TECH

Boost Inference to Follow in DeepSeek’s Footsteps, AI Veteran Tells Chinese Firms

Published: Feb. 12, 2025 6:44 p.m. GMT+8

00:00

00:00/00:00

Listen to this article 1x

The sudden rise of DeepSeek has led to a dramatic surge in SiliconFlow’s user base. Photo: AI generated

In the wake of DeepSeek’s explosive popularity, its hometown peers should strengthen their own products’ inference capabilities to offer more powerful artificial intelligence (AI) applications, according to an industry veteran.

Inference is a mechanism which large language models (LLMs) use to generate quick human-like responses to user prompts based on their training data.

You've accessed an article available only to subscribers

Subscribe today for just $.99.

VIEW OPTIONS

Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.

Subscribe to both Caixin Global and The Wall Street Journal — for the price of one.

Share this article

Open WeChat and scan the QR code

DIGEST HUB

Digest Hub

Back

Explore the story in 30 seconds

DeepSeek's popularity highlights the importance of enhancing AI inference capabilities, with Chinese companies optimizing algorithms due to U.S. chip restrictions.
SiliconFlow, established in August 2023, supports DeepSeek using both Nvidia and Huawei chips and has rapidly grown its user base.
The company completed a funding round valuing it at $200 million, backed by investors like Sinovation Ventures, contributing to its cloud AI services and inference engines.

AI generated, for reference only

Who’s Who

DeepSeek: DeepSeek has gained sudden popularity due to its strong performance in reasoning, coding, and Chinese language comprehension. It has open-sourced its large models, including V2 using Nvidia's H100 chips and V3 and R1 models with Huawei's Ascend chips. SiliconFlow, its AI inference service provider, experienced a user surge following DeepSeek's rise.

Nvidia Corp.: Nvidia Corp. provides cutting-edge AI chips used by U.S. developers to enhance AI models' inference capabilities. Due to Washington's trade restrictions, Chinese LLM developers are restricted to using Nvidia's lower-end chips or domestic alternatives. SiliconFlow used Nvidia's advanced H100 chips for DeepSeek’s open-source model V2, but later switched to Huawei’s Ascend chips for cost reasons.

SiliconFlow: SiliconFlow, founded in August 2023, is a Beijing-based provider of AI inference services for LLMs. The startup offers a cloud service platform, an AI inference engine, and a text-to-image or video acceleration library. It has worked with DeepSeek, using Nvidia and Huawei chips for AI inference. Recently, SiliconFlow completed a funding round, boosting its valuation to around $200 million. Investors include Sinovation Ventures, Glory Ventures, and MiraclePlus.

Microsoft Research Asia: The article mentions Yuan Jinhui, who is the founder of SiliconFlow and was formerly an AI scientist at Microsoft Research Asia. Other than being referenced in relation to Yuan's background, the article does not provide further details about Microsoft Research Asia.

Huawei Technologies Co. Ltd.: Huawei Technologies Co. Ltd. is mentioned in the article as a provider of Ascend chips, which SiliconFlow utilized for AI inference services for DeepSeek's V3 and R1 models due to cost concerns. Huawei's Ascend chips offer an alternative to Nvidia's advanced AI chips, particularly amid U.S. trade restrictions affecting Chinese developers' access to high-end Nvidia chips.

Sinovation Ventures: Sinovation Ventures is a venture capital firm that has invested in SiliconFlow, a Beijing-based AI inference service provider. The firm is associated with Kai-Fu Lee, a prominent figure in the tech industry. Sinovation Ventures is known for supporting innovative technology startups, fostering advancements in AI and other cutting-edge fields.

Glory Ventures: The article mentions that Glory Ventures is one of the venture capital firms that has invested in SiliconFlow, a provider of AI inference services.

MiraclePlus: MiraclePlus is one of the venture capital firms that have invested in SiliconFlow, a Beijing-based provider of AI inference services. SiliconFlow was established in August 2023 and has recently completed a funding round, bringing its valuation to around $200 million.

AI generated, for reference only

What Happened When

August 2023：: SiliconFlow was established, offering a cloud service platform and AI inference engine for LLMs

May 2024：: DeepSeek's second-generation open-source model V2 was launched, supported by SiliconFlow's AI inference services built on Nvidia's advanced H100 chips

December 2024：: DeepSeek launched its blockbuster V3 model, for which SiliconFlow provided AI inference services using Huawei Technologies Co. Ltd.'s Ascend chips due to cost concerns