Nvidia s market capitalization has repeatedly reached record highs, and Huang s next step is to find

Mondo Finance Updated on 2024-03-07

Visual China.

Titanium**Note: This article**Yu WeChat***Love Fan Er (ID: ifanr), author |Fang Jiawen, Titan** is published with permission. In the age of AI, everything is getting faster.

Last year, ChatGPT recorded the fastest user growth rate - more than one million users in 5 days and 100 million users in 2 months, completing the achievement that Facebook took four and a half years to achieve.

This year, NVIDIA's market value has soared, with a single-day market value of $277 billion, breaking Meta's single-day growth record of US ** value that was set fresh only 20 days ago.

On the way to break through the $2 trillion market capitalization, Nvidia jumped from $1 trillion to $2 trillion in just 180 trading days, while Apple and Microsoft took more than 500 trading days.

Behind the success, the company, which has been called the only arms dealer in the AI war, also needs to face the challengers that are constantly climbing up.

In a previous article, we retraced how Huang's decision and gamble allowed Nvidia, which was on the verge of bankruptcy many times, to finally become the king of chips.

Today, we will take a closer look at how NVIDIA, as the king, is facing challenges and investing in the future in search of the next AI fever .

Crisis and opportunity in the tide of core shortage.

Elon and I were begging him, and I think that's the best way to describe it, and that hour was made up of sushi and begging.

Oracle founder Larry Ellison talked about how he and Musk once invited Nvidia CEO Jensen Huang to a high-end Japanese restaurant in California for dinner, begging for a GPU.

larry ellison

Musk also has his own difficulty reference system: GPUs are now even harder to get hold of than drugs. 」

Since the rise of the new wave of AI, the lack of chips has become a theme that has continued to this day, and it will have to be maintained for some time.

When asked by Wired about the lack of AI GPUs, Huang bluntly said:

I don't think we're going to be able to catch up this year. Not this year, maybe not next year.

It's no wonder Zuckerberg was especially proud when he said last month that by the end of the year, we would have 350,000 Nvidia H100s.

After all, you have to have computing power to have a place in the AI war.

Obviously, no tech giant likes to hand over its lifeblood to another company, and self-developed chips have become the new battleground.

At the end of February this year, Microsoft announced a partnership with Intel to produce self-developed chips using Intel's 18a manufacturing technology, and showed that the company has been working to strengthen chip design capabilities.

Google's TPU has already achieved TPU v5P in December last year. Compared to TPU V4, TPU V5P has 2x more floating-point operations, 3x higher memory capacity, and 2x faster LLM training8 times.

Even OpenAI's Sam Altman is rumored to be raising trillions of dollars to reshape the global semiconductor industry.

Another up-and-comer that can't be ignored has also made its debut this year.

A start-up called Groq has built an AI chip LPU that claims to have achieved the strongest inference on the surface

LPUs have 10 times the inference power of GPUs at one-tenth the cost; The generation rate of running large models is close to 500 tokens per second, which is much higher than GPT-35 tokens per second for 40.

In the face of oncoming provocations, Huang had his own responses.

One is implicit and the other is explicit.

Jonathan Ross, CEO of Groq, said in an interview that many customers reveal the fear of being caught by Nvidia when they come to them

A lot of the people we met said that if Nvidia heard that we had a meeting, they would immediately deny it.

The problem here is that you have to pay for Nvidia a year in advance, and then you may not get the hardware until a year later or more. And then the situation is, oh, you're buying it from somebody else, so I think it's probably going to have to wait a little longer. 」

While it's uncertain whether Nvidia would actually do this, the degree of dependency is conceivable.

jonathan ross

Nvidia hasn't revealed much about the rules for inventory allocation, and Huang recently said that they are trying to allocate resources more fairly but will avoid selling chips to people who don't need them right away.

At the same time, Huang also said that Nvidia will introduce its customers to cloud computing companies and allocate a certain amount of chips.

In terms of explicitness, NVIDIA is also constantly promoting related technologies to keep itself ahead.

In the interview, Huang revealed that the company is now working on a brand new product:

We're creating a whole new kind of data center, which we're calling an AI factory.

In an existing data center, many people share a cluster of computers and store files in a large data center.

The AI factory is more like a generator. Very unique.

We've been building it for the last few years, but we have to turn it into a product now.

Huang is confident that this new product, which has not yet been named, will be ubiquitous in the future and will be needed by all companies.

All car companies will have a factory to build cars – to build actual products, at the atomic level – and there will be factories to build AI for cars – electronic grade.

In fact, you see Elon Musk doing this. He was much ahead of most of us in thinking about the future of industrial companies.

On top of that, the vision that has shaped Nvidia where it is today is also playing a role in finding the next frontier that could change everything like the AI gold rush.

Kingmaker or best buddy ?

When AI was still a poor student who was disliked by capital, Nvidia was determined to become an AI company with a unique vision.

In 2016, Huang personally delivered the world's first DGX-1 to OpenAI, where Musk was a founding member at the time.

In the following period, the iteration of hardware development and the accumulation of software by CUDA gradually formed NVIDIA's moat in the field of AI.

Today, Nvidia is investing extensively in startups to find that little bit of starlight that will shine in the future.

According to reports, Nvidia invested in more than 30 startups last year, and each investment had to be signed off by Huang himself.

Although he has the computing resources that AI startups crave, Huang doesn't like to be called the king maker

We invest in these companies because they are doing well in their field. We feel honored to invest in them, not the other way around.

Those are the smartest people in the world. They don't have to rely on us to endorse them.

Biotechnology is one area of particular focus for Huang. Recently, he also attended a seminar on health investing:

Biologists and scientists, this is an angry audience.

In the tech world, we use the words "create, improve, accelerate," but you love the words "target" inhibition.

In addition to joking to warm up, he also talked about cryo-electron microscopy techniques, X-ray crystallography and structures, and finally did not forget to call for cooperation:

I didn't mean to be pretentious, but do you think there is another CEO of a chip company who can talk like that? We have algorithms, we have experts in math and we can be your partners throughout the drug discovery process, so be sure to contact us.

Jacob Berlin has experienced NVIDIA's support firsthand.

He is the CEO of Terray Therapeutics, one of the NVIDIA-backed biotech companies. The company's algorithm was initially considered good and usable in his own opinion.

When NVIDIA started to join, Terray Therapeutics, with new computing power and engineering support, decided to just retrain the model once

We're doing much, much better right now.

If we hadn't worked with Nvidia and didn't have their support, we wouldn't have been able to do it ourselves.

Speaking of which, everyone will think, is it possible that the companies invested by NVIDIA will be given priority to get GPUs?

Not to mention the public, entrepreneurs who have received investment have also tried to ask weakly. Although I knew that there would probably not be, Imbue AI CEO Kanjun Qiu still asked the export: Can we get the goods first? 」

Nvidia has a different solution. Huang explained

They, like everyone else, have to deal with shortages ......What they get is our AI technical support, which is our engineering capabilities and the special technology that we use to optimize AI models.

We'll help them become more efficient. If you increase your throughput by a factor of five, you actually get five extra GPUs.

In addition to technical support, some of these entrepreneurs have received another enviable privilege — direct contact with Huang.

Imbue AI, which aims to build AI that can think like a human, has Huang as a mentor even though the CEO didn't get extra GPUs.

She had previously emailed Huang to ask for advice on compensation for her company's executives, and Huang was quick to get back to her.

He was more than willing to respond. I also tried not to bother him, but he was definitely there.

Even if not every investee company has this treatment, Huang's investment is still quite rare.

In a previous interview on the podcast show Acquired, Huang had a very popular answer. At the time, Huang was asked if he would choose to start a business if he were 30 now. What kind of company will you create?

Huang's answer: No. 」

In his opinion, starting a business is really difficult, and if he had known that it would all be so difficult, he might not be willing to go through these pains and hardships again. He thinks that entrepreneurship really depends on ignorance to have courage

When I meet with entrepreneurs now, they tell me that everything is easy, that I will support them and that I will not punctuate their fantasy bubbles.

But I know in my heart, oh, it's all different than you think. 」

I believe that those entrepreneurs who are fortunate enough to chat with Huang also hope that they will one day become the openai of the future

Obviously, they have a strong track record of success, and they can always choose verticals that they think will grow and take off, driving their demand for chips and software.

Now, they're trying to carve out another multi-billion dollar vertical for themselves.

Related Pages