Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/obsolescence-driving-chip-demand-ai-has-changed-all-the-rules.19119/
        )

    [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
)

Obsolescence driving chip demand? AI has changed all the rules?

Arthur Hanson

Well-known member
Could accelerating obsolescence push up chip demand? Also, could AI/ML driven automation increase chip demand? Will customers be forced to advance AI. automation and research at an increasing rate increasing the turns required to keep up with the competition? I firmly believe AI combined with advanced automation will be the greatest revolution in the history of man and we are just in the early stages of a trend that has never happened in human history, which makes all of previous thinking on trends obsolete. Any thoughts or observations in this area appreciated.
 
Just seen interesting study yesterday:
F-kVQuvWkAAemkr


Twitter thread: Paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4527336
 
I look at this question by comparing to the traditional product cycle of chips starting in the mid 80's. (I came to Silicon Valley in the early 70's). By then, the two-year cycle of technology development was establishing itself and a natural hardware product market followed, with high value leading edge innovations synchronized with the new node introduction, and the off years creating products based on the cost reductions that "came with" once the new process was running smoothly in the fab. Intel formalized this with their "tick-tock" process train. (Nothing new here for most readers of this column.)

With AI, the end products are SW, not HW (in my parlance), and both the increasing training parameter requirements and value-based real-time inference demands are going to require constantly increasing compute power at a more rapid pace than what can be upgraded in existing server farms. Lower rates of return on energy per operation with each new node are also going to make it more difficult to fit into the power envelope restrictions of existing server sites. Once the digital companion and similar markets are established (re-watch "Her"), this will lead to an increased build-out of new solar-based server farms in the next few years based on the most cost-effective currently produced processors. My expectation.
 
Back
Top