Name: Baseten's Tuhin Srivastava on China AI Coding breaking into the U.S. — 1/26/2026
Uploaded: 2026-01-26T19:58:33+00:00
Duration: 1428 s

Speakers

Deirdre Bosa — Anchor/Reporter, CNBC Tech Check

Tuhin Srivastava — Co-founder and CEO, Baseten

Summary

CNBC's Deirdre Bosa and Baseten CEO Tuhin Srivastava discuss the rise of Chinese open-source AI coding models, especially Zhipu AI's GLM and Alibaba's Qwen, and their adoption by U.S. developers and enterprises. They cover how reasoning and agentic applications are driving 100x inference growth and more compute demand, while Nvidia faces growing chip diversification from TPUs, AMD, and Huawei. Srivastava argues the Chinese open-source model surge is not a DeepSeek blip and that the massive U.S. AI infrastructure buildout is supported by global demand.

Zhipu AI limited access to its GLM 4.7 coding assistant after viral demand from China and the U.S.
Baseten's Tuhin Srivastava says multiple Chinese open-source models are competitive with U.S. closed-source frontier models.
Enterprises are adopting Chinese open-source models in production for cost, customization, and VPC control.
Inference demand has grown roughly 100x, driven by adoption, reasoning models, and agentic applications.
Nvidia remains advantaged through CUDA and chip generations, but AI workloads are diversifying to TPUs, AMD, and Huawei.
The U.S. AI data center buildout is framed as justified by global demand and the scale of the AI opportunity.

Ideas

Deirdre Bosa Anchor/Reporter, CNBC Tech Check 0:12

Zhipu AI worth watching after viral coding.

Zhipu AI is a top-ranking Chinese AI startup whose GLM 4.7 coding assistant has seen demand so strong that access was limited, with users concentrated in China and the U.S. It recently went public in Hong Kong and released a major model trained entirely on Huawei chips, making it a Chinese AI coding competitor investors should monitor.

Tuhin Srivastava Co-founder and CEO, Baseten 10:22

Chinese open-source AI models competitive, not blip.

Chinese open-source AI models are not just a DeepSeek blip: multiple labs such as Alibaba's Qwen, Moonshot's Kimi, Zhipu's GLM, and DeepSeek are producing models for real-world tasks, and the gap to closed-source U.S. frontier models is closing monthly. Enterprises are deploying them in production because they are cost-efficient, customizable, and can run inside their own VPCs, and the speaker expects open-source options to match closed-source models within days or simultaneously within a year.

Tuhin Srivastava Co-founder and CEO, Baseten 19:33

AI chip diversification is needed.

Supporting the AI application layer will require diversification across chips, not just Nvidia. The speaker sees real momentum in alternatives such as TPUs, AMD, and Huawei chips, and expects inference-specialized architectures to become a larger part of the next generation of AI hardware.

Tuhin Srivastava Co-founder and CEO, Baseten 20:19

Nvidia maintains AI chip leadership advantages.

Nvidia remains a dominant AI hardware company because of its generations of chips and CUDA software ecosystem; it is embedding itself in inference through partnerships like Baseten, announced another chip generation, and its recent acquisition shows it is pursuing LPU-type inference-specialized architectures. He expects more inference-specialized work over the next 12 months, benefiting Nvidia.

Tuhin Srivastava Co-founder and CEO, Baseten 22:22

AI infrastructure buildout justified by global demand.

The massive U.S. AI data center buildout is justified because U.S. companies serve a global AI market, not just domestic users, and because AI could become the biggest market ever created. The value delivered by inference and AI applications could grow 10x, 100x, or 1000x over five years, supporting the scale of infrastructure spending.

Baseten's Tuhin Srivastava on China AI Coding breaking into the U.S. — 1/26/2026

Summary

Ideas

Up Next