AI Infrastructure Operations
Your AI bill is about to double.
We make it predictable.
We put self-hosted models on your cloud, behind a gateway you control, with cost visibility per team — then operate it monthly. Your engineers change one environment variable. Your finance team finally gets a forecast they can use.
- OPENAI_BASE_URL=https://api.openai.com/v1
- OPENAI_API_KEY=sk-proj-… # billed per token, untraced
+ OPENAI_BASE_URL=https://llm.yourco.internal
+ OPENAI_API_KEY=sk-… # your gateway, your budget, your records
01 The problem
Someone else is paying for your inference. That ends.
Frontier labs have been selling tokens and seats below what the GPUs cost them, funded by capital rather than revenue. That gap closes eventually — through price rises, tighter limits, or both. The teams that built on a subsidy and never measured it will find out the hard way.
Nobody in the building can say what an agent run costs, which team spent the most last month, or what would happen to the bill if list prices doubled. The invoice arrives as one number.
Every laptop, CI runner and service holds its own vendor key. There is no budget ceiling, no rate limit you set, no record of which prompts left the building or what came back.
Switching models means a change-management project across every machine and pipeline. So you don't switch — and the vendor knows it. That is the whole business model.
Legal asks where the prompts go. Nobody has an answer that survives an audit, and "the vendor says they don't train on it" is not one.
02 What we build
One change for your engineers. Full control for you.
Four open-source components, wired correctly, on infrastructure you own — then kept alive by people who have done it in production.
A · ZERO DISRUPTION
Zero-disruption migration
Your AI assistants, your editors, your frameworks, your own SDK calls, your CI evals — they already speak one API shape. The gateway answers in that shape, so the migration is a base URL and a key, not a rewrite.
B · MODEL FREEDOM
Switch models without a project
A new open-weight release ships, benchmarks well on your evals, and goes live in a routing rule that afternoon. Nobody reinstalls anything. Cheap models take the bulk traffic, expensive ones take what needs them.
C · COST VISIBILITY
Cost per team, per call
Every request is traced: who asked, which model answered, how long, how much, how good. Cost per team stops being a guess and becomes a line in a dashboard you can hand to finance.
03 What you get back
This is what cost visibility looks like.
Not a monthly invoice. A span waterfall, per call, with the model, the tokens, the tool hops and the cost attached — kept as long as your retention policy says.
How the record works04 How we work
Build it once. Then someone has to run it.
The build is the easy half. The half that matters is the eighteen months after, when models deprecate, GPUs fall over and the bill drifts.
05 Why us
We ran this stack on ourselves first.
Guztia has been doing cloud operations since 2015 — migrations, Kubernetes, Terraform, 24/7 retainers, mostly across APAC and China. The discipline transfers directly. A GPU node that OOMs at 3am is not a new category of problem.
- We run this exact arrangement in our own production — it is what we operate daily, not a demonstration built for the pitch.
- Alibaba Cloud MVP eight years running, which is why deployments in Asia-Pacific are routine here rather than a research project.
- China cross-border since 2017: if your models need to answer inside the mainland, that is an old problem here, not a new one.
- Two named engineers — Alberto Roura and Xavier Macia — plus a specialist network. You always know who is on the incident.
06 Where it runs
Your account, your region, your commitment discounts.
We deploy into infrastructure you already pay for. No Guztia-hosted middle layer, no per-token margin, nothing to migrate off if you fire us.
07 Still here
The China practice didn't go anywhere.
Business VPN, site acceleration, ICP filing, cross-border connectivity, Salesforce access from the mainland. Same team, same lines, same licences — now with the added question of where your models are allowed to run and whose jurisdiction the prompts land in.
08 Writing
Nine years of notes from inside other people's infrastructure.
Testing Alibaba Cloud Model Studio (Bailian 百炼)
I’ve written about AI and LLMs a number of times in this blog in the last couple of years, but in all theoccasions I was referring to OpenAI and th...
การทดสอบ Alibaba Cloud Model Studio (Bailian 百炼)
ในช่วงสองสามปีที่ผ่านมาผมได้เขียนเกี่ยวกับ AI และ LLMs หลายครั้งในบล็อกนี้แต่ทุกครั้งที่พูดถึงจะเป็นการอ้างถึง OpenAI และโมเดล GPT3 ของพวกเขา
Xataka Nos Ha Dedicado Un Artículo!
Esto debe ser el cielo, o parecido. Xataka (la publicación sobre tecnología en Español más importante) ha dedicado unartículo a Alberto, creador de...
Migración Entre Regiones De Alibaba Cloud
El cloud, o “la nube”, es algo que tiene un potencial enorme para mejorar la eficiencia de las empresas, pero a menudohace falta hacer ajustes por ...