The most outstanding financing of the year: take Nvidia H100 mortgage loan of 16.5 billion, buy more Nvidia GPUs and expand 10 data centers

Xi Xiaoyao said to share
the source | Qubit

Nvidia H100 can be mortgaged ?

CoreWeave , a rookie in AI computing power , did not disclose how many cards it mortgaged. Anyway, it obtained debt financing of 2.3 billion US dollars (about 16.5 billion yuan).

The announcement reads, "We are excited and humbled to share the news..."

picture

This company was originally invested by Nvidia , and now it has borrowed a large sum of money with Nvidia GPUs , what do you plan to use it for?

Buy more Nvidia GPUs! 10 more new data centers will be built before the end of the year!

Netizens live in Bengbu now: Can it still be purple?

Large model research test portal

GPT-4 capability research portal (advanced/continue to visit in case of browser warning):

Hello, GPT4!

picture

Some people simply put forward that GPU is the new dollar.

picture

Nvidia is chasing Seca, Microsoft is rushing to invest

Until a few months ago, few people had heard of CoreWeave.

"We came out of nowhere", that's how Brannin McBee, Lianchuang and chief strategy officer, introduced his home to the outside world.

In fact, CoreWeave was founded in 2017. The three joint ventures were originally commodity traders , and their initial business was cryptocurrency .

After mining for more than a year, by 2018 they accidentally decided to expand the GPU acceleration business.

picture

At that time, AI was not so popular, and a large part of the business came from the CG rendering industry. So by 2022, the total annual revenue is 25 million US dollars.

Unexpectedly, in 2023, I will take the express train of AIGC, and the annual revenue is expected to exceed 600 million US dollars , which is a full 24 times increase.

The more critical point is that CoreWeave has been favored by the real winner of computing power, Nvidia Huang .

According to The Information, despite such a shortage of Nvidia H100, a large number of new cards have been allocated to CoreWeave , and the limited supply to established cloud computing companies such as Amazon and Microsoft.

External analysis is because these established companies are developing their own AI acceleration chips, hoping to reduce their dependence on Nvidia, so Lao Huang will help them.

In addition to chasing the card , Lao Huang also directly invested in CoreWeave.

It has invested US$100 million before, and Nvidia was a major participant in the US$421 million Series B financing in April this year, which raised CoreWeave's valuation to US$2 billion .

In addition to Nvidia, Microsoft also signed an agreement with CoreWeave in June to invest billions of dollars in cloud computing infrastructure construction in the next few years.

What is the magic of this company?

To build the world's largest AI server cluster

According to CoreWeave's official website, their services are 80% cheaper than traditional cloud computing vendors.

picture

How cheap is it?

Nvidia's latest HGX H100 server, which contains 8 H100s with 80G video memory and 1T memory, starts at $2.23 per hour (16 RMB).

picture

On how it did it, Coreweave revealed some details on Odd Lots, a Bloomberg podcast.

48,000 discrete connections need to be established between computers, switches, routers, etc., and more than 500 miles of fiber optic cables are required to build a cluster of 16,000 H100s... This is completely different from the way data centers were designed in the past.

as well as:

Because it has won the trust of Nvidia, CoreWeave can quickly launch new designs, while cloud computing companies usually need several quarters to obtain new chips.

picture

In addition, CoreWeave's business partners are not small.

For example , InflectionAI , a large-scale model company co-founded by Linkedin co-creator Reid Hoffman and DeepMind co-creator Mustafa Suleyman, has launched the chat assistant Pi and the basic model Inflection-1.

Previously, the two companies and Nvidia announced that they would cooperate to build a super-large AI server cluster, with the goal of running 22,000 Nvidia H100s .

If completed, it will exceed Meta's previously announced 16,000 GPU plan and become the world's largest AI server cluster .

picture

picture

picture

picture

References

 [1]https://coreweave.com/blog/coreweave-secures-2-3-billion-debt-financing-magnetar-capital-blackstone?utm_medium=social&hss_channel=tw-979803443681349632
 [2]https://www.cnbc.com/2023/06/01/microsoft-inks-deal-with-coreweave-to-meet-openai-cloud-demand.html
 [3]https://www.theinformation.com/articles/with-nvidias-help-revenue-surges-at-smaller-cloud-providers  
 [4]https://www.coreweave.com/products/hgx-h100
 [5]https://inflection.ai/inflection-ai-announces-1-3-billion-of-funding
 [6]https://www.youtube.com/watch?v=9OOn6u6GIqk

Guess you like

Origin blog.csdn.net/xixiaoyaoww/article/details/132140401