Money Always Moves Up the Tech Stack

DeepSeek reminds me of a "side project" from the 1990s called Linux. Both are open-source, both were initially dismissed by giants like Microsoft, and both reveal a timeless truth: value in technology accrues to those who build ecosystems, not just foundational tools. Today, Linux underpins 90% of the cloud, and companies like Red Hat—built on top of open-source code—turned it into a $34B business by selling support, not software. DeepSeek is following the same playbook. It allows anyone to run, tweak, or embed AI locally. Worried about data security risks, like sensitive information being routed through foreign servers? Just cut your internet connection cable—DeepSeek still works. This underscores what tech insiders already know: the real market value lies not in the model itself, but in the vertical solutions and workflows companies build around it.

Wall Street’s priorities are shifting. Investors once chased GPU manufacturers like NVIDIA but now focus on inference costs and developer ecosystems. When Meta open-sourced Llama 2, its valuation surged not because it “won” AI, but because it positioned itself as a platform for others to build on—mirroring how Android captured value in mobile. Hugging Face ($4.5B) follows the same playbook: monetizing community, not hardware.

This mirrors the cloud’s evolution. IBM dominated mainframes in the 1980s; Dell ruled PCs in the 1990s. By the 2000s, value migrated to platforms like AWS and apps like Salesforce. Today, Snowflake and Databricks extract even more value by layering efficiency atop those clouds. The lesson? Value always moves up the stack, from raw materials to refined systems—just as steam engines birthed railroads, factories, and global trade.

For enterprises, the opportunity lies in vertical solutions. Silicon Valley startups aren’t building models—they’re repurposing open-source AI into industry-specific tools for healthcare compliance, supply chains, or customer service. The winners won’t own algorithms; they’ll own workflows. Instead of asking whether “R1” or “O1” is the better model, ask: How quickly can your company build AI-powered products that extract operational value on top of these tools? If “R2” or “O2” launches tomorrow, how easily could you adapt? And the next version after that? This isn’t new. Since the dawn of the industrial age, true competitive advantage hasn’t been about owning the technology—it’s about how fast and efficiently you can build on top of it.