The Bridge Ep. 10: AI Broke the Rule of 40. Meet the Rule of 70.

Summary

David Breach makes the case that AI is not a "SaaS apocalypse" but a re-rating event for enterprise software, similar to the earlier cloud transition, with incumbents best positioned to win because they already own customer workflows, data, and trust. He argues the operating target for software is moving from the Rule of 40 to something closer to the Rule of 60 or 70, and says some Vista portfolio companies are already there, with potential EBITDA margin gains of 15 to 20 points. Breach emphasizes that the economics of AI will be determined less by frontier model capability than by the cost of inference, which he expects to be handled by more power-efficient, purpose-built infrastructure and open-weight models. He also says the main long-term value creation will occur in the application layer, where deterministic enterprise workflows can translate AI into measurable productivity, revenue uplift, and workflow automation.

Notes

• Vista’s core thesis is that GenAI is re-rating enterprise software, not displacing it.
• The episode explicitly compares AI’s effect on software valuations to the cloud/SaaS re-rating in 2015-2016.
• The market still appears to be underappreciating the speed of AI-driven value creation in software.
• Incumbent software vendors have a contextual advantage because they understand how customers operate and what their data means.
• Trusted relationships are part of the moat, not just access to data.
• Breach argues incumbents are better positioned than startups built only on LLMs.
• The traditional Rule of 40 is being replaced by a higher bar closer to a Rule of 60 or 70.
• AI is expected to improve both revenue growth and resource efficiency at the same time.
• Vista says some portfolio companies already meet Rule-of-70 economics.
• Vista is underwriting a further 15 to 20 points of EBITDA margin improvement in some businesses.
• BCG and Bain are cited as estimating a $3 trillion to $4 trillion new addressable market for software in the agentic era.
• Management believes the near-term gains from AI are easiest to realize on the cost side.
• Vista portfolio companies are reportedly seeing 80% to 85% productivity gains in software engineering.
• Sales throughput is also improving, with one cited example of 25% to 30% higher sales efficiency.
• Customer support automation is already deflecting most tickets without humans in some cases.
• More than half of Vista’s portfolio has already launched agentic products.
• Vista describes an internal enablement layer called the Agentic factory.
• The Agentic factory pairs expert resources with major model and cloud providers.
• Duck Creek is presented as a concrete example of AI workflow automation in insurance quoting.
• Vista says these products are producing real revenue uplift and TAM expansion.
• The economic bottleneck in AI is shifting toward inference, not just model training.
• Breach says training-grade GPUs are often too expensive for routine inference workloads.
• Vista is investing off balance sheet in an inference-focused neocloud.
• The inference stack is described as lower power, lower cost, and air-cooled rather than water-cooled.
• Breach expects inference infrastructure to be disaggregated from training infrastructure.
• The AI stack is expected to bifurcate between large GPU training clusters and separate inference infrastructure.
• Open-source and open-weight models are expected to matter more for inference use cases.
• The goal is to minimize inference cost so AI value accrues to customers and the software company rather than frontier model vendors.
• Token cost is becoming a new cost center that technology companies must actively manage.
• Vista’s long-term value creation lens centers on operational optimization and cost structure management.
• Breach says the long tail of value creation in AI will be the application layer.
• The application layer is where AI is translated into enterprise value.
• Enterprise software’s stickiness comes from trusted workflows and data stewardship.
• Enterprise customers want deterministic outputs, not probabilistic model behavior, for critical workflows.
• Breach draws a sharp distinction between enterprise workflow requirements and frontier model probabilism.
• The distribution base of enterprise vendors is broad, with hundreds to tens of thousands of customers per portfolio company.
• Vista says learning can be propagated across its platform quickly because it controls many companies.
• That internal platform is designed to spread operational learnings across the portfolio.
• Vista says its companies are mission-critical and defensible before AI is layered in.
• Vista’s credit business is also incorporating AI, including internal agents.
• Breach says 95% of Vista portfolio companies have seen zero impact from native GenAI startups.
• Core software remains resilient on recurring revenue and retention metrics.
• Vista says its average company has 2.5 years of contracted revenue visibility.
• Breach frames data plus context as a deeper moat than data alone.
• Frontier models are mostly trained on public data, which leaves enterprise context underrepresented.
• The combination of enterprise-scale data and its context is described as a much deeper moat.