A history of deep learning and how it's being used in more tech products than ever
Decades-old discoveries are now electrifying the computing industry and will soon transform corporate America. — Over the past four years, readers have doubtlessly noticed quantum leaps in the quality of a wide range of everyday technologies.
Context & Ripple Effects
Written at the moment deep learning stopped being a lab topic, this Fortune piece lands one year after Google open sourced TensorFlow, handing every developer the same machine-learning system the company used internally. That release is the hinge the article sits on: decades-old neural network math suddenly had free, industrial-grade tooling attached to it.
The arc since then runs through the history of generative AI — CUDA, convolutional networks, transformers — and into today's product blitz from Google and OpenAI, making this 2016 snapshot the early frame for a shift still playing out.
First-order effects
- Product teams across the computing industry gain a working recipe: embed pretrained neural networks into existing software rather than build AI features from scratch, which is why the article expects transformation to reach corporate America rather than stay in research labs.
- Google's TensorFlow open-sourcing turns its internal ML stack into a de facto standard, pulling startups and enterprises onto Google's tooling at zero license cost.
Second-order effects
- As adoption spreads, the technique's boundaries become commercially urgent — by 2018 the debate about deep learning's limits pushes research groups and startups toward alternative concepts promising more flexible AI.
- Whoever controls the training stack and compute gains leverage over the growing population of dependent developers, turning ML frameworks into competitive infrastructure rather than commodities.
Third-order effects
- If the pattern holds, value migrates with cost: as model prices fall — the dynamic behind DeepSeek's breakthroughs being called a win for app developers — margins concentrate in the application layer while intelligence itself becomes cheap plumbing.
- The long endpoint visible in the coverage is cultural as much as technical: Hinton's argument that neural networks are a better form of intelligence reframes them not as a tool inside products but as the substrate future products are built on.
The trend: Deep learning is completing its move from an embedded feature inside tech products to the default substrate of the software industry, with each cost decline shifting value toward the application layer built on top of it.