According to reports, OpenAI CEO Sam Altman has unveiled plans to raise an astounding $5 to $7 trillion to establish a network of fabs aimed at producing a sufficient quantity of chips for artificial intelligence (AI) applications.

This ambitious proposal has sparked considerable discussion.

Following Altman’s announcement, Nvidia CEO Jensen Huang stressed the importance of architectural innovation in AI processors over sheer quantity, asserting that technological advancement would help control the costs of Artificial Intelligence.

This viewpoint resonated with the esteemed Jim Keller, renowned for his pivotal roles as Chief Architect at various organizations including AMD, Apple, and Tesla. Keller’s profound grasp of both hardware and software aspects of computing has been pivotal in fueling innovation and shaping the trajectory of technology.

Notably, his work on AMD’s K8 microprocessor, the first to utilize the x86-64 instruction set, laid the groundwork for Keller’s reputation as a visionary in the field.

Jim Keller’s return to AMD to spearhead the design of the Zen architecture marks a crucial turning point in the company’s trajectory. Once on the brink of bankruptcy, AMD has experienced a remarkable revitalization under Keller’s leadership. With the success of Zen, AMD has surged back into the spotlight, asserting dominance in the CPU field and reigniting fierce competition in the market.

Keller’s visionary leadership and unmatched expertise have consistently reshaped the competitive landscape across various ventures. His potential to propel Intel to greatness was evident when he joined the company as Senior Vice President in April 2018.

However it was not to be as internal conflicts and a toxic work environment revolving around Murthy Renduchintala’s bid for the vacant CEO position led to Keller’s abrupt departure from Intel just two years later.

Jim Keller departed Intel to pursue new opportunities and take control of his own destiny. In January 2021, he assumed the roles of President and CTO at Tenstorrent, charting a new path in his career journey as they aim to produce groundbreaking AI and HPC processors.

In response to Altman’s tweet suggesting raising the fundraising amount to 8 trillion, Keller wrote in a quote tweet, “I can accomplish it for less than $1 trillion.”


Altman’s ambitious plan to raise trillions of dollars to produce enough AI processors for all workloads and emerging AI companies is indeed ambitious. However, it entails a radical expansion of the semiconductor supply chain, potentially resulting in overcapacity at foundries and the devaluation of these processors in the market.

Rather than focusing solely on increasing the quantity of AI chips, both Keller and Nvidia CEO Huang argue that the emphasis should be on enhancing the sophistication of these processors. Additionally, they advocate for simplifying the supply chain of AI hardware to reduce the costs associated with AI servers and other devices.

“Start by eliminating the margin stacking,” Keller suggested, referring to the additional costs or profit margins added by each participant in the supply chain to deliver a product to the end user. “There are two to three layers. Then, make chips significantly faster so that the hardware matches the software. That is a challenging but achievable goal.”

Tenstorrent has an ambitious roadmap focused on developing processors for AI and HPC applications. Each AI processor in the roadmap increases the number of processing units, with the units becoming more advanced, thereby enhancing performance efficiency.

Tenstorrent’s upcoming flagship product, Grendel, is poised for release later this year. This multi-chiplet solution comprises an Aegis chiplet, featuring high-performance Ascalon general-purpose cores, paired with chiplets housing Tensix cores tailored for ML workloads.

Tenstorrent has expanded its lineup to include five distinct RISC-V CPU core IPs, spanning from two-wide to eight-wide decoding, providing flexibility for its processors or for licensing to interested parties.

For customers requiring basic CPU functionality, the company offers small cores with two-wide execution. Conversely, those in need of enhanced performance for edge computing, client PCs, and high-performance computing can benefit from the six-wide Alastor and eight-wide Ascalon cores.

The out-of-order Ascalon core, equipped with eight-wide decode (RV64ACDHFMV), boasts an impressive configuration featuring six ALUs, two FPUs, and two 256-bit vector units, making it a robust option.

In contrast, contemporary x86 designs commonly employ four-wide (Zen 4) or six-wide (Golden Cove) decoders, underscoring the remarkable capabilities of Tenstorrent’s core. The aim is to deliver industry-leading performance per watt, particularly in integer math operations.

Depending on Tenstorrent’s business requirements and financial resources, they may choose to employ a 3nm-class process technology for an AI chiplet, leveraging enhanced transistor density and increasing the Tensix core count.

Alternatively, they might opt to continue using the Black Hole chiplet for AI tasks, potentially augmenting it with 24 SiFive X280 cores as necessary. The chiplets will seamlessly communicate via a high-speed 2TB/s die-to-die interconnect.

Nevertheless, the demand for AI performance is rapidly escalating, and only time will reveal whether Tenstorrent and other industry players can keep pace with this burgeoning demand in the foreseeable future.

Altman’s ambitious fundraising goal of $5 trillion to $7 trillion significantly exceeds the current valuation of the worldwide semiconductor industry. With sales totaling $527 billion last year and projected to reach the $1 trillion mark by 2030, it’s clear that the proposed funding far surpasses industry norms.

Simply pouring money into solving a problem may not always be the wisest approach, especially considering the immense energy and materials required to produce and power vast volumes of AI clusters. The energy consumption associated with artificial intelligence far exceeds that of cryptocurrency mining at its peak, yet we never complaints about it.

Furthermore, chip manufacturers allocated $99.5 billion for chip fabrication equipment in 2022 and are projected to invest $97 billion in fabrication tools this year.