Tim Urista
AI Infrastructure & Unit Economics. I make compute fleets legible, and I make their costs attributable — from $10–12M/quarter cloud spend to the per-signal economics of the AI systems I run myself.
"Nobody owns a cost they can't see."
I'm a cloud and AI infrastructure engineer who has worked in big tech at Apple, Meta and Roku, including on fleet-scale private cloud infrastructure for internal AI workloads, and I build and run my own products. My through-line for the last decade has been a single discipline: cost attribution. Cost data is an engineering artifact, and most organizations still treat it like an accounting afterthought.
The arc is rare, and it spans three scales. At Roku I built a cost-attribution pipeline handling $10–12M/quarter of AWS spend and co-led the ECS→Kubernetes migration. At Apple I worked on financial attribution correcting misattribution across roughly 700,000 servers, and on internal platforms used weekly by hundreds of people across finance, SRE, and leadership. And as a founder, I carry unit economics down to my own invoices — TrendVesting has run in production since February 2024, on an invoice I pay myself, and measuring it honestly overturned some of my own numbers.
Every product I build has teaching DNA in it: making complex systems legible to the people who have to live with them. FinOps is the community I build in; AI Infrastructure & Unit Economics is the identity I own. Home is Temecula — my wife, our two daughters, and the pets who are the reason the invisible infrastructure has to stay invisible.
What I Do
Cost Attribution & FinOps
- • Fleet-scale financial attribution (~700K servers)
- • Cloud spend pipelines ($10–12M/quarter)
- • AI unit economics: cost per inference / signal / request
- • Showback & misattribution correction
- • LLM-assisted cost-hierarchy labeling
AI & Cloud Infrastructure
- • Private cloud backbone for AI workloads
- • Kubernetes & ECS→K8s migrations
- • Health-check & failure-pattern detection
- • Multi-mode Go / Python backends
- • Production AI reliability & observability
Product & Platform (UX)
- • Internal platforms used weekly by hundreds of people
- • Finished experiences, not one-off features
- • React/Vite, Next.js, Astro, FastAPI
- • Making cost data legible to non-engineers
- • MCP integrations & markdown DSLs
Foundations
- • Cost data as an engineering artifact
- • Teaching: making complex systems legible
- • HIPAA / compliant architecture (BAA chains)
- • Internal tooling as product with users
- • Georgia Tech MS CS coursework
The Arc
Senior Cloud Engineer
Big techApple
Worked on fleet-scale private cloud infrastructure for internal AI workloads: financial attribution correcting misattribution across ~700K servers, and internal platforms that made livability, deployment state, and cost visible to finance, SRE, and leadership.
The Theme Begins
RokuCost attribution as observability for money
Built a cost-attribution pipeline handling $10–12M/quarter of AWS spend and co-led the ECS→Kubernetes migration. Where I learned that cost data is an engineering artifact, not an accounting afterthought.
The Climb
HealthTrio → MetaHealthTrio, Weedmaps, IterateAI, Meta
HealthTrio (Java/SOAP, first HIPAA exposure), Weedmaps, and IterateAI (no-code ML in the Word2Vec era). At Meta, built a CLI tool for upstream service tracing adopted well beyond my team — internal tooling is a product with users, not a script with tolerance.
Teaching Roots
Before TechBASIS Charter School, Tucson · Occidental College
Taught 6th grade and built a Python grammar game for my students — my first real product. Studied philosophy and religious studies. Every product since has teaching DNA: making complex systems legible to the people who live with them.
Values & Operating Principles
Cost is an engineering artifact
Nobody owns a cost they can't see. Cost attribution is observability for money — instrumented into the system, not reconciled after the fact.
The invoice tells the truth
The diagram tells you how the system works; the invoice tells you the truth. Every architecture gets validated against real spend.
Finished experiences, not features
Full-stack surfaces other teams operate themselves. Make the complex legible to the people who have to live with it — that's the teaching DNA.
Own the identity, rent the title
AI Infrastructure & Unit Economics is mine regardless of any job title. FinOps is the community and the moat, not the headline.
Research
Who Pays for the KV Cache? Attributing Shared AI Inference Spend Across Kubernetes and LLM Provider Bills
When several teams share one inference server, the bill does not divide itself: somebody picks a metering rule. This paper measures how much that choice moves, and where cost attribution breaks at the seams between Kubernetes allocations, gateway logs and provider billing APIs. Five case studies, including a validation run with vLLM on an NVIDIA H100.
Press Kit
Bio Variants
Tim Urista works in AI Infrastructure & Unit Economics — making compute fleets legible and their costs attributable. A cloud infrastructure engineer with past big tech experience at Apple, Meta and Roku, and founder of TendForm and TrendVesting, he has carried cost attribution from $10–12M/quarter cloud spend to ~700K servers to his own production invoices.
Tim Urista's discipline is AI Infrastructure & Unit Economics: he makes compute fleets legible and their costs attributable. At Apple he worked on fleet-scale private cloud infrastructure for internal AI workloads, including financial attribution spanning roughly 700,000 servers. Earlier, at Roku, he built a cost-attribution pipeline handling $10–12M/quarter of AWS spend and co-led an ECS→Kubernetes migration. He carries the same rigor to his own products — TrendVesting, an AI system in production since 2024, and TendForm, a HIPAA-tier SaaS. He writes about cost attribution and is active in the FinOps community. "Nobody owns a cost they can't see."