Jim Keller: I only need $1 trillion to do it

Latest update time：2024-02-18

Reads：

????If you want to see each other often, please mark it as a star???? and add it to your collection~

Source : Content compiled by Semiconductor Industry Observation (ID: i c b ank) from tomshardware, thank you.

There are reports that OpenAI CEO Sam Altman plans to raise US$5 to US$7 trillion to build a fab network to produce enough chips for artificial intelligence (AI) applications, which has attracted a lot of attention in the industry. Nvidia’s Jensen Huang said that architectural innovation in AI processors is more important than the number of these processors. And he also said that this money can buy all the GPUs in the world.

Now, legendary CPU developer Jim Keller, who now works on AI and HPC processors at Tenstorrent, said essentially the same thing. "I can do this for less than $1 trillion," Keller wrote in a tweet in response to Altman's tweet "f*ck why not 8," apparently referring to raising the fundraising amount to $8 trillion Dollar.

Altman's plan to raise trillions of dollars and build enough AI processors for all workloads and emerging AI companies is extremely ambitious. Still, it essentially involves a radical expansion of the semiconductor supply chain, which could lead to overcapacity in foundries and the devaluation of these processors in the market. But both Keller and Nvidia CEO Jensen Huang say these processors must become more complex, rather than making more chips for artificial intelligence. Additionally, the supply chain for AI hardware must become simpler to reduce the cost of AI servers and other equipment.

“It starts with eliminating margin pile-up,” Keller writes, meaning the added costs or profit margins each participant in the supply chain adds to deliver a product to the end user. "There are two or three layers. Then, make the chip faster and match the hardware to the software. That's harder, but doable."

Tenstorrent itself has a very ambitious roadmap involving the development of artificial intelligence and HPC application processors. Each AI processor on the roadmap has increased the number of processing units, and the units have become more advanced, improving their performance efficiency. However, the demand for AI performance is growing rapidly, so only time will tell whether Tenstorrent and other companies can catch up with it in the foreseeable future.

Altman's $5 trillion to $7 trillion funding target far exceeds the current valuation of the global semiconductor industry, which had sales of $527 billion last year and is expected to hit the $1 trillion mark by 2030. Meanwhile, chipmakers are investing $99.5 billion in chip manufacturing equipment by 2022 and are expected to allocate $97 billion to manufacturing tools this year.

Raising $7 trillion, the simplest part of Altman’s plan

OpenAI CEO Sam Altman's plan to reshape the global semiconductor industry envisions pouring money into solving challenges that are far more complex than money.

Chip manufacturing is a capital-intensive industry. It's also one of the world's most complex industries, with a history of wild cyclical swings that make companies wary of aggressive expansion.

It has taken decades for the world's most advanced chipmakers to reach their current heights. Some chip companies struggled during the industry's notorious downturns, such as the early and mid-2010s. Others have halted development of cutting-edge chips because of concerns about high costs and high risk of failure.

Currently, there are only three companies in the world capable of mass-producing cutting-edge chips, including the processors used to power artificial intelligence systems: TSMC, Samsung Electronics and Intel.

Altman has held discussions with chipmakers about partnering with them and using trillions of dollars to build and operate new factories, as well as investments in energy and other artificial intelligence infrastructure. Many of the world's largest chip companies, including Nvidia, design their own chips but outsource production to companies such as TSMC.

Building a cutting-edge chip factory typically costs at least $10 billion. But even so, the scale Altman discussed is extreme: Stacy Rasgon, an analyst at Bernstein Research, estimates that over the entire history of the industry, spending on chipmaking equipment has been slightly higher. Above $1 trillion.

However, money isn't the only ingredient needed to succeed in one of the most complex manufacturing industries in existence.

Industry executives say there is uncertainty about finding engineers to run the massive new plants, getting the machinery to fill them and getting enough orders to justify the plants. The Chinese government has poured money into building its chip industry for decades, but its progress has been hampered by a number of other factors.

“The semiconductor industry is not short of capital,” said Jimmy Goodrich, a semiconductor industry expert and senior adviser to the RAND Corporation. "Look at China, they're pouring huge amounts of money into their industry and they're not getting what they want. Because the fundamental challenge is that the technology is very difficult."

Even building a slew of new chip factories won't necessarily solve Altman's near-term problem — a shortage of artificial intelligence chips needed to produce systems like OpenAI's ChatGPT. The biggest bottleneck in Nvidia's AI chip production is packaging, the manufacturing step after circuits are stamped onto silicon wafers.

Altman also complained about the cost of Nvidia chips — another problem that Raymond James analyst Srini Pajjuri said may not be solved by adding more chip factories.

"To bring down the price of AI chips, we need more competition with Nvidia," he said.

More factories are under construction. All of the largest chipmakers are spending tens of billions of dollars on these chips, accelerating the pace of expansion, with global chip sales expected to exceed $1 trillion by 2030. TSMC's silicon wafer volume has increased from about 10 million wafers per year in 2016 to 16 million last year.

Chip insiders say that if Altman's plan succeeds, it could oversupply the market, drive down prices, and cause companies to operate factories with a significant shortfall in production capacity, a wake-up call for the industry due to high fixed costs. A financial death knell.

“Now they’re investing, but as online demand comes in, online demand goes away,” Bernstein’s Rasgon said, explaining how the industry has exploded. “That’s always going to happen because a lot of Sometimes the requirements you build are never real in the first place, hence the endless loop.”

The industry's current boom has been aided by governments around the world, which have invested in chips, recognizing their importance to their technological, economic and military supremacy. After passing the CHIP Act two years ago, the United States is allocating $39 billion to new factories as it seeks to rebuild an industry that has shifted to Asia in recent decades.

The scale of funding Altman is considering would dwarf incentives offered in the United States and elsewhere, including China, Taiwan, South Korea and Japan.

Altman has met with Commerce Secretary Gina Raimondo and other U.S. officials in recent weeks to discuss his ambitions, according to people familiar with the matter. Raimondo is overseeing the distribution of grants under the CHIP Act.

After a series of recent interactions with chip companies, Altman will appear at an event later this month for Intel's contract chip manufacturing operations. Intel is expected to be a major recipient of CHIP Act funding in the coming weeks.

It's unclear how Altman will find the talent to staff dozens of new factories. The Semiconductor Industry Association predicts that 115,000 jobs will be created by the end of this decade, 58% of which are at risk of becoming unfilled. and was unsure whether he could procure sufficient manufacturing equipment within a reasonable period of time. The lead time for some chip manufacturing equipment is about two years.

Industry insiders say the challenges of making chips are different from those faced by Altman in his early days as a startup, with many involving computers and software.

“In software, anything is possible—it’s really just a matter of money and coding,” Goodrich said. "However, in the world of hard technology, you actually have to deal with the laws of physics. You have to think about real-world and engineering challenges, and that's hard to do."

Original link

https://www.tomshardware.com/tech-industry/artificial-intelligence/jim-keller-responds-to-sam-altmans-plan-to-raise-dollar7-billion-to-make-ai-chips

Click here to follow and lock in more original content

END

*Disclaimer: This article is original by the author. The content of the article is the personal opinion of the author. The reprinting by Semiconductor Industry Watch is only to convey a different point of view. It does not mean that Semiconductor Industry Watch agrees or supports the view. If you have any objections, please contact Semiconductor Industry Watch.

Today is the 3679th issue shared by "Semiconductor Industry Observation" with you. Welcome to pay attention.