Investment strategies in the AI landscape

From: redpointai

Redpoint Ventures partners Scott Rainey, Patrick Chase, Alex Bard, and Jacob Efron discussed their approach to investing in the AI landscape at their annual meeting with limited partners [00:00:09]. The discussion covered where value is accruing, the dynamic between startups and incumbents, the impact of massive hardware investments, and the firm’s overall strategy in AI [00:00:12].

Market Overview and Economic Impact

The projected revenue for the NVIDIA data center division, which powers AI training and inference, is $177 bi ll i o n, 5.5 t im es t h es i zeo f I n t e l^{'} s p erso na l co m p u t in gg ro u p re v e n u e < a c l a ss = " y t - t im es t am p " d a t a - t = "01 : 47" > [01 : 47] < / a > . T o g e n er a t e a re a so nab l e [[t h e_{e} co n o mi c_{i} m p a c t_{o} f_{a} i_{i} n v es t m e n t s_{a} n d_{r} o i_{e} x p ec t a t i o n s ∣ RO I]] f ro m t hi s$ 177 billion in capital expenditure on Nvidia chips, the companies spending this money must anticipate revenues of approximately $1.2 t r i ll i o nb y 2030 an d$ 1.5 trillion by 2032 [02:06]. This projected $1.5 t r i ll i o nma r k e t f or A I b y 2032, b u i lt o v er ab o u t 10 ye a rs, co m p a res t o t h ee n t i re g l o ba l e n t er p r i seso f tw a re ma r k e t, w hi c hi s a pp ro x ima t e l y$ 1.1 trillion and was built over 50 years [02:47].

Alex Bard believes this projection might be understated, highlighting that making these investments is a strategic imperative for companies, regardless of immediate ROI, to avoid being left behind [04:11]. AI’s shift from making humans incrementally more efficient to “service as software” means it’s doing the job of a human [04:36]. Labor budgets are often an order of magnitude larger than traditional software budgets, presenting significantly larger addressable markets for AI [04:46]. For example, the customer service software market is about $35 bi ll i o n, w hi l e t h e h u man cos t i s$ 450 billion [04:58]. AI is also expanding into markets historically underserved by software due to size or user sophistication [05:22].

Investment Focus Across AI Layers

Redpoint’s investment strategy considers three main layers of the AI landscape:

1. Model Layer

The value of model companies is increasingly tied to the products built on top of them [06:49]. Building a cutting-edge foundation model enables the creation of unique products that require state-of-the-art models [07:03]. While the cost to enter the state-of-the-art LLM game is prohibitively expensive, adjacent model categories, such as robotics, biology, and material sciences, are of interest due to different data requirements [07:22].

The DeepSeek announcement highlighted that models are becoming cheaper, which benefits application companies by improving margin structures, as inference and training costs drop approximately 10x per year [08:13]. This also showed that scale is not an enduring moat for model companies; they will likely build moats through distribution (moving up the stack with apps and agents) or specialization [08:36]. Switching costs between models are very low, as companies can switch quickly, for example, from Anthropic to DeepSeek, achieving 80-90% cost reductions [09:36].

2. Infrastructure Layer

Historically, Redpoint expected significant opportunities in the infrastructure layer, similar to the cloud wave, which provided compute, storage, and networking infrastructure as a service, along with crucial tooling [10:06]. However, this area has been slow for two main reasons:

The model layer is changing rapidly, causing builder patterns to shift at the same speed [10:59].
Early AI adopters focused on use case discovery, leveraging the most powerful and recognized models like OpenAI and Anthropic [11:10]. Opportunities are emerging as agents come forth, which could lead to common patterns in how agents access the web and use tools [11:37].

3. Application Tier

The advent of AI has caused an explosion in companies building applications to deliver differentiated experiences [12:19]. A key differentiator for AI applications is the shift in business model from charging for a seat to charging for work done [13:34]. This creates a moment of disruption similar to the early days of SaaS [13:42].

Opportunities exist in:

Horizontal Applications: AI-native solutions competing with established players like HubSpot or Salesforce, benefiting from the new business model [13:52].
Vertical Markets: A “Cambrian explosion” of hundreds of new vertical AI SaaS businesses are emerging in markets previously underserved by compelling SaaS solutions [15:15].

Competing with Incumbents

Incumbents like Salesforce possess significant advantages including customer relationships, existing workflows, and proprietary data for fine-tuning models [29:47]. However, startups can leverage speed as an attack vector because large companies often have more lawyers than engineers, hindering their ability to rapidly adopt fast-moving underlying models and embrace new business models [31:30]. Incumbents struggle to integrate AI into old databases, infrastructure, and UX, making them vulnerable where AI fundamentally changes workflows [31:51]. For example, AI can eliminate complex logic trees in systems like customer service routing engines by using more data for better decisions [32:51]. The greater the change in workflow, the better the opportunity for startups [33:30]. Markets where the workflow does not change significantly (e.g., AI-native Notion or PowerPoint) are less attractive for startups as incumbents can deliver these features [34:16].

The primary losers in this shift are traditional BPOs and legacy services businesses, as both startups and incumbents are consuming their market share [34:55].

Navigating AI Valuations

AI valuations are high, with Series A rounds seeing incrementally larger sizes and substantially higher pre-money valuations [35:30]. This is partly due to the belief that AI markets will be larger, accessing labor markets and new verticals [35:58]. AI companies are also growing at an unbelievable pace and can operate much more efficiently, potentially reaching hundreds of millions in revenue with only 20-40 employees [36:15]. This capital efficiency means they might need to raise less future capital, reducing dilution for early investors [36:53].

However, high valuations come with significant challenges:

False Positives: Companies can spike quickly, making it hard to distinguish enduring businesses from short-term successes [37:25].
Increased Risk: The rapid pace of market change adds inherent risk to these investments [37:46].
Fund Construction: Higher prices and faster follow-on rounds impact fund size and ownership [39:26].
Corporate Maturity vs. Revenue Maturity: Companies can achieve significant revenue quickly (e.g., $8 million ARR in 8 months with 10 people) but lack the corporate maturity, systems, and processes of a traditional business at that revenue scale [40:11]. This means revenue, traditionally a lagging indicator, can be a leading or even misleading indicator in AI [41:12].

Investment Discipline

To navigate these challenges, Redpoint maintains discipline:

Focus on “Tail Opportunity”: Prioritizing end markets with massive potential if successful, rather than smaller, crowded verticals [38:00].
Rigorous Due Diligence: Not being swayed by large numbers alone, but focusing on core fundamentals and first-principle diligence [42:42].
Co-processing Deals: The firm is increasingly co-processing deals between its early-stage and growth funds, as many companies raising larger first institutional rounds blur traditional stage definitions [40:50].

Key Investment Considerations

First Mover Advantage: Speed in achieving market prominence is crucial, as being synonymous with a category can quickly attract customers, partnerships, and capital [23:42].
Velocity: The ability to move and innovate quickly is paramount [24:20].
Moats in AI Apps: While grand data asset moats are rare, competitive advantages often derive from “a thousand little things” like user experience, product breadth, and overall user experience, similar to traditional SaaS [24:38].
Founder Market Fit: At early stages, prioritizing founders with deep experience and unique insight into the long-term market problem over just “great builders” who move fast but lack market experience [22:31].
Product Depth: Looking for solutions with significant product depth that are challenging to replicate, often characterized by a higher ratio of workflow to model (e.g., 80% workflow, 20% model) [23:01].
Domain Expertise vs. AI Expertise: While being technical enough to understand model trends is important, deep domain expertise is increasingly valued for understanding end-user problems [25:56].
Durable Revenue: Distinguishing between experimental budget (companies willing to try new AI products) and securing long-term business line budgets [27:58]. User engagement and usage are key metrics for identifying durable revenue [28:50].
Quality Matters: Focusing on industries and use cases where quality is paramount (e.g., regulated industries like healthcare or law), as these markets are less susceptible to a “race to the bottom” on price for 80%-good solutions [17:01].

In conclusion, the AI landscape presents massive opportunities and challenges, requiring venture capitalists to be highly disciplined, analytical, and focused on long-term business fundamentals amidst rapid growth and high valuations [42:55].

Tubegraph

Explorer

Table of Contents