From: redpointai
This article summarizes discussions from an annual meeting of Redpoint Ventures partners, including Scott Rainey, Patrick Chase, Alex Bard, and Jacob Efron, focusing on their approach to investing in the evolving AI landscape [00:00:09]. The conversation delves into where value is accruing, how startups compete with incumbents, the impact of massive hardware investments, and Redpoint’s overall investment strategy in AI [00:00:12].
Market Overview and Investment Landscape
The projected revenue for the AI landscape is staggering, anticipated to reach 1.5 trillion by 2032 [00:02:30]. This represents a market being built over roughly 10 years, compared to the world enterprise software market, which is about $1.1 trillion built over 50 years [00:02:50].
Many major technology leaders, including Marc Benoff, Bill Gates, Jeff Bezos, and Eric Schmidt, express incredible excitement for AI’s potential [00:03:39].
Investing in AI is seen as a strategic imperative for companies, regardless of immediate ROI, to avoid being left behind [00:04:16]. The shift is from “Software as a Service” to “Service as Software,” where AI performs human jobs, potentially unlocking significantly larger budgets [00:04:36]. For example, the customer service software market is about 450 billion, indicating a massive opportunity for AI [00:04:58]. AI is also expanding into new markets and those historically under-penetrated by software due to size or user sophistication [00:05:21].
AI Landscape Layers and Investment Focus
The AI landscape is conceptualized into three layers:
- Model Layer: The “brains” (LLMs and other models) powering applications [00:06:08].
- Infrastructure Layer: “Picks and shovels” bridging models and applications, used by developers [00:06:19].
- Application Layer: Horizontal and vertical SaaS solutions leveraging AI capabilities [00:06:28].
Model Layer
The value of model companies is increasingly tied to the products built on top of them [00:06:52]. Building a cutting-edge foundation model allows companies to create unique products that require state-of-the-art models [00:07:03]. However, the cost of entering the state-of-the-art LLM game is prohibitively expensive [00:07:20].
Redpoint is interested in adjacent model categories like robotics (e.g., Physical Intelligence), biology, and material sciences, which require different datasets [00:07:35].
The Deepseek announcement highlighted two key trends:
- Models are becoming cheaper (costs dropping about 10x a year for inference and training) [00:08:13]. This is beneficial for application companies, leading to better margin structures [00:08:27].
- Scale is not an enduring moat for model companies [00:08:36]. Model companies will likely build their moats through distribution (moving up the stack with apps and agents, like OpenAI) or specialization [00:08:53].
Switching costs between models are very low, as seen with Redpoint’s portfolio companies quickly moving from Anthropic to Deepseek, achieving 80-90% cost reductions within days [00:09:21].
Infrastructure Layer
Redpoint historically invested heavily in the infrastructure layer during the cloud wave [00:10:07]. However, this layer has been slower for AI for two main reasons:
- The model layer is changing so rapidly that builder patterns also change frequently [00:10:59].
- Early AI adopters were in “use case discovery mode,” primarily using powerful, branded models like OpenAI and Anthropic [00:11:10].
While data centers and the inference market (e.g., Modal, Livekit) have seen investment, the current year is seen as interesting for infrastructure as common patterns emerge with the rise of AI agents [00:11:37].
Application Layer
The advent of AI has led to an explosion in application companies building differentiated experiences [00:12:19]. The last five years of innovation in large language models and infrastructure are now poised to be deployed at scale [00:12:50].
A key factor enabling disruption in the application layer is a fundamental business model change: charging for work rather than for seats [00:13:34]. This creates a moment of disruption similar to when cloud computing enabled new startups to challenge incumbents [00:13:21].
Horizontal Applications
There are significant opportunities in horizontal applications. Companies like Adio (targeting HubSpot/Salesforce.com) and Level Path (targeting Coupa) are examples of AI-native solutions with this business model advantage [00:13:52]. The prize is large if these startups can even capture a small percentage of incumbents’ revenue [00:14:14].
The challenge lies in competing with tech giants who are well-capitalized and also adopting AI [00:14:24]. The primary attack vector for startups is speed in adopting fast-moving models and embracing the new business model disruption, as incumbents face legacy issues and cannot easily abandon existing SaaS pricing models [00:14:30].
Vertical Applications
There has been a “Cambrian explosion” of vertical AI SaaS businesses, with over 500 companies started in the last few years, targeting industries previously underserved by compelling SaaS solutions [00:15:19].
When evaluating these vertical markets, Redpoint asks:
- Is there an effective wedge into the market? They look for meaningful, viral product-market fit rather than just experimental revenue [00:16:09].
- How much more can these companies do? Beyond replacing a single FTE, the focus is on building a large standalone company, especially in large industries like healthcare, law, and finance where the prize is significant [00:16:36].
- How much does quality matter? In competitive markets where price undercutting is common, quality differentiation is crucial to avoid a “race to the bottom” [00:17:01]. Regulated industries (e.g., healthcare, corporate law) are good examples where 100% quality is paramount [00:19:00].
The potential to unlock labor budgets in these verticals means markets traditionally considered too small for venture-scale investment might become attractive [00:17:25]. However, a lot of attrition and consolidation are expected, similar to the early days of e-commerce [00:20:12]. It’s unclear if these markets will be “winner take most” or highly fragmented, given that much of the value is created by underlying LLMs that many companies can leverage [00:20:44].
Investment Strategy & Challenges
Founder Profile
At the earliest stages, the focus is heavily on founders [00:22:13]. Two types are observed:
- Young, great builders: Move fast, may have first-mover advantage, but lack market experience [00:22:18].
- Founders with “founder market fit”: Collective experiences provide unique insight into long-term market problems (e.g., Motif, co-founded by former Autodesk co-CEO, building an AI-native Autodesk 2.0) [00:22:31]. Redpoint typically indexes towards the latter [00:22:42].
The preference is for companies with significant product depth that would be challenging for others to replicate, where the solution is primarily workflow-driven (e.g., 80% workflow, 20% model) rather than model-driven [00:23:01].
Success Characteristics of AI Businesses
Common characteristics seen in successful AI businesses include:
- First-mover advantage: Companies can become synonymous with a category rapidly (6-9 months), gaining early customer access and partnerships [00:23:42].
- Velocity: The ability to move and innovate quickly is crucial [00:24:20].
- Moat: While unique data assets were once thought to be key, the moat often comes down to “a thousand little things” like user experience (UX), product breadth, and overall user experience, similar to traditional SaaS [00:24:41].
AI Expertise vs. Domain Expertise
Initially, there was a strong desire for founding teams with deep AI research expertise (e.g., former DeepMind or OpenAI researchers) [00:25:51]. However, this is seen as less critical now [00:25:56]. Technical understanding is important to stay aware of the rapid pace of model changes and potential “extinction-level events” if a company cannot adapt quickly to new models [00:26:02].
Domain expertise is important for understanding end-user problems, but the market has become more democratic, allowing new entrants to engage with customers if they have a good AI product [00:26:21]. Ideally, a combination of both AI and domain expertise is preferred [00:26:51].
Lessons from Stalled Businesses
Companies that experience early success but then hit a wall often share traits:
- They were easy to replicate and rip out due to a lack of integrated solution or “gravity” [00:27:20].
- They capitalized on “AI tourists” or experimental budgets, rather than securing durable business line budgets [00:27:29]. Buyers are often mandated to explore AI, leading to initial engagement but not necessarily long-term adoption [00:28:07].
- Key metrics for durability include user engagement and usage [00:28:50].
Early-stage AI companies are reaching significant revenue milestones (e.g., 0 to $3-10 million ARR) much faster than traditional SaaS companies [00:29:14]. However, this rapid revenue growth can mask a lack of corporate maturity (e.g., systems, people, process) [00:30:28]. Revenue can be a misleading indicator, creating “early stage” companies with meaningful scale but lacking the foundational elements of a well-built business [00:41:18].
Valuations and Fund Strategy
AI valuations are high, with substantially larger pre-money valuations and round sizes, a trend seen across Series A, B, and C rounds [00:35:36]. This is partly due to the belief that AI markets will be significantly larger, accessing labor markets and previously untapped verticals [00:36:00]. The speed of growth for successful AI companies also justifies higher valuations [00:36:15].
AI companies are expected to build products more efficiently, potentially reaching hundreds of millions in revenue with a small number of employees (e.g., 20-40) [00:36:40]. This implies they may need to raise less future capital, leading to less dilution for early investors [00:37:06].
The challenge for VCs is picking the right companies, as the market is changing fast and comes with inherent risks [00:37:44]. There’s a need to focus on “tail opportunities” – massive end markets where success could lead to enormous returns [00:38:01]. Firms must be disciplined, even if it means missing opportunities, to avoid false positives from companies that spike quickly but lack endurance [00:37:31].
The increased valuations and faster fundraising cycles (companies coming back for new rounds sooner) imply that initial investments are even more expensive [00:38:48]. This impacts fund construction and the need for rigorous first-principles diligence to differentiate durable businesses from hype [00:39:31]. Redpoint is increasingly co-processing deals between its early and Omega funds to manage this dynamic [00:40:50].
Ultimately, building a $50 million AI SaaS business is different from a traditional SaaS business of the same size, with less maturity in the company’s structure [00:42:05]. This requires investors to verify that revenue is indicative of long-term sustainability and not just a temporary surge [00:42:47].