Udabur Investment：The mysterious acquisition of Nvidia hides a good business？｜ Jiazi Guangnian

The independent commercialization prospects are unknown on the GPU cost reduction.

Author ｜ Tian Siqi

After Meta, Google, Amazon and other technology giants have officially announced their own research chip plans, is it necessary for industry overlord Nvidia to be nervous?

Between the changes, Nvidia is still promoting its strategic layout in a low -key.Judging from its latest acquisition, Nvidia seems to know the philosophy of "being safe and dangerous".

A few months after rumors, Nvidia finally acquired the Israeli AI startup Run: AI in the blog at the end of April.This is a GPU -oriented supplier based on Kubernetes.According to media reports, the acquisition price is about $ 700 million.Udabur Investment

At the same time, Nvidia was revealed to buy another Israeli AI startup DECI, but the transaction amount was unknown. Only the Israeli media released the news that the price was $ 300 million.

The total $ 1 billion acquisition form is enough to confirm Nvidia's most concerned field at present: cost optimization and efficiency improvement.In order to cope with the growing market demand and competition, Nvidia is trying to build a more efficient and scalable AI ecosystem.

As the energy crisis brought by AI power consumption is increasingly urgent and energy consumption -it has become the core of no dispute between the GPU battlefield.So can the GPU cost reduction and efficiency, can it really be a good business?

1. Different tracks of GPU ecology

Both of the above two Israeli companies are committed to reducing the cost of artificial intelligence, but their operating models are different.

According to Nvidia Blog, artificial intelligence deployment is becoming more and more complicated, and customers' workloads are distributed in the infrastructure of clouds, edges and local data centers.Management and arranging generatory AI, recommendation systems, search engines, or other workloads require complex scheduling in order to optimize the performance of the system and underlying infrastructure.

Run: The open platform built on Kubernetes on Kubernetes can gather GPUs or sharing power for independent tasks -this includes a mode of only occupying some GPUs, or running multiple GPUs on different clusters.Corporate customers can better manage the GPU cluster of the data center, and achieve more efficient GPU cluster resource utilization.

This startup was established in 2018.Before being acquired by Nvidia, it completed Series C financing in March 2022, and the total funds raised were $ 118 million.

The two founders Omri Geller and Ronen Dar met at the School of Electrical Engineering, the University of Tellaviv.CEO Geller worked at the technical department of the Israeli Prime Minister's office, and Chief Technology Officer Dar was the Bell Lab and Apple.

"Since 2020, Run: AI has been closely cooperating with Nivine. We are keen to help customers make full use of their infrastructure."

HPCWire article in the field of high -performance computing is that Nvidia has its own Kubernetes plug -in, but Run.ai will bring more fine control to AI container management and arrangement.Therefore, Nvidia is willing to use this tool more than relying on the configuration of cloud suppliers.It can also help Nvidia with demand for complete software stacks.

In March of this year, Nvidia just launched the reasonable microservices (NIM) platform for pre -configuration and optimization of the deployment of AI models.After the acquisition of Run: AI, Nvidia can also use its Kubernetes arrangement to manage the deployment of NIM in its GPU infrastructure.

Image source: DECI

Another company, a company in the report, was reported by Nvidia. The business focus is to adjust the AI ​​model to make it easier to access and expand to improve the construction and deployment of AI models.The company established four years ago has raised $ 55 million so far.

Deci co -founder and CEO Yonatan Geifman said: "Our technology can automatically design new neural network structures, and optimize itIt runs faster on the hardware.

Israeli media Globes reported that Microsoft has integrated DECI into Azure Ai Studio, where it must not only face up with Openai, but also compete with Meta and French company Mistral and Nvidia.

"Today's trend is the collaborative integration of software and hardware," said Gavanman, "Some companies also hope that we can help them design the next -generation chips. They want to understand how our algorithm views will improve their next -generation products."

When it comes to the origin of DECI, Gavman recalled his student days and the time of the headquarters of Google Mountain."During the doctoral degree from the Israeli Institute of Technology, I developed many deep learning models. In the process of building these models, I found that many trial and error iterations need to be performed, and each iteration needs to consume a lot of computing resources to converge to one to oneA good model solution, "he said.Later, after entering Google's work, he more strongly felt the necessity of scalability, running time and memory such as the model in the production environment.

Similar companies also have open source artificial intelligence platform Lightning AI.In March of this year, it collaborated with Nvidia a new generation of AI compiler called "Thunder", which aims to accelerate the training speed of the AI ​​model.

"We found that customers did not make full use of the existing GPUs, but invested more GPUs to solve the problem," said Luca Antiga, chief technology officer of Lightning AI."Thunder and LightNing Studios and other analytical tools can make customers effectively use GPUs when running larger and faster models. As the AI ​​industry starts training cross -multiple models and stronger capabilities, to obtain the best and best acquisitionThe performance of GPU becomes crucial. "" "."

Zhang Binlei, the research director of Xinmou Research, summarized the "Jiaziguang Years" that Run: AI is essentially cooperating with GPUNew Delhi Investment. In the GPU, it is to allocate tasks to distribute, coordinate GPU's resource computing power, and perform multiple tasks.Application -side products are measured and evaluated when obtaining GPU space.

However, the independent commercialization prospects of such companies are unknown.Zhang Binlei pointed out that Run: AI relies on GPU enterprise applications, and DECI may obtain more startups.Because the future software is different from the small programs on the current mobile phone, it may be extremely power -consuming large AI models, which requires more vigorous services for DECI.

2. Nvidia's "shopping strategy"Mumbai Wealth Management

Many people may not realize that Nvidia may be the most active investors and sellers in the current AI entrepreneurial field, and even none of them.Pitchbook data from venture capital statistics agencies show that Nvidia has invested in more than 30 startups in the past year.

According to people attending the demonstration, a few months before the completion of these two transactions, Nvidia asked at least one large venture capital company to organize 12 "promising" AI entrepreneurial enterprises to concentrate on demonstration to Nvidia executives. CEOHuang Renxun is also included.

Huang Renxun, CEO of Nvidia, Picture Source: CBS 60 Minutes

This is not Nvidia for the first time to "purchase" in Israel.In 2019, Ningwei Dahao tossed $ 6.9 billion to acquire Israeli chip manufacturer Mellanox, becoming the company's largest acquisition in Israel to date.

Thanks to Intel's investment in Israel since the 1970s, and the local attention to engineering and technical talents, although Israel has no wide -ian giant companies, it has a large number of stealth champions in semiconductors and artificial intelligence fields and a good talent environment in the fieldEssenceAt present, Israel has at least more than 3,000 Nvidia employees, accounting for about 10%of the company's total.Run: AI employees will also join Nvidia's Israeli office.

As for the pursuit of GPU cost reduction and efficiency, it may be a measure of Nvidia to find a new income growth engine.

Earlier, Nvidia's main customers were the technology giants with strong financial resources to buy their expensive GPUs.Gil Luria, an analyst at DA Davidson Investment Bank, said that in the second half of 2023, orders from Microsoft and Meta accounted for about a quarter of Nvidia sales.

However, large manufacturers such as Google and Meta have officially announced their own AI chips, hoping to reduce dependence on Nivine.

Pierre Ferragu, an analyst of New Street Research, said that although Nvidia sold 2.5 million chips last year, Google spent 2 billion to 3 billion U.S. dollars to create about 1 million self -developed AI chips.Amazon is expected to spend 200 million US dollars to produce 100,000 dollars.Microsoft also said at the end of 2023 that it started testing its first AI chip.

The research company Gartner said that by 2027, the artificial intelligence chip market is expected to double to more than $ 140 billion.Well -known chip manufacturers such as AMD and Intel, as well as startups such as CEREBRAS and Sambanova, are also making special artificial intelligence chips.

After acquiring the two companies mentioned above, Nvidia can manage computing resources more effectively, reduce energy consumption, provide customers with more cost -effective platforms, develop to the small and medium -sized enterprise market, and also attract the residence of large customers.

Zhang Binlei said: "The key is not how expensive the price of the chip is, because this is a one -time costGuoabong Stock. The important thing is the energy consumption of long -term operation. If the energy consumption is much worse than others, then the self -research is a long -term harm for the enterprise.. In the GPU track, only the best, no better, that is, the strong. "

For other technology giants, Zhang Binlei said that it is not difficult to make its own products to replace Nvidia.However, from the perspective of commercial costs, the best chip to purchase must be the most cost -effective.Therefore, what Nvidia has to do is to make the chip ecology the best and greatly lead other companies.Provide customers with the best services, the lowest cost, and lowest energy consumption at all levels of energy efficiency, ecology and application.Other technology giant customers will lose their motivation to develop the GPU.

It is worth mentioning that in mid -April, Nvidia's stock price had encountered a significant decline, which caused a short concerns about the market's "AI bubbles".

But the market panic is fast and fast, and Nvidia's stock price has quickly recovered from $ 750 per share to about 900 US dollars.Zhang Binlei pointed out that in the current AI wave, Nvidia's stock market performance is the same as that of explosive AI products.After the release of AI products, Nvidia's stock price rose all the way.And the stocks falling significantly is just the corresponding emotional decline.

At this stage, Nvidia's chip prices are temporarily maintained at a high level, with gross profit of 70%and a net profit of about 50%.Zhang Binlei believes that if other technology giants can develop their performance, it is certainly valuable for self -research.But then again, if there are some companies in the market that can reach or approach Nvidia, Nvidia can also play price wars to make market explorations and continue to maintain a competitive advantage.

On April 30, AMD, one of Nvidia's main opponent, released the first quarter financial report, raising the annual sales forecast of the MI300 AI accelerator chip this year from $ 3.5 billion to $ 4 billion at the beginning of the year.But investors are still dissatisfied with this number because it is estimated to reach $ 6-8 billion.The disappointment caused AMD to fall by 7%after the day.

In this regard, Zhang Binlei also believes that energy efficiency is a very important factor, because companies consider long -term expenditure. If energy consumption is higher than others, the level of technology is more backward than others, and the ecology is not perfect, and sales will be difficult to meet expectations.Only when the price of the chip is appropriate and the service is better than the competition, can the opportunity to seize the market.

"NVIDIA TO Buy Two Israeli Startups That Make ITS AI Chips Cheaper To USE", "The Information"

"How Nvidia Could Use $ 700M Run.ai Acquisition for Ai Consumption", HPCWIRE

"Optimizing AI With AI", "Calcalist"

"NVIDIA’ s Big Tech Rivals Put their Own A.I. Chips on the table "," The New York Times "

"NVIDIA to Buy Israeli Deep Learning Co Deci AI -Report", "Globes"

(Cover map source: Analytics India Magazine)

Jaipur Investment