Deep Dive
The scale of AI infrastructure spending
ARK opens with a stark metric: technology capital expenditure as a percentage of GDP has now surpassed spending levels during the 2000 tech and telecom bubble. The big four hyperscalers—presumably Amazon, Google, Meta, and Microsoft—are projected to spend more than $700 billion in 2026 alone, a figure that's expected to climb in 2027. This massive wave of spending is creating opportunity in chip design, where AMD is positioned to compete for a meaningful slice.
AMD's proven playbook in server CPUs
AMD's track record in data center CPUs offers a credible template for challenging Nvidia in AI accelerators. In 2017, AMD reentered the server CPU market with nearly zero share. By Q1 2026, it had captured 46% of server CPU revenue—a stunning climb achieved by shipping faster, more power-efficient processors on an annual cadence while Intel repeatedly missed deadlines. Each cloud instance built around AMD's EPYC processor line locks in multi-year procurement commitments, creating durable moats. AMD designs the chips and relies on TSMC for manufacturing, meaning the competition is purely in design and systems architecture.
Helios and the accelerator opportunity
AMD is deploying the same competitive strategy against Nvidia that worked against Intel. The company is launching Helios, a rack-scale solution, in the second half of 2026 with direct competition to Nvidia's offerings. ARK's research indicates AMD chips already deliver better performance-per-dollar than Nvidia on some workloads, a crucial advantage when hyperscalers are deploying at this scale. Customer announcements from OpenAI, Meta, and Oracle signal real demand, not theoretical positioning. The broader AI evolution toward more capable agents requiring traditional compute resources also plays into AMD's strength in CPU architecture.
The competition picture ahead
ARK expects AMD to mount a better competitive challenge than Intel ever did, though they acknowledge risks. Nvidia's execution quality matches AMD's across the past five years, and major customers like Google and Amazon are rapidly developing proprietary custom chips to reduce reliance on third-party accelerators. Still, ARK is optimistic that AMD's breadth across CPU and GPU architectures, combined with proven execution, positions the company to capture meaningful share of the AI compute TAM from a very small starting base.