AI as infrastructure: the demand that changes its name
When AI becomes infrastructure, the job title does not say "AI". It says "platform engineering", "MLOps", "data engineer". Demand grows invisible to those who track only by title.
What AI as infrastructure means
In the first phase (2020–2023), AI was a competitive differentiator: whoever had AI stood out. In the second phase (2024–), AI becomes basic infrastructure: whoever doesn't have it falls behind. In this transition, demand does not disappear — it dilutes into other job titles and requirements embedded in positions that previously would have been "just backend".
The signal: companies that build silently (high hiring, low narrative) are those in the infrastructure phase. Those making noise are still selling differentiation.
Recent VAIA signals
657 vs 38
Anthropic vs OpenAI jobs
Anthropic has 33× more jobs per HN mention than OpenAI. Builds infrastructure; OpenAI builds narrative. Signal: cross-hn-openai-anthropic (91).
6,3%
tech hires are AI/ML
vs 80% of investment narratives. The difference reveals that AI is already in the requirements for roles that don't use "AI" in the title.
$5B
Anduril — AI as defense platform
Defense-Tech replaced consumer tech as the primary late-stage capital destination. AI is not a product — it is national defense infrastructure.
HN vs TC
utility vs capital — two conversations
HN discusses technical utility (tool fatigue, latency, inference cost). TechCrunch discusses mega-rounds. These are conversations about different phases of the same transition.
What to track
- →MLOps, data platform and feature store engineering jobs — indicate AI infrastructure in production, not pilot.
- →GitHub repos for model serving and observability — the ecosystem for maintaining models in production.
- →Anthropic/OpenAI/Google DeepMind hiring ratio — who is in the building phase vs the marketing phase.
- →Capital allocated in inference chips vs training chips — indicates infrastructure maturity.