Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/sam-altman-wants-washingtons-backing-for-his-7-trillion-ai-chip-venture.19632/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Sam Altman wants Washington's backing for his $7 trillion AI chip venture

Daniel Nenni

Daniel Nenni

Admin
Staff member
OpenAI CEO Sam Altman was fired by the company's board on Friday. Justin Sullivan/Getty Images

OpenAI CEO Sam Altman was fired by the company's board on Friday. Justin Sullivan/Getty Images© Justin Sullivan/Getty Images
  • Sam Altman wants US government approval for his trillion-dollar AI chip venture, Bloomberg reported.
  • Altman has been attempting to raise up to $7 trillion to boost his GPU-chip supply.
  • Nvidia CEO, Jensen Huang, has expressed skepticism about the hefty sum.
Sam Altman is trying to get Washington's backing for his trillion-dollar AI chip venture.


The OpenAI CEO is working to secure US government approval for the project as it risks raising national security and antitrust concerns, Bloomberg reported.

Altman has reportedly been attempting to raise up to $7 trillion to boost his GPU-chip supply. The Wall Street Journal reported that the tech CEO had held talks with investors including in the United Arab Emirates.

Altman told potential investors that he can't move forward without Washington's approval, Bloomberg reported.

Altman's proposals are aimed at helping to solve the global chip shortage, according to the reports. He's said to be presenting his pitch as a partnership with OpenAI, chip makers, and investors who can finance GPU chip plants, the report says.

Representatives for OpenAI did not immediately respond to a request for comment from Business Insider.

Jensen Huang, the CEO of leading chipmaker Nvidia, hasn't been overly positive about Altman's plan.


He told the United Arab Emirates' AI minister, Omar Al Olama, that developing AI wouldn't cost as much as the amount Altman was seeking to raise.

Huang said: "There's about a trillion dollars' worth of installed base of data centers. Over the course of the next four or five years, we'll have $2 trillion worth of data centers that will be powering software around the world."

He also took an indirect shot at Altman, joking that $7 trillion could buy "apparently all the GPUs."

 
blueone said:
I'm still mystified that Altman has any credibility at all.
Click to expand...
Returning to my comment about this last week (what could you possibly spend the $7trn on).

Some others pointed out that you could buy out a few of the magnificent seven. But that creates nothing new.

I missed the possibility that you could hire 1 million new engineers for 7 years at an annual salary of $1m each (!). Except there aren't 1 million such engineers out there just waiting to be hired.

I can't see how you can hose the industry with so much money without creating both huge wage inflation and overcapacity.

On the bright side, taking 10s of billions out of the Gulf States and spending it in the US with no reasonable expectation of paying most of it back might be popular in Washington.
 
Let me do some simple calculations to estimate how many AI wafers/chips will be created after $7 Trillion investment.
1) If half of that is used to build fabs in 10 years, which means $350B/year in new fabs. Another $3.5T is assumed for product design/development and others.
2) Current WFE TAM will be ~$100B. Then it will imply ~26% CAGR in 10 years.
3) If that is for 2nm or 3nm AI chips, the fab cost will be ~$3B/10kwpm. $350B means ~1.16Mwpm more and ~14Mwpy. This is 10x TSMC CapEX intensity.
4) If AI chip is big with ~100DPW and good enough D0 to produce ~60GDPW, then there will be 840M chips/ year produced.
5) Not mentioned supported memory and other assembly/test requirements.

To achieve this, it implies creating AI robots to build fabs and doing manufacturing by AI with very good yield and more. It is a dream and still far far away from us.
 
Last edited:
hskuo said:
Let me do some simple calculations to estimate how many AI wafers/chips will be created after $7 Trillion investment.
1) If half of that is used to build fabs in 10 years, which means $350B/year in new fabs. Another $3.5T is assumed for product design/development and others.
2) Current WFE TAM will be ~$100B. Then it will imply ~26% CAGR in 10 years.
3) If that is for 2nm or 3nm AI chips, the fab cost will be ~$3B/10kwpm. $350B means ~1.16Mwpm more and ~14Mwpy. This is 10x TSMC CapEX intensity.
4) If AI chip is big with ~100DPW and good enough D0 to produce ~60GDPW, then there will be 840M chips/ year produced.
5) Not mentioned supported memory and other assembly/test requirements.

To achieve this, it implies creating AI robots to build fabs and doing manufacturing by AI with very good yield and more. It is a dream and still far far away from us.
Click to expand...
Chip design R&D, even including software enabling like CUDA, is a much smaller investment compared to the combination of fab construction and fab process development.
 
Seven Trillion Dollars ($7T)! Altman’s vision is childish and, not to put too fine a point on it, absurd. The H100 devices he covets are FP64 TeraFlops devices. What is needed are FP64 PetaFlops, or better stil FP64 ExaFlops devices and their concomitant Ai/ML algorithmic architectures for Deep Learning Networks, rather than the general purpose high performance scientific computing architectures currently being ‘misused’ by the hyperscalers for Ai/ML workloads. Hopefully, physical scaling would allow Chiplets and Heterogeneous IC Integration to enable massive data throughput at a reduced energy footprint.

Assuming his calculations are correct, instead of $7T, Altman would need $7B for FP64 PetaFlops devices and $7M for FP64 ExaFlops devices! Suggests he pumps $5B into domain specific Ai/ML algorithmic architectures R&D. But wait, that would democratize the Ai/ML landscape, decimate the hyperscalers’ rental business and most significant deliver affordable Academic Compute!

Geoff Boyd, San Jose, CA
 
The US National Debt clock is at $34T currently. That's $101,847 per citizen. https://www.usdebtclock.org/

The main/only way to ever get out from under this is through technology and growth. AI is the main vehicle under consideration to achieve enough growth to pay back those trillions.

The idea would be to spend $7T, grow GDP more than, I don't know, $70T (10x being the usual multiplier for advanced R&D), so the $34T is dwarfed and easily paid off.
 
For $7T, Sam could buy up all the chip companies - NVDA, TSM, Broadcom, AMD, Samsung, Intel, ASML, ARM, QCOM, SNPS, MU, CDNS and still have $2T of spare changes.

Actually, he can buy every single semiconductor companies in the world. There are 119 chip companies with a total market cap of $6.419 T
 
is 7 trillion going to be used on Nvidia's chip? why not build OpenAI's own AI chip with 7 trillion? Sam can hire the best to design the next-generation AI chip and a better platform than Nvidia's CUDA and hire Daniel as a consultant on how to spend that 7 trillion. :)

​

 
soAsian said:
is 7 trillion going to be used on Nvidia's chip? why not build OpenAI's own AI chip with 7 trillion? Sam can hire the best to design the next-generation AI chip and a better platform than Nvidia's CUDA and hire Daniel as a consultant on how to spend that 7 trillion. :)
Click to expand...

Seriously, I cannot imagine anyone raising trillions of dollars for anything.
 
You must log in or register to reply here.
Back
Top