around 2013–2014, when AJ Bruno's previous company brought in a new CEO, he argued three early hires deserved more equity. The new CEO didn't know them and resisted, so AJ offered to forfeit founder shares to fund it, which caused conflict at the board meeting. It happened: Lindsay stayed six years, James similarly, and Max left after three years without vesting everything. AJ still works with James and Max. His lesson was that equity can be meaningful if granted 'very focused' and 'with purpose'.
“you can create a world where those shares are really meaningful, but you have to do it very focused and you have to do it very clear with purpose.”
One host argued that equity today rewards tenure rather than outcomes, and hoped AI will make quantitative performance measurement possible in every function, not just sales. Listen
A host compared current equity to per-seat pricing or token usage, paying for tenure rather than outcomes, sometimes because a manager never had the heart to fire someone. They said performance is measured well only in sales, because it is tied to revenue, while functions like engineering and HR rely on qualitative assessments. They hope that as intelligence enters businesses, they can measure performance quantitatively and then structure equity very differently. In what they called an employer's market, they said the alternative is to spin up agents.
“It feels like how we do equity right now is AI tokens. It's usage of a thing like that. It's not the outcome.”
He would feel very negative if an AI vendor handling his finance or bookkeeping announced it was downgrading to a cheaper model. Listen
Sam Jacobs posed a hypothetical: a fast-growing AI finance tool, 'Schmarvy', announces it is moving back to a cheaper model. AJ Bruno said his reaction would be very negative. QuotaPath is currently dealing with bookkeeping and accounting issues tied to phishing scams, and he considers that the most important part of his business, so as CEO he is closely attuned to anything AI touches there.
“Very negative and I'll tell you why because we're dealing and this isn't FP&A but we're dealing with a bunch of bookkeeping accounting stuff on the phishing scams right now and it's like the most important part of my business”
AJ Bruno expects a major catastrophic AI-related event before 2031, though not the downfall of mankind. Listen
In the Bulls and Bears segment, the question was whether a rogue agentic swarm or similar AI will cause a global economic catastrophe within five years. AJ Bruno said he believes it will happen but won't bet against the human spirit. He described himself as a prepper with a panic room stocked with six months of food and water.
“I do believe we are headed towards a major catastrophic event as it relates to AI”
Sam Jacobs expects AI-driven incidents such as cybersecurity breaches, but not a global economic catastrophe, before 2031. Listen
Sam Jacobs called himself an optimist and acknowledged he may be naive. He sees AI as a tool and does not expect it to create bioweapons. He found compelling the argument that existing product-safety principles and laws already require that products not harm consumers.
“I think there'll be incidents like cyber security incidents fo sho, but I do not think there will be some global economic catastrophe.”