SaaStr 868: Software Isn't Dead. It's Gotten Harder with Scale...

Full Title

SaaStr 868: Software Isn't Dead. It's Gotten Harder with Scale Venture Partners' Rory O'Driscoll

Summary

This episode discusses the evolving landscape of software investment in the age of AI, arguing that while software isn't dead, it has become significantly more complex and challenging to navigate.

The discussion focuses on the massive capital expenditure in AI development versus its current revenue generation, the emergence of new investment models, and how traditional software companies must adapt or face obsolescence.

Key Points

• The current AI investment landscape is characterized by enormous capital expenditure ($688 billion in AI CapEx by hyperscalers in 2026) vastly exceeding current revenues ($110 billion), indicating a prolonged "invest mode" for the industry.
• The success of AI is underpinned by observable trends like scaling laws proving that increased compute leads to better results and the widespread user adoption demonstrated by the ChatGPT moment, signaling strong market demand.
• While foundation models like those from OpenAI and Anthropic are dominant, opportunities exist for companies building "harnesses" – the software stack that makes LLMs dependable systems – and for those creating defensible businesses through unique data, marketplaces, or specialized vertical solutions.
• Pre-2022 software companies face varying impacts from AI: some are insulated, some are augmented by AI features, and a significant portion, particularly "plain vanilla SaaS," are directly threatened and must adapt or risk extinction.
• The investment thesis has shifted from traditional SaaS growth metrics to understanding the capital intensity of compute, the business models of AI-driven companies, and the potential for new venture-backed software opportunities built on top of dominant foundation models.
• The rise of AI requires a new framework for building software applications, analogous to the LAMP stack of the past, where the "harness" components are crucial for integrating LLMs into reliable systems.
• Compute costs are highly variable across AI-related businesses, ranging from 70-80% for model companies to as low as 10% for application-focused AI companies, highlighting a shift in cost structures and value propositions.
• The question of "is software dead" is reframed by Rory O'Driscoll to focus on how economic value will be allocated between foundation model companies and newer software companies, and whether older SaaS models remain viable.

Conclusion

Software is not dead, but it has become significantly more complex and requires a new approach to investment and development.

Companies must focus on category conviction and build defensible businesses, whether through specialized solutions, unique data, or effective "harnesses" that integrate AI.

Investors need to adapt their strategies to account for the massive capital expenditure in AI and the evolving value chain, looking for opportunities that leverage AI effectively rather than those simply adding AI features to existing models.

Discussion Topics

• Given the immense capital flowing into AI, what are the most promising areas for venture-backed software companies to innovate and create defensible value?
• How should traditional SaaS companies evaluate their long-term viability in the face of AI advancements, and what strategic pivots are most critical for survival?
• With compute costs becoming a major factor, what are the key metrics and considerations for investors assessing the financial health and scalability of AI-native businesses?

Key Terms

SaaS

Software as a Service, a software distribution model that hosts applications and makes them available to customers over the internet.

LLM

Large Language Model, a type of artificial intelligence algorithm that uses deep learning techniques and massive amounts of data to understand, generate, and manipulate human language.

CapEx

Capital Expenditure, funds used by a company to acquire, upgrade, and maintain physical assets such as property, buildings, or equipment.

Hyper-scalers

Large cloud computing providers that can scale their services to millions of users and massive data loads, such as Amazon Web Services, Microsoft Azure, and Google Cloud.

Foundation Models

Large AI models trained on a vast dataset that can be adapted to a wide range of downstream tasks, such as OpenAI's GPT-3 or Anthropic's Claude.

LAMP Stack

A widely used open-source software stack used for hosting dynamic websites and web applications, consisting of Linux (operating system), Apache (web server), MySQL (database), and PHP (programming language).

Plain Vanilla SaaS

Standard, straightforward Software as a Service offerings that typically automate a specific workflow without significant customization or unique technological differentiation.

Timeline