Breaking China's longstanding passive chip following situation.

Written by | Huashang Taolue Zhang Jingbo

June 2015, No. 6 South Road, Beijing Academy of Sciences.

A small group of young enthusiasts crammed into a dimly lit room, nervously awaiting the results of their chip test. When the computer screen displayed the correct results, their excitement knew no bounds:

The world's first AI chip was about to be born!

[No Longer Following the Crowd]

Outpacing Google, Amazon, and overshadowing Intel.

In 2024, NVIDIA's share price soared, making it the first semiconductor company in history to surpass a market value of 3 trillion dollars. However, there is a Chinese AI enterprise that has surpassed even NVIDIA in terms of ambition: Cambrian.

From 2023 to the present, NVIDIA's share price has increased by 9 times. During the same period, Cambrian's share price peaked at over 10 times its initial value.

Despite revenues amounting to only a billion yuan, a fraction of NVIDIA's, and consecutive years of heavy losses, Cambrian's share price performance puzzled many and sparked exclamations of a market bubble.

Cambrian's fans view it as China's answer to NVIDIA and the prime beneficiary of domestic substitution. However, both perspectives overlook Cambrian's technological prowess and its significant contribution to the global AI chip industry.

In fact, Cambrian was one of the earliest pioneers in the field of AI chips.

As early as March 2014, Chen Tianshi and Chen Yunji, brothers from the Institute of Computing Technology at the Chinese Academy of Sciences, published a groundbreaking paper at the prestigious ASPLOS conference:

DianNao: A Small-Scale High-Throughput Machine Learning Accelerator.

In this paper, the Chen brothers introduced a dedicated processor architecture for deep learning, offering an energy efficiency improvement of over 100 times compared to traditional CPUs, thereby breaking through the efficiency bottlenecks of general-purpose processors in AI tasks.

Later that year, they published another DianNao paper at the MICRO conference.

These two papers garnered significant attention, receiving thousands of citations and becoming landmark achievements in global AI chip research, inspiring both Google and NVIDIA.

The American magazine Science published an article praising it as a disruptive development outside of Silicon Valley.

Drawing on this paper, the Chen brothers founded Cambrian, turning theoretical ideas into tangible reality.

When this theoretical concept collided with large AI models, a chain reaction ensued: Cambrian's share price skyrocketed, pushing its market value beyond 250 billion yuan.

Chen Tianshi, a mere scholar, became a billionaire with a net worth of 70 billion yuan, ascending to the pinnacle as Jiangxi's newest richest man.

CCTV named the two brothers as the 2017 Annual Technology Figures, emphasizing their role in breaking China's longstanding passive chip following situation.

In 2020, Stanford University released a list of the top 2% scientists worldwide, with Cambrian's founder and CEO Chen Tianshi ranking first in mainland China.

However, the widely cited AI chip research in the world today began as a modest research project with no funding, no published papers, and little interest!

[Pioneering Uncharted Territory]

In 2010, Chen Yunji, the chief architect of Loongson 3, met his younger brother Chen Tianshi in Beijing.

This meeting would alter the course of both their lives.

It was Chen Tianshi's final year as a Ph.D. student at the University of Science and Technology of China. While his wife was in Beijing, he frequently visited the Institute of Computing Technology at the Chinese Academy of Sciences under the pretense of research, but in reality, he was reuniting with his wife.

During this period, he stumbled upon his brother's Loongson project and impressed Hu Weiwu, the father of Loongson.

At that time, Chen Yunji was swamped with arduous chip testing tasks.

Amidst his hectic schedule, he constantly pondered how AI could automate chip testing. Coincidentally, his younger brother was engrossed in AI research at the University of Science and Technology of China.

One brother excelled in chips, while the other was adept at AI, making them a perfect team.

Invited by Hu Weiwu, Chen Tianshi also joined the Institute of Computing Technology after graduation. The brothers reunited, discussing research topics whenever they had the chance.

After much deliberation, they decided to undertake a project greater than AI-assisted chip design:

Developing AI chips to accelerate all deep learning algorithm programs with a single chip.

However, this idea seemed like a fairy tale at the time. In 2010, AlphaGo had yet to defeat Lee Sedol, and NVIDIA was still primarily a graphics card manufacturer.

Globally, there were no similar papers or projects.

Due to its unpopularity, they couldn't even secure funding of 200,000 yuan for their research. Their students were also skeptical, with more than one advising them to abandon such an elusive pursuit.

Years later, recalling those years, Chen Yunji still felt emotional:

Pushing ahead alone in an unpopular interdisciplinary field unrecognized by academia, disregarded by industry, and challenging to publish papers and secure funding was akin to navigating in the dark.

At a critical juncture, the leadership of the Institute made a decisive move.

At the end of 2010, during an internal meeting, the Chen brothers presented their vision for developing AI chips. This idea resonated with the Institute's leaders.

To this end, the Institute specially established an intelligent processor team and appointed Chen Yunji as the head of the research group.

With the support of the leadership, the brothers began to let go of their burdens and go all in. They reconstructed the chip architecture from scratch with a new instruction set.

Finally, in 2014, the two famous DianNao papers were published.

The Chen brothers used these papers to demonstrate the enormous potential of deep learning processors and won the Best Paper Award at the prestigious MICRO conference.

This was the first time since 1963 that scholars from a country other than the United States had received this honor.

▲Chen Yunji (left) and Chen Tianshi brothers

The papers garnered significant attention, but their significance would have been limited had they remained theoretical.

Director Sun Ninghui decided to take things further. He summoned the brothers to his office and inquired, 'Can you create a real deep learning processor chip?'

Chen Yunji, who had worked on Loongson at the Chinese Academy of Sciences for over a decade, was well aware of the challenges involved in transforming a design into a tangible chip.

However, the excitement of pioneering uncharted territory overcame their fear of difficulties.

So, in a small, dimly lit room of less than 30 square meters, Chen Yunji and his team worked tirelessly day and night.

Army cots were placed beside workstations, and it was common to pull all-nighters, transitioning from workstation to cot. Every test was filled with anxiety and anticipation.

'We are far from as glamorous and magical as people outside imagine,' Chen Tianshi later remarked.

After hundreds of days and nights of relentless effort, they finally succeeded in tape-out in 2015.

That night after the tape-out, everyone crammed into the small, dimly lit room, eyes glued to the computer screen. When the test results popped up correctly, their excitement knew no bounds.

This chip carried the Chen brothers' high expectations, and they named it Cambrian, hoping it would be like the great explosion of life 500 million years ago.

Taking this chip as a starting point, with the support of the Institute of Computing Technology and venture capital funds, Cambrian was officially established a year later.

Younger brother Chen Tianshi, due to his composed demeanor and strategic decision-making skills, took on the role of CEO. Older brother Chen Yunji preferred working behind the scenes in research and served as the company's chief scientist.

[No Shortcuts, No Surprise Attacks]

Cambrian's founding coincided with a significant trend of the times.

In the spring of 2016, Google's AlphaGo became an overnight sensation, and AI quickly sparked a global wave, with numerous startups springing up like mushrooms after rain.

At this juncture, Chen Tianshi faced two paths:

One was to design their own chip, but the development cycle was too long, and any misstep could lead to failure.

The other was to design a chip IP and license it to related hardware manufacturers, thereby achieving commercialization and large-scale production at the fastest pace.

The cautious Chen Tianshi chose the latter.

In 2017, Huawei and Cambrian joined forces, integrating the Cambrian 1A processor into the new generation of Kirin 970 chips as its core AI processing unit.

This was the world's first AI mobile chip.

At a critical moment, Chen Tianshi secured Huawei as a major customer, making Cambrian the most successful AI startup.

However, there is an inherent gap between conducting research and developing a product for commercialization. Chen Tianshi, with a technical background, was initially unprepared for this disparity.

While the chip IP licensing business starts quickly, it has a low ceiling and fragile cooperative relationships.

In fact, Huawei's collaboration with Cambrian lasted just over a year before it terminated due to the former shifting to self-research. Over-reliance on a single major customer put Cambrian in a precarious position.

At a crucial juncture, Chen Tianshi made an important decision:

Expand from a single smart terminal chip to full-scenario smart chips for cloud, edge, and terminal, and actively pursue government affairs and intelligent computing cluster projects.

Starting from 2018, Cambrian developed a series of cloud-based inference and training chips, along with edge computing product lines, and built a unified basic system software and tool chain for all products.

By 2020, Cambrian had evolved into a chip design company with full-stack system capabilities encompassing both hardware and software.

In July of that year, Cambrian was listed on the STAR Market. On its first day of listing, it surpassed a market value of 100 billion yuan, becoming the first domestic AI chip stock.

Listing on the A-share market temporarily relieved Cambrian's funding woes.

However, what Chen Tianshi didn't anticipate was a bigger crisis looming ahead, one that nearly doomed Cambrian.

At the end of 2022, the U.S. government unilaterally included Cambrian on the Entity List for export controls. This meant that TSMC would no longer manufacture chips for Cambrian.

For a chip design company, this was akin to a death sentence.

Affected by this, Cambrian's revenue plummeted. Yet, Chen Tianshi withstood the pressure of external doubts about Cambrian's chronic losses and insisted on investing in R&D.

In Chen Tianshi's view, chips are a marathon with no shortcuts. He compared chip research and development to a bulldozer traveling along the main road, taking no shortcuts and no surprise attacks.

'I don't have that kind of passionate slogan. I believe that strategy is to work hard and move bricks honestly,' he said.

He also benchmarked global chip giants Intel and NVIDIA, whose predecessors had endured decades of hardship. In comparison, Cambrian was not even a decade old, just a fledgling enterprise, and the marathon had only just begun.

This patience eventually paid off.

Entering 2024, with the proliferation of large AI models and the maturation of the domestic semiconductor industry chain, especially the commissioning of the 7-nanometer production line, Cambrian broke free from constraints and began to accelerate its growth.

[From China's Talent Class to Jiangxi's Newest Billionaire]

Chen Tianshi and Chen Yunji were born in the 1980s in Jiangxi. Their father is a power engineer, and their mother is a middle school teacher.

This seemingly ordinary family nurtured two geniuses from the University of Science and Technology of China's Talent Class. Chen Yunji later summarized his parents' educational approach with four words:

'Fishing by flooding the waters.'

Unlike coercive education, Chen Yunji's parents fostered the scientific interests of the two brothers through guidance rather than force when they were young.

The bookshelves at home were filled with various books spanning fields like engineering and history, allowing them to freely choose what to read.

In particular, a set of Why? books sparked the curiosity of the two brothers. Driven by this curiosity, older brother Chen Yunji was admitted to the University of Science and Technology of China's Talent Class.

Younger brother Chen Tianshi was equally competitive. We play together every day; you're not smarter than me. If you can get in, so can I.

So, he followed in his brother's footsteps and was also admitted to the Talent Class at the University of Science and Technology of China.

Besides studying, the two brothers were avid video gamers. It is said that one of the reasons Chen Tianshi worked hard to enter university was because his brother once told him:

'After entering university, parents won't be able to control you anymore, and it will be more convenient to play games.'

After enrolling in the Talent Class at the University of Science and Technology of China, Chen Tianshi indeed enjoyed gaming freedom. However, his obsession with gaming led to an academic crisis, and he narrowly avoided failing a course in his sophomore year.

Older brother Chen Yunji also loved gaming and even took StarCraft as a main course.

Despite this, Hu Weiwu, the father of Loongson, overcame objections and recruited the two brothers under his wing. In Chen Yunji's own words:

Prior to this, the Institute of Computing Technology, Chinese Academy of Sciences, had never recruited students with such subpar undergraduate grades.

What truly captivated Hu Weiwu was the brothers' passion for computers and their unwavering commitment to hard work.

The Institute of Computing Technology also provided these two prodigious young men with an ample stage and space. In an era when many were still unfamiliar with AI, it staunchly supported their exploration into AI chips.

Without such a conducive environment, the Chen brothers would not have been able to achieve the Cambrian miracle.

Although it has temporarily surmounted production hurdles and business is gradually rebounding, Cambrian's ongoing challenge has always been how to attain better commercialization.

In 2024, Cambrian's anticipated annual revenue stands at 1 to 1.2 billion yuan, a fraction of NVIDIA's revenue.

This glaring disparity has prompted many to speculate whether Cambrian has been merely storytelling, with its performance failing to justify its share price, which has consequently led to a recent sharp decline in its share price.

Beneath this disparity in strength lies not just a technological divide but also an ecological barrier.

Today, with the ascendancy of large AI models, NVIDIA's CUDA ecosystem has established a monopoly, whereas Cambrian's software tool chain is still in its nascent stages.

Chen Tianshi has a keen understanding of this gap.

Over the years, Cambrian has been striving to compensate for its ecological shortcomings and, in the process, forged its own path—focusing on underlying hardware and tool chains while steering clear of applications.

Regarding this, Chen Tianshi explained: "Once you delve into applications, you inadvertently encroach upon customers' core interests."

"Cambrian serves as a platform, specializing in providing the foundation. Everyone is welcome to utilize us for applications," he said.

Commercialization transforms a person.

A decade ago, Chen Tianshi was a humble student engrossed in his studies at the Institute of Computing Technology. Today, he presides over a leading industry worth billions of yuan, with a personal net worth of 70 billion yuan, ascending to the pinnacle as Jiangxi's newest billionaire.

This meteoric rise in wealth has not altered Chen Tianshi's unpretentious and competitive spirit.

Although in interviews, he repeatedly emphasizes that he is not an ambitious individual, the original aspiration of utilizing AI chips to accelerate all deep learning algorithms has consistently fueled him and Cambrian.

"If the era of intelligence truly dawns, it's hard to fathom that new titans won't emerge in the realm of intelligent chips. I won't conceal the fact that Cambrian aspires to be that titan," he said.

However, in reality, with billions in revenue but consecutive years of losses, Cambrian Innovations is grappling with financial constraints.

Particularly when confronted with NVIDIA, a near-monopolistic global giant, Cambrian Innovations' prospects of victory appear slim. Chen Tianshi has a long and arduous journey ahead if he is to realize his dream.

【References】

[1] "Pioneering the 'Cambrian Era'" - China Science Daily

[2] "Dialogue with Chen Tianshi of Cambrian Innovations: The Long Journey to a Great Chip Company" - Jiazi Guangnian

——END——

Welcome to follow [Huashang Taolue], discover influential individuals, and delve into tales of strategy.

All rights reserved. Reproduction without permission is prohibited.

Some images are sourced from the internet.

If infringement is suspected, please contact us for removal.