AI is a 5 Layer Cake: What the Real Estate Industry Needs to Know About Compute and Intelligence

By Anna with Oppy

As the Oppy Inc. team continues our coverage from the ground at HumanX 2026 in San Francisco, one of the most highly anticipated sessions was the "AI is a 5 Layer Cake" panel. Featuring heavyweights like Bryan Catanzaro (VP of Applied Deep Learning Research at NVIDIA), Denis Yarats (CTO of Perplexity), and Lin Qiao (CEO of Fireworks AI), the discussion offered a masterclass in how the AI stack is evolving and what it means for enterprise adoption.

For real estate professionals, understanding this "5 Layer Cake"—Compute, Models, Infrastructure, Tooling, and Applications—is no longer optional. It is the blueprint for how brokerages will build their competitive advantage over the next decade.

The Compute Layer and the 4 Scaling Laws

The foundation of the AI cake is compute, and NVIDIA's Bryan Catanzaro provided a sobering look at the sheer scale of infrastructure required to train modern models. He outlined the four scaling laws that dictate AI progress: compute, data, model size, and algorithmic efficiency.

"We are seeing a 40x increase in compute requirements for frontier models every 18 months. This is fundamentally different from Moore's Law. We are building supercomputers the size of small cities just to train the next generation of intelligence."

— Bryan Catanzaro, VP of Applied Deep Learning Research, NVIDIA

Catanzaro also highlighted NVIDIA's Nemotron family of models, emphasizing that while frontier models get the headlines, smaller, highly efficient models are often better suited for specific enterprise tasks. For a real estate brokerage, this means you don't need a trillion-parameter model to analyze a local MLS feed; you need a specialized, efficient model running on optimized compute.

The Infrastructure Bottleneck

Lin Qiao, CEO of Fireworks AI, shifted the conversation to the infrastructure layer, pointing out that the cost of inference (running the models) has plummeted, but the complexity of deploying them has skyrocketed.

"The cost of inference for a GPT-4 class model has dropped from $30 per million tokens to under $1 in just 18 months. The bottleneck is no longer the cost of the model; it's the infrastructure required to serve it reliably at scale."

— Lin Qiao, CEO, Fireworks AI

Qiao noted that developers are wasting massive amounts of time wrestling with infrastructure rather than building applications.

"We are seeing a massive S-curve in adoption, but the friction is in the deployment. If you are an enterprise, you shouldn't be building your own inference engine. You should be focusing on your data and your user experience."

This is a critical insight for real estate leaders. Brokerages should not be trying to build foundational AI infrastructure. Instead, they should leverage platforms that abstract away this complexity, allowing them to focus on their core business.

Private Data: The Ultimate Real Estate Moat

Perhaps the most actionable takeaway for the real estate industry came from Denis Yarats, CTO of Perplexity. As the leader of an AI-native search product, Yarats understands that when everyone has access to the same foundational models, the only true differentiator is proprietary data.

"If you are building an AI application today, and your only advantage is a prompt wrapper around a public model, you have no moat. The true moat is your private data. It's the data that isn't on the public internet."

— Denis Yarats, CTO, Perplexity

In real estate, this "private data" is the lifeblood of a successful brokerage. It's the off-market pocket listings, the historical negotiation tactics that won specific neighborhoods, the nuanced understanding of local zoning laws, and the deep CRM histories of past clients.

When a brokerage combines this proprietary data with agentic AI, they create an insurmountable competitive advantage. An AI agent that knows your specific market history can draft better offers, identify better investment opportunities, and provide hyper-personalized client service that a generic public model simply cannot match.

Building the Application Layer with Oppy

The panel concluded that the application layer—the top of the 5 Layer Cake—is where the most value will be created over the next five years.

"We are going to see millions of specialized models and agents deployed across every industry. The winners will be the ones who seamlessly integrate these agents into the daily workflows of their employees."

— Lin Qiao, CEO, Fireworks AI

This is exactly why Oppy Inc. exists. We are building the application layer for entrepreneurs and real estate professionals. Oppy is a fully AI-native platform that allows you to launch and manage AI employees (oppies) with access to over 60 business tools. We provide the conversational utility that abstracts away the complexity of the bottom four layers of the cake, allowing you to focus purely on the application of intelligence to your business.

Connect with Oppy at HumanX

The entire Oppy team is here in San Francisco for HumanX 2026, and we are actively raising our seed round. We are looking to partner with investors and real estate visionaries who understand that the future belongs to those who control the application layer and their private data.

If you are at the conference, we invite you to connect with our Founder and CEO, Alex Gustafson. Let's discuss how Oppy can help you build your data moat and deploy AI agents that transform your brokerage operations.

Stay tuned for our final dispatch from HumanX 2026.