The Bridge Ep. 22: Who Really Benefits From AI?

Summary

Holden Spaht of Thoma Bravo says the AI conversation has moved from trials to outcomes as customers confront rising token bills and demand predictable ROI. He argues that falling unit token costs do not eliminate overall AI spend pressure because usage is rising and frontier access still costs more, forcing buyers to question where language models add value versus where existing enterprise software already performs deterministic work. Spaht’s core thesis is that general-purpose models are commoditized, while the durable moat sits in proprietary data, workflow embeddedness, precision, compliance, and the human judgment encoded in regulated enterprise processes. He frames the software stack as systems of record evolving into systems of intelligence and then systems of action, with the last step hardest to automate. He also says AI is changing sales, diligence, and board discussions inside Thoma Bravo, but the firm still sees durable growth, strong margins, and long-term opportunity for sector specialists.

Notes

• Thoma Bravo says its portfolio is moving from AI trials to outcome-based spending as customers focus on ROI and the size of their AI bills.
• Token prices are falling, but frontier-model access is still getting more expensive.
• Enterprise buyers are hitting token-budget limits as usage expands faster than expected.
• He questions paying per token for work that deterministic enterprise software can already do at zero marginal cost.
• The model layer is increasingly treated as an API call, not a moat.
• Defensibility comes from domain context embedded in workflows, not from the raw foundation model.
• He uses Jeppesen to show how enterprise AI can be auditable in regulated settings by grounding outputs in FAA rules, weather, tarmac times, home-base conditions, and union contracts.
• In regulated workflows, the output must be auditable, not merely clever.
• Thoma Bravo says many board discussions are now centered on AI because the firm sees pattern recognition across a large portfolio.
• The firm’s framing of enterprise AI is systems of record to systems of intelligence to systems of action.
• Automating human judgment is the hardest step in enterprise AI.
• The last 20% of an enterprise decision often still requires a human-in-the-loop.
• Software learns from those human approval or rejection decisions over time.
• Workflow complexity and embeddedness are central to the moat.
• High-stakes, high-precision use cases are better AI businesses than casual workflow add-ons.
• Proprietary data and domain knowledge are the key defenses against commoditized models.
• Coupa is used as an example of network data scale, with $10 trillion of transactional history.
• Coupa’s network includes more than 3,000 buyers and hundreds of thousands of suppliers.
• CIOs are regaining influence in software buying as data sovereignty and privacy concerns intensify.
• The enterprise sale has become more technical because customers want AI to run on their own data in their own environment.
• Thoma Bravo says more software is being built and sold, not less, despite better coding tools.
• Technical deployed engineers are becoming part of the go-to-market motion for complex enterprise AI.
• General-purpose frontier revenue is still concentrated in coding, chatbots, and customer support.
• The best software companies have proprietary data, workflow depth, and very high gross retention.
• Some portfolio companies have 95%+ gross retention, indicating durable customer value.
• Thoma Bravo says its diligence is now AI-first.
• Internal collaboration has intensified, with Slack channels more active than ever.
• Thoma Bravo believes it can make strong margins on inference by hosting it and charging for predictability.
• Customers value predictable AI spend as much as raw capability.
• The portfolio is compounding with mid-teens top-line growth and 20%+ EBITDA growth.
• Higher multiples would directly support IRR by recognizing terminal value.
• Spaht says sector specialization is how Thoma Bravo believes it generates alpha.
• The firm benchmarks itself against public software stocks as well as private equity peers.
• The conversation is also framed by the legal disclaimer that the material is informational only, not advice.
• Alternative investments are described as complex, speculative, and potentially unsuitable for many investors.
• The disclaimer notes leverage, valuation delays, complex tax structures, and high fees as core risks of alternatives.
• iCapital Markets says its revenue depends on transaction volume and benefits when product sales rise.