Six investors managing over $2 billion in committed capital gathered at a recent Corum-sponsored tech conference. What they revealed about AI investment strategy contradicts almost everything founders believe about building fundable companies.
Here’s what they say actually matters.
The Core Misunderstanding
Steve Schlenker runs DN Capital. His fund has deployed $700 million and returned about $1 billion in exits. His firm’s AI portfolio recently crossed $400 million in aggregate revenues.
When asked about AI investment strategy, he said something surprising:
“Ultimately none of us are investing in AI companies. We’re investing in great companies that just use AI to be even better.”
This isn’t semantic hedging. It’s the central insight that separates funded companies from unfunded ones.
Every panelist — representing firms across Stockholm, Silicon Valley, Dallas, London, and Santa Monica — rejected the premise that AI itself creates investable businesses. Their portfolios prove the point. The pattern that emerges may be uncomfortable for most founders to hear.
Why Most AI Companies Have Zero Gross Margins
Schlenker explained the economic problem with brutal clarity:
“If you’re spending your modeling costs on building your model, and then you run a proprietary model, you have 90% gross margins. If you’re using someone else’s model for your inference, you might have zero gross margins.”
Translation: Wrapping a nice interface around ChatGPT or Claude doesn’t create a business. It creates a cost structure that collapses under scrutiny.
This matters for three reasons:
First, strategic acquirers care about margin structure. A SaaS company trading at 8–12x revenue assumes 70–85% gross margins. If you drop to 20% gross margins, that multiple compresses to 2–3x. Your $50 million revenue company just lost $300–400 million in enterprise value.
Second, financial buyers model cash flow. Private equity firms structure acquisitions around debt service coverage. Businesses with compressed margins cannot support leverage, which eliminates an entire class of potential acquirers.
Third, competitive moats erode. Companies dependent on third-party AI platforms face three simultaneous threats: platform risk (pricing changes, access restrictions), commoditization risk (competitors can build identical solutions), and substitution risk (the platform itself may compete directly).
The Only Moat That Survives
Brian MacMahon runs Expert Dojo. His firm has invested in 300+ companies across Africa, India, Latin America, and the US. He distilled investment strategy to a single sentence:
“AI does not work without accurate data. Period.”
Every panelist converged on this conclusion. Data, specifically proprietary data that improves continuously through usage, represents the only defensible competitive advantage.
McMahon’s fund invested in Matchbook AI, a data matching and data enrichment platform. The company does $3–5 million in annual recurring revenue. They’ve penetrated 10% of Fortune 200 companies with essentially zero marketing spend.
Why? They solve the foundational problem that enables AI deployment. Their product generates proprietary data with every customer interaction, and competitors can’t replicate the dataset without equivalent customer volume. This means that the business compounds in defensibility over time.
Contrast this with most AI startups. They process data, generate insights, or automate workflows — all valuable functions. But they don’t generate proprietary data through their core operations. Their product improves through engineering effort, not through usage. This creates a linear scaling problem rather than a compounding advantage.
Three Portfolio Examples That Illustrate the Pattern
Schlenker shared three investments that returned $1.4 billion in exits this year:
Incode Technologies: Trust Infrastructure
Revenue growth: $6 million → $170 million
Use case: Biometric authentication for identity verification
The moat: Every verification attempt generates training data. The model becomes more accurate with scale. Competitors starting from zero face an accuracy gap they cannot overcome without equivalent transaction volume.
The market thesis: Identity verification is currently a $5 billion market. It will reach $150 billion within 20 years as AI-generated fraud proliferates. The company that owns the most comprehensive biometric dataset captures disproportionate value.
Brigit: Financial Inclusion
Revenue: $150 million at exit
Use case: AI-powered overdraft protection for low-income Americans
The moat: Proprietary spending pattern data from millions of transactions. The model predicts financial distress with increasing accuracy as the user base grows. Saved users $1.1 billion in bank fees, creating both social impact and customer retention.
Accent Softening Technology: Empathy Preservation
Bookings: $10 million per quarter (enterprise sales)
Use case: Real-time accent moderation for call center workers
The moat: Syllable prediction data accumulated from thousands of hours of live calls. Workers quit call center jobs after 6–7 months due to accent-based harassment. This technology extends tenure, improves satisfaction scores, and increases calls completed per hour — creating measurable ROI for enterprise buyers.
Notice the pattern. Each company generates proprietary data through core operations. The product improves automatically as usage scales while competitors face structural disadvantages that engineering talent alone cannot overcome.
The Capital Reality That Nobody Discusses
McMahon delivered what was probably the most uncomfortable truth of the session:
“LPs who used to be rich are now poor. The US hasn’t had any decent IPO activity since 2001, except for sporadic stuff. Real estate is in a hell hole with interest rates at five-plus percent. Everyone’s a VC now, everyone’s an angel, everyone has the same thesis, everyone wants to go after AI companies.”
This creates a paradox. More investors chase fewer viable deals and so capital concentrates in proven companies. Early-stage founders who would have raised $2–5 million easily now struggle to secure any financing.
The implication: Companies without institutional backing may represent acquisition opportunities at compressed valuations — provided they’ve achieved organic revenue traction and built defensible data assets.
For founders, this means the optimal strategy may involve bootstrapping to $5–10 million in revenue, then positioning for strategic acquisition rather than pursuing traditional venture scaling.
Geographic Arbitrage Still Works
McMahon’s portfolio spans four continents and the investment pattern reveals distinct regional opportunities:
Africa: Fintech dominates. Winich Farms (approaching unicorn valuation) and two undisclosed fintech platforms solving remittance friction received significant follow-on capital. The insight: “Sending money across developed economies was broken. Something was going to fix it.”
India: Convenience economy drives returns. Despite producing 23,000 engineers annually, Indian unicorns emerge from service delivery — massage therapists, drivers, domestic services. The demographic dividend creates massive addressable markets.
Latin America: Both diaspora capital flows and financial infrastructure gaps create fintech opportunities. The ecosystem differs fundamentally from other regions due to remittance patterns and regional integration challenges.
Europe: Tanya Horowitz of Butterfly Ventures operates a €50 million deep tech fund. Her portfolio company Zenniz exemplifies the European thesis: hardware/software integration solving vertical-specific problems. In this case, AI-powered tennis coaching analysis.
The lesson: Viable AI opportunities distribute geographically based on regional infrastructure gaps, regulatory environments, and labor cost arbitrage — not technological sophistication.
When the Next VC Exit Window Opens
Both Brian McMahon and Steve Schlenker project the next significant VC liquidity cycle occurring in 2027–2028.
Schlenker’s reasoning: “VC exit markets go in every seven-year cycle. We had a big one around 1999–2000, a small one before the global financial crisis, a small one in between, a pretty good one in 2021. So, 2028 feels about right.”
McMahon was more direct: “Prepare for the greatest economy the world has ever seen. Trump won’t stop himself from creating it. Ego dictates. Drive growth like crazy today. Become a media company. Create a movie. Be the hero in that movie. Don’t beg for money today. Beg for customers today. Then get ready towards the end of 2026 to go hard.”
The strategic implication: Companies should optimize for customer acquisition and operational metrics through 2025-early 2026, then position for growth equity or strategic acquisition during 2027–2028.
What Growth Equity Actually Funds
Tara Sahiri manages investments for Mainsail Partners; a growth equity firm focused on B2B software. Her criteria provide some insight into what later-stage investors actually evaluate:
- Vertically oriented market leaders
- Product-market fit with long-term growth potential
- Mission-critical software for unique workflows
- Recurring revenue and solid unit economics
- Sticky products with high retention rates
Notice what’s absent: AI differentiation. Sahiri confirmed that generative AI hasn’t altered their investment thesis. Instead, it “expanded the scope and speed of what’s possible for our companies.”
Two of her portfolio companies, RentVine (property management) and Nspire (fire inspection), launched AI assistants enabling natural language task execution. The AI enhanced existing workflow software rather than creating new categories.
This matters for positioning. Mature software companies implementing AI thoughtfully may command higher multiples than pure-play AI startups, particularly when buyers seek immediate revenue accretion.
The Vertical Market Advantage
Sahiri’s emphasis on vertical specialization is an important insight. “Mission-critical software tailored for unique workflows” creates switching costs that horizontal AI platforms cannot replicate.
As AI capabilities commoditize through model proliferation, vertical depth combined with proprietary customer data creates defensible positions. Compare this with generic AI tooling, which faces continuous competitive pressure and margin compression.
For founders, this suggests a clear path: Solve a specific, deep problem for a defined customer segment. Use AI to solve it better than alternatives. Generate proprietary data through solution delivery. Scale within the vertical before attempting horizontal expansion.
The Transaction Readiness Imperative
Gabriela Smith of LatAmplify Ventures delivered practical guidance as well:
“Your M&A journey already started. Act like you’re going to be exiting tomorrow because you don’t know when the buyer or investor is coming. They just show up.”
This matters particularly for AI companies because technical due diligence extends beyond traditional software evaluation. Buyers examine:
- Data provenance and ownership
- Model training infrastructure and costs
- Inference cost structure and scalability
- Dependency on third-party AI platforms
- Proprietary algorithm development
- IP protection for training data
- Compliance with emerging AI regulations
Companies unprepared for this scrutiny face valuation haircuts or deal attrition.
The European Context
Horowitz provided candid assessment of European challenges: “Early stage is really hard right now. There’s a lot of money being poured into larger growth companies, the more profitable ones.”
However, she identified a couple of viable paths: Public funding through the European Innovation Council (€4 billion) and strategic US market expansion. European founders increasingly target US revenue traction to attract transatlantic buyers operating under US valuation frameworks.
Her portfolio company Zenniz illustrates the approach. They secured partnership with Mouratoglou Academy in France, expanded internationally, and achieved 3.5x revenue growth year-over-year. The company is now raising a Series A with strong investor interest — because they built a real business first, then used AI to enhance it.
What Actually Gets Funded
Smith of LatAmplify noted that 90% of her portfolio includes AI components. But she emphasized a critical distinction: “You either implement AI in some shape or form even if you’re a manufacturing company, or you have a solution that uses AI technology.”
AI becomes a feature layer rather than a product category, and this has direct implications for positioning:
Wrong approach: “We’re an AI company that solves X.”
Right approach: “We solve X. AI enables us to do it better, faster, and more cost-effectively than alternatives.”
The difference seems subtle, but it determines fundability.
The Discipline of AI Deployment
Sahiri emphasized disciplined AI adoption: “There’s so much out there. We look at tons of M&A opportunities as add-ons for our portfolio companies. It’s hard not to get overwhelmed. Taking a disciplined approach and thinking through where we can leverage AI to drive the most value for customers — using that as the driving force.”
This counsel matters for both builders and acquirers. AI capabilities should solve specific customer problems with measurable ROI. So, feature bloat driven by technological possibility rather than customer need, will destroy value.
The Uncomfortable Conclusion
The venture capital market has entered a phase of what is being called “selective brutality.” Capital is concentrating in proven business models enhanced by AI, not speculative AI-native companies lacking sustainable margins.
For founders, this creates both challenges and opportunities:
Challenge: Traditional venture paths have narrowed and so most AI startups will fail to secure institutional capital.
Opportunity: Companies that build real businesses solving real problems can leverage AI as a valuation multiplier — provided they generate proprietary data, achieve defensible margins, and target specific verticals.
The question is not whether AI matters. It obviously does.
The question is whether you’ve built a business that would succeed without AI, then used AI to accelerate that success.
Only the latter survives diligence. Only the latter commands premium valuations. Only the latter creates sustainable competitive advantage.
Everything else is narrative and hope. The market has stopped rewarding either.
This analysis based on insights from the 2025 GXS AI Tech Conference investor panel. It featured DN Capital, Butterfly Ventures, LatAmplify Ventures, Mainsail Partners, and Expert Dojo — collectively managing over $2 billion in AI-focused capital deployment. My sincere thanks to the participants and to Corum Group.
By Ivan Ruzic, Ph.D. · Originally published on Medium