While Huang Renxun and Su Zifeng are making money, Zuckerberg and Musk are burning it, and Ultraman also wants a piece of the pie.

Wallstreetcn
2024.01.20 08:34
portai
I'm PortAI, I can summarize articles.

Developing large-scale machine learning models requires a significant amount of computational resources, especially relying on the computing power of high-performance AI chips. AI startups that are crazily raising funds may continue to support the growth of GPU demand.

Tech stocks will continue to shine in 2024, with OpenAI's biggest "shovel seller" NVIDIA seeing an 8.7% cumulative increase this week, and both AMD and NVIDIA hitting new all-time highs for two consecutive days.

OpenAI struck gold by selling shovels, and in the face of sustained high demand, "the king of computing power" NVIDIA has become the biggest beneficiary, with expected record-breaking revenue of $58.86 billion in 2023, a YoY growth of 118%. AMD is also raking in the profits, with GPU revenue expected to exceed $2 billion in 2024.

However, most of the money earned by NVIDIA and AMD comes from the rapidly growing artificial intelligence companies that burn through cash. This includes not only rising stars in artificial intelligence like OpenAI and Anthropic, but also tech giants like Alphabet-C, Microsoft, Meta, and Tesla, all of whom are heavily investing in the development of large-scale artificial intelligence models.

The "unicorn" OpenAI's crazy fundraising may support continuous growth in GPU demand

When the global primary market suffered a comprehensive collapse last year, the "unicorn" OpenAI went on a crazy fundraising spree, propping up half of the financing for startups in the early part of last year.

According to data from Crunchbase, AI startups raised nearly $50 billion in 2023, with 8 rounds of financing exceeding $50 million.

The latest reports from the media state that x.OpenAI, an AI company under Musk's umbrella that competes with OpenAI, has completed a $500 million first round of financing, half of its $1 billion target, with a post-investment valuation of $15-20 billion. Musk stated in November last year that equity investors in x.OpenAI would own 25% of the company, and it is expected that Musk and the investors will finalize the terms in the coming weeks.

However, this report was quickly "debunked" by Musk, who responded on Twitter that it was fake news.

Canadian AI startup Cohere is also in talks for a new round of financing, with the funding amount expected to be between $500 million and $1 billion. Tech giants such as NVIDIA, Oracle, and Salesforce are participating in this round of financing.

Cohere has not yet determined the valuation or exact funding target for the new round. The company raised $270 million in a Series C financing round in June last year, with a valuation of $2.2 billion.

In addition, a month ago, there were reports that Anthropic, OpenAI's biggest competitor in the AI startup space, was in talks to raise $750 million in funding led by Menlo Ventures. Developing large-scale machine learning models requires a significant amount of computing resources, especially relying on the computational power of high-performance OpenAI chips. The rapidly growing OpenAI startup may continue to support the increasing demand for GPUs.

Giants Rush to Purchase NVIDIA Chips

At the same time, tech giants are also burning money, with most of these huge funds being converted into purchases of OpenAI chips.

Looking towards AGI, Mark Zuckerberg announced on Thursday that by the end of this year, Meta will have approximately 350,000 NVIDIA H100 chips. If other GPUs are included, the equivalent computing power will be around 600,000 H100 chips.

600,000 is an astonishing number, and the cost of purchasing these chips may be even more astonishing, conservatively estimated to be around 9 billion US dollars.

Analysts from Raymond James pointed out that the price of each NVIDIA H100 chip is between $25,000 and $30,000. Although Huang Renxun has said "the more you buy, the more you save," even with large-scale purchases, it may still cost over 9 billion US dollars.

A recent report from market research firm Omdia Research shows that in 2023, Meta and Microsoft purchased 150,000 H100 GPUs each from NVIDIA, ranking first in terms of procurement volume among all companies.

Tesla currently uses NVIDIA's OpenAI chips to support its autonomous driving software. However, NVIDIA's order volume is already close to full capacity, making it difficult to meet Tesla's chip requirements. This is also one of the important reasons why Tesla has started to develop its own OpenAI chips.

OpenOpenAI also Wants a Piece of the Pie

However, this wave of "unprecedented wealth" brought by chips is not only benefiting the two chip giants, NVIDIA and AMD. Downstream, OpenOpenAI also wants a share.

According to media reports on Friday, OpenOpenAI CEO Sam Altman is in discussions with the Abu Dhabi G42 Fund and SoftBank Group from Japan to raise billions of dollars to establish a new OpenAI chip company. Altman's goal is to use the funds to establish a factory network for chip manufacturing, directly competing with NVIDIA.

The report states that OpenOpenAI is discussing raising $8 billion to $10 billion with G42 alone, but the negotiations are still in the early stages and the complete list is not yet confirmed.

Currently, OpenAI still prefers to purchase chips from NVIDIA and AMD. However, considering that most of the profits flow into the pockets of "shovel sellers," if they do not explore other chip supply sources, they will forever be "working for Huang Renxun and Su Zifeng." Perhaps for this reason, almost all of NVIDIA's major customers, except for Tesla and OpenAI, have announced that they will develop their own OpenAI chips.