The CPU dominated computing for 40 years. Then the GPU displaced it for AI workloads and Nvidia became worth more than Intel by a factor of ten. Andrej Karpathy says it is happening again — and whoever owns the model wins the third era.
In 1980, floating point math was an add-on. Engineers bolted an Intel 8087 coprocessor onto an 8086 to do real calculations. It was specialist hardware for specialist work. By 1989, the 486DX had swallowed the math chip whole. The standalone FPU business disappeared inside a decade, absorbed into the main processor it once augmented.
In 1996, graphics was an add-on. 3dfx shipped the Voodoo card to accelerate one task: rendering 3D games. Nvidia repurposed the GPU for general parallel compute with CUDA in 2007. Then deep learning landed, and the GPU became the substrate for the entire AI economy. The accelerator card became worth more than the computer it plugged into. Intel, which owned the CPU it was supposed to accelerate, watched Nvidia grow to a $5 trillion market cap while Intel sat at $425 billion — a gap built entirely from the second flip.
Each flip follows the same logic. An accelerator handles a specialist workload. That workload becomes the dominant workload of the era. The accelerator becomes the substrate. The old chip gets demoted.
ChatGPT has 900 million weekly active users. Most of them never open a traditional application or touch a file system. They type into a text box and the model decides which tools to call, which code to run, which interface to render. The application layer is collapsing into the model layer.
Claude Code now authors around 4% of all public GitHub commits. Projections place that figure at 20% by the end of 2026. The integrated development environment — the tool that defined software engineering for three decades — has not disappeared. It has become a thin shell wrapped around the model that is actually doing the work.
Microsoft sold its entire spare CPU inventory to Anthropic and OpenAI last quarter. AWS tripled its server CPU buys year over year and still cannot meet demand. The shortage is not a supply problem. It is the third flip pricing itself in.
The workload that defines the third era is language and reasoning over raw multimodal input. The dominant interface for actually using a computer is now natural language directed at a model. When the workload of the era runs on a specific substrate, that substrate wins. The model is becoming the substrate.
Karpathy's framing of Software 3.0 makes the third flip concrete. Software shipped today is fixed. You install Photoshop and you get whatever Adobe shipped in that version. The interface is static. The functionality is bounded by what the engineers wrote.
In Software 3.0, the interface itself gets generated. Raw video, audio, and intent go in. A custom UI comes out. The "app" exists for one prompt and then disappears. There is no installation. There is no version. There is no vendor roadmap to wait for.
The first two flips ate the companies that owned the old chip. They did not eat the people who learned to work with the new substrate. Intel engineers who understood CUDA moved to Nvidia. Systems programmers who understood parallel compute became the most valuable engineers in AI. The people who adapted ahead of the flip were fine. The institutions that didn't were not.
The third flip will distribute the same way. The people who learn to work with the model layer — who understand how to deploy it safely, who know how to build processes around it, who can tell a board of directors whether an AI deployment is actually safe — those people will be in enormous demand for the next decade.
The skills required are not the same as the skills that won the second flip. CUDA expertise and distributed training knowledge built the GPU era. The third era requires different fluency: production safety frameworks, agent orchestration, governance, and the judgment to operate AI systems that none of your organisation's existing processes were designed for.
Intel won the first flip. Nvidia won the second. Both won by being the substrate — the thing that everything else runs on. The third flip is being won by whoever owns the model that everything else runs through.
The current answer is Anthropic, OpenAI, and Google — the three labs whose models are embedded in the most workflows, the most tools, and the most enterprises. But the substrate in the third era is different from the substrate in the first two. Chips are manufactured goods. Models are knowledge goods. Knowledge compounds differently.
The winner of the third flip may not be the lab that builds the best model. It may be the organisation that builds the best production infrastructure around it — the safety systems, the governance frameworks, the operator tooling, the certification of the humans who run it. The third flip creates a layer above the model that nobody has built yet.
That layer is what Production AI Institute is building — starting with the practitioners who will operate it.
The AIDA examination tests applied PSF knowledge across all eight domains — exactly the gaps and strengths covered in this assessment. 15 minutes. No charge. Ever.