TARKA

AI Infrastructure for Nepal

[TARKA @tarkahq]

https://t.co/rT2LCrxRlj

[TARKA @tarkahq]

Every Nepali AI product calls foreign model API and pays for it in dollars. That layer now exists at home. Tarka is open to the public: one OpenAI-compatible API for inference, agents and batch jobs. One API. One bill. #NepalAI #OpenAICompatible #LLMAPI

[TARKA @tarkahq]

@HimalayaAILabs Sign up for generous usage tiers across these models, our broader model lineup, and our AI and agent infrastructure.

Log in today to start building your AI apps, and join us on Discord to share what you’re working on.
https://t.co/10Lvf9ffKZ

[TARKA @tarkahq]

Tarka is proud to announce our partnership with @HimalayaAILabs , an organization dedicated to building AI in and for Nepal.
We’ve supported the training of HimalayaAI's Gen AI Models.
Now we’re hosting them on Tarka for developers and businesses to build with. ↓

[TARKA @tarkahq]

HimalayaAI × Tarka — Building with Nepali AI : https://t.co/10Lvf9ffKZ

[TARKA @tarkahq]

Sign up for generous usage tiers across Himalaya AI models, our broader model lineup, and our AI and agent infrastructure.

Log in today to start building your AI apps, and join us on Discord to share what you’re working on.
Link Below↓
#sovereignAI #Tarka #HAL

[HimalayaAI @HimalayaAILabs]

Nepal's sovereign AI stack now has its two layers joined.

Today we announce our partnership with @tarkahq in building sovereign AI infrastructure for Nepal.

We, Himalaya AI Labs, build the sovereign AI models for Nepal.
@tarkahq , as our model inference partner, helps us host them for public and enterprise application.

And this isn't a cold start — Tarka has helped us train our 1st Pretrained Model HimalayaGPT 0.5B, and our GLM-OCR-Nepali and Nepali-Parler-TTS models since early days of HimalayaAI.
(You will find our models in their Model API tab).

This MoU formalizes our continued partnership to accelerate sovereign AI in Nepal.

हाम्रो AI, राम्रो AI #HAL #TARKA #NepaliAI

[TARKA @tarkahq]

That's the layer Tarka is building: OpenAI-compatible inference, priced by the token, billed in NPR. https://t.co/Ts2hX4nZbV

[TARKA @tarkahq]

That's the gap Tarka is building into: OpenAI-compatible inference built to serve large open-weight models like these, priced by the token, billed in NPR.

https://t.co/3QM8KRUqoV

[TARKA @tarkahq]

A 2.8-trillion-parameter open-weight model shipped this month. A 2.4-trillion-parameter one followed three days later. Open and closed AI are no longer two different conversations.

[TARKA @tarkahq]

For Nepal's builders, that changes the calculation. Running a genuinely frontier-grade model no longer means a closed-API subscription billed in USD — it means GPU compute and an inference layer that can actually serve a model that size.

[TARKA @tarkahq]

At Tarka HQ, we are actively driving this shift—building the tech infrastructure and ecosystem to transition Nepal into a global AI powerhouse.

Let’s connect and build the future together: https://t.co/TbxvroRTVw

[TARKA @tarkahq]

Nepal is sitting on an energy goldmine, but we’re playing the wrong game. Exporting raw electrons to our neighbors is a low-margin trap. The real play? Turn that power into compute. Here’s why powering the AI boom domestically is Nepal’s ultimate economic unlock: 🧵 (1/5)

[TARKA @tarkahq]

This isn’t just about hosting data centers. Nepal’s local tech talent pool is growing exponentially. By combining green electrons with AI & bright minds, we scale from a primary-commodity exporter into a major player in the global digital High value chain.

[TARKA @tarkahq]

Geography is our unfair advantage. Nepal is sandwiched directly between the two biggest consumers of AI workloads on the planet. Processing data locally means we can sell ultra-low latency AI inference straight into these massive neighboring markets.

[TARKA @tarkahq]

Why AI Data Centers makes sense here?

High-Altitude Natural Cooling: Our mountain climate acts as a natural heatsink, drastically slashing cooling costs and driving down PUE.

24/7 Baseload: Consistent, uninterrupted power flow perfect for non-stop computing.

[TARKA @tarkahq]

The baseline scale is massive: Nepal already has 30,000 MW of hydro projects sanctioned within its national capacity pipeline. Instead of allocating this 30GW potential to low-margin, cross-border commodity exports, we need to inject it straight into local AI Data centers.

[TARKA @tarkahq]

The global AI bottleneck is energy. Grids in the US and Europe are completely maxed out, driving data center power costs up to $0.14–$0.20 per kWh. Nepal’s clean hydro? Just $0.03–$0.05 per kWh, and we are already wasting (spilling) massive amounts of it during peak seasons.

[TARKA @tarkahq]

Nepal's AI developers deserve infrastructure that was built for them. That's what Tarka is.

GPU compute. Model training and fine-tuning. Deployment and inference. OpenAI-compatible APIs. NPR billing.

Built for this market. Building now.