LectureHSE · May 23, 2026
Cloud Infrastructure for AI Business
How managers choose a platform, govern complexity and evaluate infrastructure economics
/ Cloud Infrastructure for AI Business · HSE 2026
Slide contents
1. Cloud Infrastructure for AI Business
How managers choose a platform, govern complexity and evaluate infrastructure economics
2. Alexander Polomodov
Technical Director & Fellow, T-Technologies
Architecture and engineering R&D.
AI adoption in SDLC.
Focus: cloud platform choice.
3. Here cloud is a management choice
Options, criteria and regular governance
No basic recap — Cloud, IaaS/PaaS/SaaS — in notes.
We look like managers — Options, acceleration, risk.
We calculate economics — We keep the resource, hardware, payback and FinOps examples.
4. 01. Management frame
Before choosing technology, understand which speed, control and accountability the business is buying
5. Which decisions managers actually make
Platform: public, managed, private, hybrid.
Where speed matters; where control matters.
Owners: platform, product, security, finance.
Loop: cost, reliability, reversibility.
6. A platform portfolio beats one answer
The rational answer is usually a mix
Public / managed — Speed and services; provider-rule dependency.
Private platform — Scale, compliance, predictable load.
Hybrid strategy — Core controlled; spikes stay in cloud.
7. Evaluate speed against control
What accelerates business
Launch time and change frequency.
Managed services: DB, queues, AI, observability.
Reduced infrastructure load on product teams.
What contains risk
Data, compliance, access, audit.
Predictable cost under growing load.
Critical-loop exit strategy.
8. 02. Migration as a business program
Move not servers, but the model of speed, risk, accountability and change governance
9. Migrate only for business pull
Migrate when
Faster product and environment launch.
The existing platform blocks delivery speed.
Clear domain, owner, success criteria.
Wait when
Stable, cheap, non-blocking service.
Team cannot own the new ops-loop.
Technical fashion without business impact.
10. Migration options: rehost / replatform / refactor / replace
At each step recalculate not only progress, but also accountability, cost and reliability
11. Executive anti-patterns
Migration without priority and owners.
Move itself beats speed/reliability/economics.
Unclear: on-call, changes, risk, bill.
Old delivery process in new cloud.
12. 03. Platform governability
Repeatable safe infrastructure change
13. IaC is management control
Infrastructure as code (IaC) matters when change trace, guardrails and environment repeatability matter
14. Kubernetes must earn its complexity
Orchestration pays off only with a platform team, mature practices and real service variability
15. Responsibility model: platform, product, security, finance
Unnamed accountability appears in incidents
Platform — Standards, templates, baseline reliability, observability, security and the change path.
Product — Architecture decisions, load, user value and product unit cost.
Governance — Security, access, audit, forecast, allocation and TCO.
16. 04. Money calculation example
Compare order of magnitude: cloud resources, buying a server and the payback point
17. Price per cloud resource unit
720 hours per month, public Yandex Cloud rates for vCPU, RAM and network storage
18. Cloud equivalent of a regular server
Rough monthly cost of compute, RAM and two storage scenarios
19. Cost of similar hardware
Catalog example
Dell PowerEdge R750: 24 physical cores.
64 GB RAM in base configuration.
Catalog price: from 566,050 ₽.
What is not included
noHDD: price disks separately.
No DC, power, network, licenses, ops.
Educational order-of-magnitude estimate.
20. Payback point and private cloud at scale
At large predictable load, managers must calculate when owning the platform is better than buying it as a service
21. 05. FinOps and regular governance
Recurring accountability, forecast and team behavior
22. Cost drivers and expense allocation
You need to see which decisions create spend: compute, storage, network and managed services
23. FinOps loop: showback/chargeback, optimization, forecast
Finance, product and engineering must discuss cost in one regular loop
24. Unit economics instead of total invoice size
Calculate cost per useful product unit.
Unit economics shows margin decay earlier.
Reliability has price and risk.
Mature FinOps changes team behavior.
25. Final decision matrix
Key question: when is owning the baseline platform cheaper and more reliable than buying it as public cloud?
26. References and materials
Core sources
FinOps Foundation - FinOps Framework: finops.org/framework/
CNCF Kubernetes docs: kubernetes.io/docs/ · cncf.io
Cloud architecture frameworks.
Yandex Cloud pricing and Servermall R750.