If the AI Money Dries Up, Which Companies Burn?
The Top Line podcast debate centers on whether AI-driven companies can sustain negative gross margins through rapid growth, examining Harvey's collapse from 50% positive to 50% negative margins, discussing M&A trends showing 85% YoY increases in AI acquisitions, and exploring how equity structures and performance management need reformation as the tech ecosystem becomes more volatile.
Summary
The episode opens with a critical examination of AI company unit economics, particularly Harvey's dramatic margin collapse from +50% to -50% due to agentic adoption consuming expensive tokens. The hosts discuss whether extreme growth (200%+ annually) can justify terrible margins temporarily, with consensus that path clarity matters—Harvey's strategy to build proprietary models and move away from frontier labs like OpenAI/Anthropic provides that path. The discussion reveals Iconic's data showing AI companies averaging 41-52% gross margins versus traditional SaaS's 70-85%, with application layer companies even lower at 33-45%.
A major debate emerges between Asad (optimistic) and the others about market sustainability. Asad argues that current market conditions differ fundamentally from 2021 because: (1) founders now get secondary liquidity along the way, not just at exit, (2) employees benefit more broadly through equity participation, and (3) the balanced incentive structure reduces downside risk while enabling generational ambitions. Counterarguments emphasize that open-weight models create deflationary pressure on frontier labs' economics, and that most companies lack the choice to pursue high-burn strategies—they're simply out of favor compared to AI darlings like Harvey and Legora.
The conversation on model selection reveals a critical tension: customers want the best models (currently Claude/OpenAI) for ROI, but this locks application companies into expensive inference costs and complete dependency on frontier lab pricing decisions. AJ notes lawyers using Harvey appreciate the technology but the domain expertise matters more than model choice, yet cheaper models might degrade outcomes enough to reduce usage.
Regarding M&A trends, Austin presents data showing 68% deal increases from 2024-2025 and 85% YoY increases specifically in AI acquisitions, with Q2 2026 showing 49% YoY growth. Companies acquire for both capabilities and talent, with acqui-hires targeting founders while leaving other employees behind (Google/Windsurf example). This creates equity precarity for employees—companies might be acquired within years of hire, before vesting completes.
The equity discussion proposes monthly continuous vesting instead of cliff vesting to protect employees from acquisition timing risk, with AJ advocating for performance-based equity supplements (125% market rate for 18-month milestones) rather than pure tenure-based systems. Sam pushes back on perpetual equity holding by poor performers, proposing forfeiture agreements and buyback options. The group agrees performance measurement across all functions (not just sales) must improve to make equity meaningful and management effective.
On sustainability of the AI ecosystem, the consensus emerges that everything depends on continued revenue growth matching investment rates, cheaper model inference eventually arriving, and frontier labs maintaining profitability. If markets turn choppy or interest rates rise further, the system lacks structural resilience. The phrase 'this shit burns' encapsulates the existential fragility—businesses built on exuberant markets cannot survive austere ones.
The episode concludes with a bulls vs. bears segment where AJ expresses being bullish on a catastrophic AI event within 4 years (though not civilization-ending), citing his prepper mentality and panic room preparation. Sam remains optimistic, comparing doomsday predictions to disaster movie tropes, arguing that cybersecurity laws and product liability frameworks already exist to prevent worst-case scenarios.
About this episode
<p>Harvey's gross margin fell from roughly positive 50% at the start of 2026 to negative 50% by June, so every $1 of revenue cost $1.50 to deliver. Sam Jacobs, CEO of Pavilion, AJ Bruno, CEO of QuotaPath, and Asad Zaman, CEO of STA, go hosts-only on when AI growth justifies margins like that, why open weight models are no substitute for frontier intelligence, and what happens to the application layer if the exuberant market funding it goes away. Plus, AJ's 100x raise from 2021, a 68% jump in tech M&A and what it means for employee equity, and a Bulls and Bears round on AI catastrophe before 2031.</p> <p>Key Takeaways:</p> <p>- Negative gross margins can be a phase if there's a way out. Sam's straw man: "Above 200% growth, I can tolerate bad margins temporarily if you can show me you have a path to a good business." Between 100% and 200% he wants "a clear improving trajectory," and below 50% the margins have to be good.</p> <p>- Cheaper models fix the margin line but can break the product. Asad likened the switch to open weight models to "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." That leaves application companies "completely at the behest of these model companies."</p> <p>- The 2021 hangover is still on cap tables. AJ raised QuotaPath's Series B "at a 100x valuation in '21," but with SaaS multiples at "1x or less than 1x," he said, "If I wanted to go sell QuotaPath today, I really couldn't find a buyer." Asad countered that secondaries changed the math: "The founders are making a lot of money along the way."</p> <p>Connect with the Hosts:</p> <p>Host: Sam Jacobs, CEO at Pavilion - https://www.linkedin.com/in/samfjacobs/<br /> Host: AJ Bruno, CEO at QuotaPath - https://www.linkedin.com/in/ajbruno3/<br /> Host: Asad Zaman, CEO at STA - https://www.linkedin.com/in/azaman1/</p> <p>Topline is more than a podcast:</p> <p>Subscribe to Topline Newsletter: https://toplinemedia.substack.com/<br /> Check us out on YouTube for the #1 video podcast for founders, operators, and investors in B2B tech: https://www.youtube.com/@TOPLINE-Media<br /> Join the free Topline Slack channel to connect with 600+ revenue leaders to keep the conversation going beyond the podcast: https://www.joinpavilion.com/topline-slack</p> <p>Chapters: <br /> 00:00 The Math Of AI, Deals, And Equity<br /> 02:03 SaaS Margins Were A Law Of Nature<br /> 03:24 Harvey's Margins Went To Negative 50%<br /> 04:18 Sam's 200% Growth Straw Man<br /> 05:46 Open Weight Models Aren't Harvard<br /> 14:13 Open Models Vs. Frontier Lab Economics<br /> 19:01 What Would You Spend $1M On?<br /> 30:08 Bullish Or Bearish On The AI Trade?<br /> 34:15 Walrath's Warning And A 100x Round<br /> 36:37 Secondaries Changed The Founder Math<br /> 42:35 Most Companies Never Get The Choice<br /> 47:32 Corporate M&A Is Surging<br /> 52:04 Should Equity Vest Monthly?<br /> 57:40 AJ's Case For Performance Equity<br /> 1:02:49 Bulls And Bears: AI Catastrophe</p>
Key Insights
- Harvey's gross margin fell from +50% to -50% in June 2025 due to rapid agentic adoption, meaning each dollar of revenue destroyed 50 cents of gross profit while revenue grew extraordinarily quickly.
- Sam proposes a tiered growth tolerance framework: above 200% growth justifies bad margins temporarily if path to profitability is clear; 100-200% growth requires improving margin trajectory; below 50% growth requires good margins immediately.
- Asad argues that secondaries have fundamentally rebalanced founder, employee, and investor risk distribution compared to 2021, allowing employees to capture upside earlier and reducing the downside cliff of total loss at acquisition.
- The hosts debate whether open-weight models create deflationary pressure that threatens frontier lab economics, with Asad asserting that only data center supply increases (not software alternatives) will bring model prices down.
- AJ contends that application layer companies using cheaper models are reselling 'mediocre intelligence' and that customers won't accept substituting Claude for ChatGPT when high-value workflows require the best available models.
- M&A activity increased 68% from 2024-2025 overall and 85% YoY specifically in AI acquisitions, with acquisitions happening at all lifecycle stages, sometimes within the first year of an employee's tenure before vesting.
- The current system traps employees—an acquisition 10 months into employment with year-one cliff vesting means zero equity upside despite contribution, whereas monthly continuous vesting would provide prorated value for time worked.
- Performance-based equity (125% market rate for achieving 18-month milestones) could motivate alignment better than tenure-based vesting, but requires quantifiable performance metrics that most companies lack outside of sales.
- The entire AI ecosystem sustainability depends on a 50-50 bet: revenue growth continues matching investment, or some 'crater' (not merely a speed bump) causes systemic collapse because 'you cannot sustain these businesses on anything other than an exuberant market.'
- The hours claim that if frontier labs stop subsidizing token prices to capture market share and instead capitalize on pricing power, application layer companies with high inference costs face 'pretty ongoing gross margin concerns."
- AJ identifies that open-source models consistently lag frontier models by six months and that every new frontier release resets the competitive gap, meaning there's always incentive to use premium models for important workflows.
- The group concludes that most companies don't have a genuine choice between slow/steady growth and fast/high-burn strategies—only the AI darlings get funding at hypergrowth rates, leaving everyone else 'out of favor.'
Topics
Transcript
You cannot sustain these businesses on anything other than an exuberant market. If that goes away, this s*** burns. In today's episode, our first debate. Is there a safe path ahead for tech companies that incur massive AI inference bills? Case in point, this year, Harvey, the AI legal startup, doubled its revenue while its gross margin tanked. Is this a reasonable growth model or a dead end for all but a privileged few companies? Yeah, but you're at negative 50% margins. You don't have a business. Also on deck, what can we learn about current sky-high valuations from past experience? AJ tells the story of how his business got funded at 100x multiple in 2021 and what happened after…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Topline
CRM As A Business World Model: The Future For GTM Teams? | Keith Peiris, CEO @ Lightfield
Keith Peiris, CEO of Lightfield, discusses why his AI-native CRM is disrupting the go-to-market space, having raised $47M Series A and already serving 5,000+ customers. The conversation covers how Lightfield tackles AI reliability through hybrid deterministic/probabilistic approaches, debates whether the SaaS partner ecosystem is dead in the age of AI, and explores whether ecosystem partnerships or integrated platforms will dominate the future of revenue tech.
SPOTLIGHT: What It Takes to Step into the CEO Role | Eric Anderson, CEO @ Walnut
Eric Anderson, newly appointed CEO of Walnut, discusses his transition from a 28-year go-to-market career to his first CEO role and shares lessons learned in his first six months, including the importance of decision-making, product quality, and building aligned leadership teams across geographically distributed offices.
RevOps for Hypergrowth (Fireworks AI, Perplexity, Exa & More) | CEO @ Go Nimbly, Jen Igartua
Jen Igartua, CEO of GoNimbly, discusses how fast-growing AI companies are scaling despite operational gaps, emphasizing that product strength drives growth while go-to-market becomes a competitive advantage. She argues that RevOps is fundamentally a support function, advocates for technical talent in the form of RevOps Architects rather than siloed GTM Engineers, and shares lessons from building a services business that prioritizes craftsmanship and employee tenure over transactional relationships.
SPOTLIGHT: Skipping the SMB Trap and Building for Big Brands | Matt Allison, CEO & Founder @ Handraise
Matt Allison, CEO and co-founder of Handraise, discusses how his AI-powered media intelligence platform solves enterprise communications challenges by automating manual data processing that plagued his previous company TrendKite. He emphasizes the importance of building for high-value enterprise customers from day one and maintaining discipline in go-to-market strategy rather than chasing rapid growth metrics.
Forget Growth. Your Valuation Depends On Your AI Story | Tomasz Tunguz, GP @ Theory Ventures
Tomasz Tunguz, a venture capitalist at Theory Ventures, discusses how AI is fundamentally reshaping SaaS valuations, sales metrics, and go-to-market strategies. He argues that the market now values AI narrative and enterprise budget increases over traditional growth metrics, while warning that observable productivity gains may actually reflect demand-side changes rather than supply-side innovations.