Running AI Workloads on Alibaba and Tencent Instead of Nvidia DGX Cloud

Description and use

Migrating non‑GPU AI workloads to Alibaba and Tencent reduces costs to <$15/year for indie developers, maintaining performance for data pipelines, vector databases, and inference APIs.

Problem

Indie developers and bootstrapped startups face prohibitive costs when using premium Western cloud infrastructure for AI workloads, especially for non‑GPU tasks.

Outcome

By shifting non‑GPU AI workloads to Alibaba and Tencent, the author reduced annual compute costs from ~$50/month to <$15/year, enabling faster iteration and lower burn rate for indie projects.

Workflow

• 1. Identify non‑GPU workloads (data preprocessing, vector DB, inference APIs).
• 2. Compare pricing of Western clouds vs Alibaba/Tencent for CPU instances.
• 3. Create new accounts on Alibaba Cloud and Tencent Cloud to access new‑user deals.
• 4. Grab Alibaba Cloud Lightweight Server (¥38/year) during daily flash sales at 10:00/15:00 Beijing time.
• 5. Optionally use Alibaba ECS Economic‑e 2C2G 3M (¥99/year) for price‑locked renewal through 2029.
• 6. For memory‑heavy vector DBs, select Tencent Cloud 2C4G (¥188/year).
• 7. Provision instances via official deal pages (aliyun.com/minisite/goods?userCode=tzlh4rrj and cloud.tencent.com/act/cps/redirect).
• 8. Deploy Dockerized AI inference servers or vector DBs on these instances.
• 9. Monitor usage and renew before expiry to maintain locked pricing.

Evidence

For massive enterprises with bottomless budgets, sure. But for indie developers, bootstrapped startups, and solo hackers? That’s just a non‑starter.

Source

I completely ditched the premium tiers of my old providers and moved my infrastructure to Alibaba and Tencent Clouds.

Source

For less than $15 a year, I can host a robust data preprocessing pipeline that would cost me $50 a month on a Western provider.

Source