OpenAI Secures $51 Million in Chips from Rain AI to Fuel AI Revolution

OpenAI’s power struggle has just ended, and a key deal has quietly surfaced.

According to foreign media “Wired”, when Sam Altman was the CEO of OpenAI, OpenAI signed a letter of intent worth US$51 million with Rain AI, promising to purchase Rain AI’s chips after they were launched.

Rain AI is an AI chip startup that aims to significantly reduce the cost of AI computing power. By developing an AI chip that mimics the way the human brain works, NPU, it provides “low-cost, energy-efficient hardware” for AI companies such as OpenAI and Anthropic.

The company said, “Compared to traditional GPUs, NPU will provide AI developers (such as OpenAI) with potentially 100 times the computing power and provide 10,000 times energy efficiency in training .”

Given that OpenAI has been suffering from a shortage of computing power, it is not difficult to understand that it would be willing to spend a lot of money to ensure the supply of chips required for its AI projects.

What are the characteristics of the chips developed by Rain AI? How did the company come to prominence? What does this investment reveal about the layout of Altman and OpenAI in the chip field?

01 “Brain-like” AI chip

The core product of Rain AI is the NPU, a “brain-like” AI chip based on neuromorphic technology. The chip is designed to process information with low consumption and high efficiency to meet the demanding computing needs of AI tasks.

It mimics the structure and function of the human brain, similar to the neural connections in the brain, and is built on a network of interconnected artificial synapses . This architecture allows the NPU to process information in a parallel and distributed manner, making it ideal for “computation-intensive tasks” in AI applications.

Moreover, Rain AI is the first to adopt the digital memory computing (D-IMC) model, which further improves the efficiency of AI processing, data movement and data storage.

In addition, Rain offers intellectual property (IP) licensing opportunities for digital in-memory computing tiles and software stacks that are specifically tailored for AI workloads on devices that require ultra-low latency and high energy efficiency, covering long-term A range of computing use cases for Long Reach Ethernet (LRE), including smart cars, smart watches, etc.

For its own products, Rain’s solgan is “redefining the limits of AI computing” and promotes “our AI accelerator achieves a record-breaking balance between speed, power consumption, area, accuracy and cost.”

Whereas, the “brain-like” chip (NPU) designed by Rain promises efficient and low-consumption operation, which is crucial to overcoming the “bottlenecks” associated with heavy-duty chips made by companies such as Nvidia and AMD.

Gordon Wilson, one of the founders of Rain AI, said bluntly on LinkedIn, “NPU chips will define the new AI chip market and significantly subvert the existing market.”

However, it is worth noting that although Rain AI claims to have better energy efficiency than Nvidia’s GPUs, Rain’s initial chips are actually based on the traditional RISC-V open source architecture supported by Google, Qualcomm and other technology companies . It is intended for use in so-called edge devices far away from data centers, such as mobile phones, drones, cars and robots.

However, most current edge chip designs, such as those used in smartphones, focus on the inference stage of neural networks. The goal of Rain is to provide a chip that can be used not only for model and algorithm training, but also for subsequent inference operations.

Currently, Rain AI has launched its first AI platform for AI reasoning and training. It also claims that the “brain-like” chip (NPU) will allow AI models to be customized or fine-tuned in real time based on the surrounding environment.

In this regard, Sam Altman has also publicly stated, ” This neuromorphic method can significantly reduce the cost of AI development and is expected to help realize true AGI. ”
It is reported that OpenAI hopes to use these chips to reduce the cost of data centers and deploy its models in devices such as mobile phones and watches. Then “brain-like” chips (NPU) are undoubtedly very attractive to OpenAI.

Still, these are just guesses. At present, it is still unknown how OpenAI will use the Rain chip.

02 Close connection with OpenAI

Rain AI was founded in 2017 to build a “low-cost” computing platform for future AI.
Rain AI has three co-founders, namely Jack Kendall, Gordon Wilson, and Juan Claudio Nino, who met at the University of Florida. In addition, it also hired OpenAI software engineer Scott Gray as a company consultant.

Currently, Rain AI has about 40 employees, including experts in AI algorithm development and traditional chip design.

Interestingly, Rain AI is also headquartered in San Francisco, less than a mile away from OpenAI.
The year after the company was founded, Rain AI received US$5 million in seed round financing from investors including the well-known startup incubator Y Combinator .

At that time, Altman was serving as the CEO of Y Combinator and also personally invested $1 million in Rain AI . A year later, OpenAI passed the chip purchase agreement worth $51 million.
As of April 2022, after a $25 million round of financing led by Saudi Arabia-affiliated fund Prosperity7 Ventures, Rain’s total financing reached $33 million and its valuation reached $90 million.

At the beginning of this year, the company “boasted” its progress to potential investors, saying that it expected to launch a “test” chip this month, which would mean that the chip design was completed and manufacturing could begin.

Rain AI also stated that the first batch of chips could be available to customers as early as October next year, and even emphasized to investors that it has entered into advanced negotiations with technology giants such as Google, Oracle, Meta, Microsoft and Amazon to sell systems to them. Microsoft declined to comment, and other companies did not respond to requests for comment.
In short, Rain AI is still in the development stage, and it’s unclear when it will be commercially available. Although the company’s “brain-like” chip (NPU) technology is promising and its backers are high-profile, it still faces many challenges.

03 Open AI’s ambitions

Regardless of whether Altman’s investment in Rain AI is selfish or not, the chip shortage is indeed a major problem facing OpenAI.

In fact, a year ago, less than a week after ChatGPT was released, Altman felt that the computing cost was “horrible.” After that, he publicly complained more than once about the “cruel shortage” and “eye-popping” cost of AI chips.

In May of this year, Altman reluctantly admitted, “OpenAI is experiencing a serious shortage of computing power, and many short-term plans have been postponed.”
As we all know, OpenAI leverages the powerful cloud services of its main investor Microsoft, but often turns off certain features of ChatGPT due to hardware limitations.

In this regard, Altman said, “The speed of AI progress may depend on new chip designs and supply chains.” After all, computing power is everything now.

In fact, Altman himself has invested in chips a long time ago. In addition to Rain AI, around 2021, he also participated in investing in Cerabras, an AI company with a chip as big as a plate that requires two hands to hold.

At the beginning of this year, Altman also paid attention to Atomic Semi, which was established by “Silicon Wizard” Jim Keller and “Silicon Prodigy” Sam Zeloof (by simplifying and shrinking semiconductor factories and integrated circuit prototypes to quickly manufacture affordable chips), and OpenAI Startup Fund also participated in the investment .

In addition, just weeks before Altman was fired from OpenAI, there was news that he was trying to raise billions of dollars to start a new chip company.

The details of the project are not yet known, except that it is codenamed “Tigris” and aims to compete with Nvidia in the field of AI chips .

It is reported that Altman raised funds in the Middle East for the “Tigris” project. The “coincidence” of the location makes people wonder if there is any connection between this project and Rain.

In addition, Altman has also held discussions with semiconductor executives, including chip design company Arm, to discuss how to design new chips as early as possible to reduce costs for large language model companies like OpenAI.

Moreover, not only Altman, OpenAI is also looking for the possibility of building large models at a lower cost, thereby getting rid of dependence on NVIDIA.

In addition to looking for chip suppliers like Rain AI and making external investments, OpenAI also began to try to develop its own chips some time ago, evaluate potential acquisition targets, and recruit hardware-related positions.

Not long ago, OpenAI appointed the former head of Google TPU, Richard Ho, as hardware director. It also hired many experts in compilers and kernels, and is recruiting “data center facility design experts.”

Richard Ho will lead OpenAI’s new division and help optimize partners’ data center networks, racks and buildings.

However, these forward-looking investment layouts are still difficult to solve the current GPU shortage problem. Currently, OpenAI still uses Nvidia chips on a large scale.
It has been observed that OpenAI is dynamically adjusting the capabilities of products such as ChatGPT to save computing power. This is not difficult to understand. Some netizens recently discovered that GPT-4 is easier to “be lazy” than GPT-3.5.

With the emergence of large models, people have begun to pay attention to the power consumption of large AI model data centers. Rain and other chip startups aim to reconfigure the way data is processed to reduce transmission requirements and reduce power consumption.

Google, Microsoft, AMD, Intel, Amazon, and startups such as Cerabras, Sambanova, and Rain have successively entered the field of AI chips. Will the market for AI computing power supply change in the future? And can OpenAI get rid of the situation where computing power is controlled by others? Judging from the ultra-long cycle of the chip, these problems will continue to exist for a long time.

error: Content is protected !!