Misunderstanding Economics: AI Consumes More Than Just Software — It's Eating the Entire Internet

SaaS becomes AaaS — and people are the fuel

At GTC 2026, Jensen Huang of NVIDIA reportedly told the audience that most SaaS firms will become AaaS — Agentic as a Service. Short sentence: software will stop asking for clicks and start doing work on its own. That shift matters because it changes the unit of value from human operators (seats, logins) to autonomous agents that run workflows independently. It has been reported that Meta plans deep cuts — about 20% of staff, roughly 16,000 people — and is quietly recording engineers’ screens and prompts to extract knowledge and turn processes into trainable data. Why keep a role that can be encoded and run by an agent?

The internet is not tooling anymore; it’s the feedstock

For Western readers: think of the internet not as a marketplace of tools but as the raw material for large language models. OpenAI’s GPT, Anthropic’s Claude, Google’s Gemini and domestic Chinese efforts reportedly including Tongyi Qianwen (通义千问) from Alibaba (阿里巴巴) — and other local models — are all trained on the same trove: Common Crawl, Wikipedia, forums, code repositories, books and academic papers. The point is McLuhan-esque: new media turn old media into content. Newspapers became web pages; now the web becomes the stomach that digests every prior medium and serves it up as model cognition. Software doesn't vanish; it is downgraded into instruments agents call.

Geopolitics, chips and the cost of automation

There is a geopolitical angle too. Training and serving these agentic systems requires vast GPU fleets, and export controls and trade policy have made advanced chips a strategic bottleneck. US curbs on cutting-edge AI silicon have accelerated Chinese efforts to build domestic stacks and raised the infrastructure bill for Western platforms. Companies like Meta are cutting headcount partly to reallocate spend to AI infrastructure — it has been reported that those savings are intended to pay for the compute needed to deploy “AI-assisted” workers at scale. The result: a realignment of labour, capital and supply chains across borders.

The takeaway is stark. This is not a story about individual layoffs or a clever new product. It is a structural change: internet-era software becomes the substrate of a new cognitive medium. Will jobs and companies simply vanish? No — they will persist as data, APIs and training signals — but their economic role will be diminished, repurposed and priced differently. Who wins? Those who control models, data access and compute. Who loses? The middlemen whose value relied on process friction.