From: redpointai
The landscape of Artificial Intelligence (AI) is rapidly evolving, driven by significant investments in hardware and the development of sophisticated models and infrastructure. Projections suggest the AI landscape could generate revenues of 1.5 trillion by 2032, a market built over approximately 10 years, which is staggering when compared to the world enterprise software market of $1.1 trillion built over 50 years [02:30:00]. This growth signals immense excitement for AI and the unlocking of new opportunities with every advancement in AI capability [04:00:00].
The AI Landscape: A Three-Layered Model
The AI landscape can be understood through a simple three-layered model [06:06:00]:
- Model Layer: The “brains” like Large Language Models (LLMs) that power AI capabilities [06:10:00].
- Infrastructure Layer: The “picks and shovels” that bridge models and application vendors, used by developers to build AI applications [06:19:00].
- Application Tier: Horizontal and vertical SaaS solutions delivering unique AI capabilities, often replacing services with software [06:28:00].
The Model Layer: From Foundation to Specialization
The value of model companies is increasingly tied to the products built on top of them [06:52:00]. Building a cutting-edge foundation model allows companies to create unique products that require state-of-the-art models [07:03:00].
However, the cost to enter the state-of-the-art LLM game is prohibitively expensive [07:20:00]. Investment focus is shifting to adjacent model categories that require different data, such as robotics (e.g., physical intelligence), biology, and material sciences [07:29:00].
Trends in Model Development
- Cost Reduction: Models are becoming cheaper. Inference and training costs are dropping approximately 10x per year [08:13:00]. This benefits application companies by improving margin structures [08:27:00].
- Commoditization: The Deepseek announcement highlighted that scale is not an enduring moat [08:36:00]. Having the biggest GPU cluster doesn’t guarantee the best or most unique model [08:40:00].
- Building Moats: Model companies are seeking to build moats through:
- Distribution: Like OpenAI moving up the stack by launching apps and agentic products [08:53:00].
- Specialization: Focusing on specific domains like robotics [09:02:00].
- Low Switching Costs: It’s easy for companies to switch between models. A new model can be plugged in with similar performance without moving much code, unlike cloud migrations [09:36:00]. This was demonstrated when many portfolio companies switched from Anthropic to Deepseek, seeing 80-90% cost reductions [09:21:00].
The Infrastructure Layer: Slower but Emerging Opportunities
Historically, investment in the infrastructure layer during the AI wave has been slower than anticipated [10:39:00].
Challenges in AI Infrastructure Investment
- Rapid Model Evolution: The model layer changes so quickly that the patterns and tools used by builders change at the same speed [10:59:00]. Every three months, a new model or approach emerges [11:03:00].
- Early Use Case Discovery: In the early wave of AI, people focused on use case discovery, often leveraging the most powerful and recognized brand-name models (like OpenAI and Anthropic), which slowed the adoption of new infrastructure tools [11:11:00].
Emerging Opportunities
Despite initial slowness, there are areas of interest and growing opportunity:
- Data Centers and the Inference Market are significant areas of investment [11:27:00].
- Agent Emergence: The current year (referring to 2024 at time of discussion) is seen as particularly interesting for infrastructure, as agents begin to emerge. This is expected to lead to common patterns in how agents access the web and use tools, creating new infrastructure needs [11:37:00].
The evolution of AI models, especially the commoditization of models and the reduction in training and inference costs, directly impacts the viability and profitability of the infrastructure and application layers. As models become more accessible and affordable, the focus shifts to robust and adaptive infrastructure that can support diverse and rapidly evolving AI applications, driving future trends in AI hardware and software development.