Cerebras: The AI Chip Gem Valued at $60 Billion
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Cerebras: An Impressive Valuation in the AI Chip Sector
Cerebras Systems, now a public company, has recently made waves in the stock market, reaching a valuation of $60 billion. This company, specializing in the manufacture of chips for artificial intelligence, counts giants like OpenAI and AWS among its clients. The IPO not only propelled the company into the spotlight but also turned its co-founders into billionaires.
However, in 2019, Cerebras's future was far from assured. The company, then three years old, faced significant financial challenges, spending about $8 million per month. It struggled to solve a technical problem that many in the semiconductor industry deemed insurmountable.
A Major Technical Challenge
Andrew Feldman, CEO and founder of Cerebras, shared the difficulties the company faced at that time. Indeed, nearly $200 million had been invested without success to overcome a crucial technical obstacle. Each board meeting became a trial, marked by the announcement of new failures and significant expenditures.
Cerebras's core idea was to create a single, powerful chip from an entire silicon wafer. Traditionally, the microprocessor industry had spent over 50 years making CPUs faster and cheaper by integrating more transistors onto a silicon wafer and cutting them into smaller and smaller pieces. However, to meet the growing demands of AI, a different approach was needed.
The Complexity of the Mega-Chip
Realizing this vision proved to be extremely complex. Although the team succeeded in designing the mega-chip and manufacturing it with the help of TSMC, they encountered a major hurdle: packaging. This crucial process, which includes assembling the chip onto a motherboard and managing power and cooling, turned out to be a real puzzle.
Cerebras's chips were 58 times larger than those of their competitors and consumed 40 times more energy. No off-the-shelf heat sinks were available, and no manufacturing partner could assist them. The brightest minds in microprocessor engineering had tried for decades to build chips that were both this large and denser, but had failed. The Cerebras team had to invent a machine capable of simultaneously screwing in 40 screws to attach the wafer to a board without cracking it.
An Unexpected Success
After numerous trials and errors, the team finally overcame these technical challenges. In July 2019, the chip functioned for the first time, a moment that Feldman describes as one of the greatest of his life. This success is all the more remarkable given that the founding team had previously sold a cloud server startup, SeaMicro, to AMD for $334 million in 2012.
The day the chip finally worked also occurred about two years after OpenAI had discussed acquiring Cerebras, which Feldman confirmed. However, these discussions did not lead to an agreement due to disagreements among the founders of OpenAI, several of whom are angel investors in Cerebras.
Complex Relationships with OpenAI
Cerebras's success also attracted the attention of OpenAI, which considered acquiring the company. Although these discussions did not materialize, OpenAI became a key partner, lending $1 billion to Cerebras in exchange for warrants for 33 million shares. These shares, at a closing price of $279, are worth over $9 billion.
As part of this agreement, Cerebras committed not to sell its products to certain competitors of OpenAI, although this restriction is temporary. Feldman did not confirm that the obvious company involved is Anthropic. He explained that this limitation allows Cerebras to focus on its current capabilities before expanding to other clients.
A Measured Growth Strategy
Feldman compares selling AI computing capacity to an all-you-can-eat buffet. Rather than trying to serve all potential customers, Cerebras chose to focus on a portion of the market to better manage its operations. This cautious approach may well be the key to its future success as the company continues to navigate the ever-evolving landscape of AI technologies.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.