*

High-Flyer created DeepSeek in 2023 to focus on artificial

general intelligence (AGI)

*

DeepSeek's AI models praised, trigger global tech share

selloff

*

High-Flyer invested heavily in Nvidia ( NVDA ) chips before U.S.

export

ban

By Eduardo Baptista

BEIJING, Jan 29 (Reuters) - Quantitative hedge fund,

High-Flyer, built a 100 billion yuan ($13.79 billion) portfolio

using artificial intelligence models to make investment

decisions, but in 2023 decided to change track to focus on

developing the most cutting-edge AI.

In a post on its official WeChat account, Hangzhou Huanfang

Technology Ltd Co., as the company is officially called, said it

would focus on pursuing artificial general intelligence (AGI).

"High-Flyer will concentrate its resources and strength,

wholly devote itself to serve AI technology that benefits all of

humanity, create a new independent research group, and explore

the essence of AGI," the company said.

Microsoft ( MSFT )-backed OpenAI, which developed ChatGPT, defines

AGI as autonomous systems that surpass humans in most

economically valuable tasks.

It's the next generation of AI models and in a post on X

last week OpenAI CEO Sam Altman said his company had not yet

achieved that milestone.

The independent research group envisaged by High-Flyer was

DeepSeek, whose models have rocked the global technology sector

in recent weeks. High-Flyer's founder and controlling

shareholder, Liang Wenfeng, doubles as DeepSeek's low-profile

leader.

The sophistication of DeepSeek's models has been widely

praised by its Silicon Valley competitors, a first for a Chinese

AI model, but the startup's claims that it used a fraction of

the computing power deployed by leading U.S. firms for their own

models triggered a selloff of tech shares worldwide.

It is unclear how close DeepSeek is to developing an AGI

model.

While DeepSeek's success appears to have happened almost

overnight, High-Flyer shows how this meteoric rise has been over

a decade in the making.

Under Liang's leadership, the fund spent years studying and

experimenting with overseas AI models, applying this technology

to their business, and investing tens of millions of dollars in

high-end Nvidia ( NVDA ) chips to provide the computing power necessary

to support this AI-centric strategy, according to a Reuters

review of High-Flyer's websites and official WeChat accounts.

SUPERCOMPUTING CLUSTERS

This includes building two AI supercomputing clusters,

entirely made up of Nvidia's ( NVDA ) powerful A100 chips, which

Washington banned from export to China in September 2022.

High-Flyer's A100 clusters were built and put into operation

long before the export controls were announced. Its first

cluster, made up of 1,100 A100 chips, cost 200 million yuan and

was put into operation in 2020, while its second cluster, made

up of around 10,000 A100 chips, was completed a year later with

a cost of 1 billion yuan, according to the company's website and

several WeChat posts.

In 2022, High-Flyer AI researchers presented a strategy at

an Nvidia ( NVDA ) conference that the firm had developed to maximize the

second cluster's efficiency when training AI models.

It is unclear how much High-Flyer has invested in DeepSeek.

High-Flyer has an office located in the same building as

DeepSeek, and it also owns patents related to chip clusters used

to train AI models.

Liang has a 55% stake in privately held High-Flyer and holds

99% of the voting rights, according to Chinese corporate

records. The remaining shares are held by other executives in

the fund.

DeepSeek has so far only claimed to use Nvidia's ( NVDA ) much less

powerful H800 and H20 chips for training its DeepSeek-V3 model

and its predecessor DeepSeek-V2, which triggered an AI model

price war in China when it was released last May.

However, some tech executives have publicly claimed DeepSeek

has far more computing power at its disposal.

Scale AI CEO Alexandr Wang said during an interview with

CNBC on Thursday, without evidence, that DeepSeek has 50,000

Nvidia H100 chips, which he claimed were not disclosed because

that would violate Washington's export controls, which banned

the export of H100 chips to China at the same time as the less

powerful A100.

DeepSeek did not respond to a request for comment on the

allegation. Nvidia ( NVDA ) also did not immediately respond to an email

asking for comment.

But Liang's concern with computing power when discussing

DeepSeek's future echoes his quant fund's massive investment in

AI clusters. Asked in an interview with Chinese media outlet

Waves last July if High-Flyer planned to split DeepSeek from the

company and take it public, Liang answered:

"We have no plans to raise money in the short term, the

problem we face has never been money, but the embargo on

high-end chips."

($1 = 7.2507 Chinese yuan renminbi)