How the rise of AI and machine learning has spawned a new chips arms race involving tech giants and a plethora of chip development startups
Andy Patrizio / Ars Technica : Tweets: @eladgil Tweets: Elad Gil / @eladgil : The most exciting area of AI no one ever talks about.... http://twitter.com/...
Context & Ripple Effects
This 2018 piece marks the opening of the arc the rest of the coverage tracks: Andy Patrizio frames AI and machine learning as the force pulling both tech giants and a wave of chip startups into silicon. The startup side was already visible — a Cerebras Systems profile months earlier showed a little-known Los Altos company raising $112M for AI training chips.
By 2021 the challengers had names — Google, Amazon, Graphcore, Cerebras all positioning against Nvidia's dominance in AI chips — and the race has since widened from accelerators to CPUs themselves, with Intel and AMD now defending data-center ground against Nvidia, Qualcomm, Arm, and cloud players.
First-order effects
- Tech giants designing their own AI silicon immediately reduce their dependence on Nvidia for training hardware, while startups like Cerebras convert venture capital into a shot at the same workloads.
Second-order effects
- Nvidia's largest customers become its competitors — the dynamic later flagged when rivals and key customers began shipping their own AI chips — squeezing the very demand that powered its boom, while many of the 2018-era startups struggle to monetize against incumbents' distribution.
Third-order effects
- If the pattern holds, AI compute fragments across vendors and architectures rather than consolidating around one supplier — reinforced by Epoch AI's expectation that the number of AI chips in use will double every nine months, and by CPUs re-entering the data-center fight.
The trend: AI workloads are pulling the semiconductor industry from a single-accelerator market toward multi-vendor competition spanning giants, startups, and eventually CPUs again.