Investing in AI and its emerging trends

From: redpointai

This article explores the dynamic landscape of AI investment, drawing insights from a conversation with Redpoint Ventures partners Scott Rainey, Patrick Chase, Alex Bard, and Jacob Efron. The discussion covers where value is accumulating, the interplay between startups and incumbents, the economic implications of massive hardware investments, and Redpoint’s investment approach in AI [00:00:09].

The Staggering Scale of the AI Market

The projected revenue for the NVIDIA data center division, which powers AI training and inference, is $177billionforthecurrentyear.Thisis5.5timeslargerthanInte{l}^{\prime }sclientcomputinggroup,whichgenerates$32 billion from powering personal computers [00:01:45]. This significant capital expenditure on NVIDIA chips implies an anticipated revenue generation of approximately $1.2trillionby2030and$1.5 trillion by 2032 in the AI landscape [00:02:26]. For context, the global enterprise software market, built over 50 years, is about $1.1 trillion [00:03:00].

Industry leaders like Marc Benoff, Bill Gates, Jeff Bezos, and Eric Schmidt express “incredible excitement and exuberance” for AI’s potential, seeing it as unlocking more incredible opportunities with every new rung of capability [00:03:55]. Investing in AI is viewed as a “strategic imperative” for companies, regardless of immediate ROI [00:04:16].

A key shift is from “software as a service” to “service as software,” where AI is not just making humans incrementally more efficient but is “actually doing the job of a human” [00:04:36]. Labor budgets are often an order of magnitude larger than historical software budgets, presenting a massive market for AI to penetrate [00:04:48]. For example, the customer service software market is roughly $35billion,whilethehumancostis$450 billion [00:04:58]. AI is also expanding into new markets historically underserved by traditional software due to small market size or unsophisticated users [00:05:22].

The AI Investment Landscape: Three Layers

The AI landscape can be understood in three layers:

1. Model Layer

This layer includes large language models (LLMs) and other “brains” that power AI applications [00:06:10].

• Value Accrual: The value of model companies increasingly lies in the products built on top of them [00:06:52]. A state-of-the-art model (e.g., 10 IQ points higher than open-source alternatives) provides a unique opportunity to build differentiated products [00:07:12].
• Cost and Commoditization: The cost of entering the state-of-the-art LLM game is prohibitively expensive [00:07:22]. However, models are becoming cheaper, with inference and training costs dropping approximately 10x per year, leading to better margins for application companies [00:08:21].
• Moats: Scale is not an enduring moat [00:08:36]. Model companies are seeking moats through distribution (moving up the stack with apps and agents) or specialization [00:08:53]. For example, DeepSeek’s emergence led to 80-90% cost reductions for some companies that switched from Anthropic within days due to low switching costs [00:09:28].
• Adjacent Categories: Beyond LLMs, areas like robotics (requiring different data to enable real-world action), biology, and material sciences are seen as interesting for future investment [00:07:35].

2. Infrastructure Layer

This layer comprises the “picks and shovels” that bridge models and application vendors, used by developers to build AI applications [00:06:19].

• Initial Expectations vs. Reality: Redpoint initially expected significant opportunities here, similar to the cloud wave (AWS, tooling), but investment has been slow [00:10:33].
• Challenges:
  • The model layer is changing rapidly, causing building patterns to shift quickly [00:10:59].
  • Early AI adopters focused on “use case discovery,” often using powerful, branded models like OpenAI and Anthropic, rather than open-source or requiring extensive new infrastructure [00:11:10].
• Emerging Opportunities: This year (2024) is becoming more interesting for AI infrastructure as agents emerge, creating common patterns for web access and tool use [00:11:37].

3. Application Layer

This layer involves horizontal and vertical SaaS solutions leveraging AI to deliver unique capabilities and replace services with software [00:06:28].

• Reinvigorated Application Space: After a period where the application space “got long in the tooth” (2015-2018), AI has caused an “explosion” in companies building applications [00:12:09].
• Business Model Shift: Similar to the cloud wave disrupting incumbents with a new software delivery mechanism, AI brings a new business model: charging for “work” rather than for a “seat” [00:13:34]. Labor budgets are an order of magnitude larger than historical software budgets, providing significant opportunities [00:04:48].
• Horizontal Applications: Opportunities exist to disrupt large horizontal players like HubSpot or Salesforce.com by building AI-native solutions with a business model advantage [00:13:52]. Speed is a key attack vector against incumbents, who may have more lawyers than engineers and face challenges in adopting fast-moving underlying models and new business models [00:14:30].
• Vertical Applications: There’s a “Cambrian explosion” in vertical AI SaaS businesses, with over 500 companies started in the last few years [00:15:19]. These target verticals that historically lacked compelling SaaS solutions [00:15:26].
  • Investment Criteria: When evaluating vertical AI startups, key questions include:
    1. Is there an effective “wedge” into the market, demonstrating meaningful traction and viral end-user love? [00:16:09]
    2. How much more can the company do beyond replacing a single FTE? Focus is on large industries like healthcare, law, and finance where the prize is significant [00:16:33].
    3. How much does quality matter? In areas where quality is critical (e.g., regulated industries like healthcare or law), it’s less likely to be a “race to the bottom” on price [00:17:01]. Markets where customers are already willing to trade quality for lower cost (e.g., outsourced BPOs) are less attractive [00:19:39].

Challenges and Opportunities in AI Integration

Competition and Fragmentation

The rapid proliferation of AI companies, particularly in vertical markets, raises concerns about market crowding and a potential “carnage” of similar offerings [00:18:00]. Many may experience initial growth but then “top out” [00:18:13]. The question remains whether these markets will see clear winners (winner-take-most) or remain incredibly fragmented because value is created by underlying LLMs accessible to many [00:20:45].

Incumbents vs. Startups

Incumbents like Salesforce possess significant advantages:

• Customer Relationships: They own existing workflows and have proprietary data for fine-tuning models [00:29:51].
• Strategic Imperative: All incumbents are driven to leverage AI to remain competitive, and public market valuations favor companies with an AI story [00:30:04].

However, incumbents face challenges:

• Legacy Systems: They sit on old databases, infrastructure, and user experiences (UX) that are difficult to change with AI [00:31:54]. Integrating AI into workflows developed 20+ years ago can be “heart surgery” [00:32:05].
• Business Model Inertia: Large companies struggle to shift from established SaaS pricing (seat-based) to new models (work-based) without abandoning billions in existing revenue [00:14:42].

Startups’ advantages include:

• Speed: They can move quickly, innovate, and outflank competitors [00:21:26].
• Native AI Design: They can build from scratch with AI at the core, designing new workflows that fundamentally change how tasks are done, unlike incumbents integrating AI into existing, outdated processes [00:33:00]. The “more the workflow changes, the better opportunity there will be for startups” [00:33:30].

The BPO Loser

A significant amount of budget for AI applications is expected to come from business process outsourcing (BPO) and traditional services businesses, as both startups and incumbents will “eat into that” market [00:34:55].

High Valuations and False Positives

AI valuations are high, with larger round sizes and substantially higher pre-money valuations for Series A companies [00:35:30]. This is partly rationalized by the belief in larger addressable markets, access to labor budgets, and unprecedented growth speeds [00:35:58]. Additionally, AI-native companies can build products more efficiently, potentially achieving hundreds of millions in revenue and billions in enterprise value with significantly fewer employees (e.g., 20-40 employees), reducing the need for future capital raises [00:36:40].

However, these high valuations come with risks:

• Immature Companies: Companies are raising significant rounds with high revenue traction but may lack corporate maturity (e.g., systems, people, processes), meaning revenue can be a “misleading indicator” [00:40:11].
• False Positives: Companies can spike quickly, making it hard to distinguish enduring businesses from those with transient success [00:37:25]. This can be exacerbated by “AI tourists” who experiment with products based on experimental budgets, not core business line needs [00:27:29].

Investment Strategy

Redpoint’s approach to investing in AI emphasizes:

• Founder Market Fit: Prioritizing founders with collective experiences that provide unique insights into long-term market problems, rather than just “young, great builders” who move fast but lack market experience [00:22:31].
• Product Depth: Seeking significant product depth that would be challenging for other teams to replicate, focusing on solutions that are “80% workflow, 20% model” rather than the reverse [00:23:01].
• First-Mover Advantage: The speed at which a company can become synonymous with a category (e.g., in healthcare AI) matters significantly, leading to customer access, partnerships, and capital [00:23:40]. However, it’s acknowledged that sometimes being “last” can be more advantageous (e.g., Google not being the first search engine) [00:25:10].
• Evolving Moats: Traditional moats like proprietary data assets may not matter as much. Instead, the moat often lies in “the thousand little things” that make a SaaS product better, such as UX, product breadth, and user experience [00:24:38].
• Domain vs. AI Expertise: While AI expertise is important for understanding model evolution (to avoid “extinction-level events”), deep domain expertise is crucial for understanding end-user problems and navigating regulated industries [00:26:00].
• Durable Revenue: Differentiating between “experimental budget” and “business line budget” by focusing on user engagement and usage metrics [00:28:16]. True traction means “meaningful traction” that can translate into a large, standalone company [00:16:33].
• Massive Tail Opportunities: Prioritizing end markets with a “massive tail opportunity” where the prize is large and can sustain competition, rather than smaller vertical markets that might become overcrowded [00:38:01].
• Discipline and Due Diligence: Maintaining “incredible discipline” when evaluating high valuations and fast-growing, but potentially immature, companies [00:39:37]. This involves extensive “first principle diligence work” and focusing on “core fundamentals” rather than hype [00:42:56].
• Co-processing Deals: Redpoint’s early and Omega funds are increasingly co-processing deals for companies raising larger first institutional rounds with significant traction [00:40:50].

The impact of AI advancements on business models means a $50 million AI SaaS business is very different from a traditional SaaS business, especially in terms of company maturity and the underlying work required to achieve that revenue [00:42:05]. The firm remains focused on identifying those opportunities that can build truly large, enduring businesses.