Episode 823 | Hot Take Tuesday: Is A.I. Killing B2B SaaS?, ChatGPT Ads, OpenClaw

What It Covers

Rob Walling, Tracy Osborne, and Einar Vollset debate whether AI is killing B2B SaaS, analyze ChatGPT's move into advertising, and evaluate OpenClaw's potential. The episode covers M&A market dynamics, AI model improvement rates, and early-mover advertising opportunities on AI platforms for SaaS founders.

Key Questions Answered

• •B2B SaaS survival: Subscription software is not dying — it is being renamed. SaaS previously cycled through labels including "downloadable software," "ASPs," and "cloud software." Founders who treat AI agents as the next delivery mechanism rather than an existential threat position themselves to adapt, just as early SaaS founders adapted from desktop software distribution.
• •M&A deal structure threshold: Roughly 90% of B2B SaaS acquisitions handled by Discretion Capital are stock transactions, not asset purchases. Stock deals become standard around $1M ARR and a $5M exit price. Founders pursuing QSBS tax benefits — which require stock sales to qualify — should structure as C-corps early, since asset purchases disqualify them from those savings.
• •Private equity dominates SaaS acquisitions: 70% of B2B SaaS acquisitions in the $2M–$20M ARR range involve private equity or PE-owned buyers. Founders who misunderstand this buyer landscape risk selling for roughly one-third of actual market value. Discretion Capital's free guide at discretioncapital.com/guide breaks down buyer types and deal mechanics chapter by chapter.
• •AI advertising early-mover window: ChatGPT ads represent the same early-stage opportunity that Google AdWords offered in 2005–2006 and Facebook ads offered circa 2010–2012, when clicks cost $0.05–$0.25. SaaS founders should pursue early access to ChatGPT's ad platform now, before pricing becomes competitive and before established consultants and playbooks dominate the channel.
• •AI model improvement plateau: The rate of capability improvement in LLMs is slowing due to two constraints: training data exhaustion and exponentially increasing compute costs. Each successive model upgrade requires roughly 10x the compute budget of the prior version, making continuous step-change improvements mathematically unsustainable without a new architectural breakthrough beyond scaling existing approaches.