JPMorgan: Chip Replacement Cycle to Sustain NVIDIA Demand Through 2030
0xBroomberg
JPMorgan forecasts GPU and AI chip spending will rise from 50% to 60% of total AI capex by 2030, driven by short chip replacement cycles; Nvidia benefits directly, but whether AI investment can generate sufficient returns remains the biggest unproven link in the entire thesis.
Why does JPMorgan think chip spending will keep climbing?
The core logic: data-center infrastructure — buildings, power equipment — lasts a long time and is largely a one-off build. That spending curve is expected to flatten around 2028.
But GPUs and AI chips have a useful life roughly one-tenth that of the infrastructure. In plain terms = the building is a one-time cost, but the chips inside it need replacing every few years, creating a recurring purchase cycle.
This means → chip spending's share of total AI capex will rise from about 50% to 60% by 2030. JPMorgan projects total AI chip and core hardware financing will exceed $3 trillion over the next five years.
How big is the money at stake?
JPMorgan raised its pre-2030 total AI capex forecast from $5.1 trillion last November to $5.5 trillion.
Within that, silicon spending is projected to grow from $340 billion in 2026 to roughly $800 billion four years later — more than doubling.
Nvidia posted $81.6 billion in FY2026 Q1 revenue, up 85% year-over-year. CEO Jensen Huang positions the company as the "core hub" of the AI transition; CFO Colette Kress forecasts AI spending will reach $3–4 trillion per year by decade's end.
How far ahead is Nvidia — and how fast are rivals closing in?
JPMorgan expects Nvidia to ship 8.9 million GPUs this year, well ahead of Google's 4.5 million TPUs and Amazon's 1.9 million Inferentia/Trainium chips.
But the gap is narrowing fast: next year Google TPU shipments are projected at 8 million, versus Nvidia's 9.9 million. This means → Google moves from less than half of Nvidia's volume to near parity in one year.
Stock performance reflects the shift: Nvidia is up over 12% year-to-date in 2026, while AMD has more than doubled — partly on rising expectations for CPU demand growth, diverting capital within the AI trade.
What is the biggest uncertainty?
Whether the money being spent on AI can actually earn a return — this is the key unproven link in the entire investment thesis.
U.S. Bureau of Labor Statistics data shows Q1 2026 labor productivity growth at just 0.3%. MIT economist Daron Acemoglu estimates AI will lift total factor productivity by no more than 0.71% over the next decade.
In plain terms = spending keeps rising, but the question "how much faster is AI actually making people work?" still lacks a convincing answer from the data.
How real is the overbuild risk?
Former Deloitte chief cloud strategist David Linthicum told CNBC that actual build-out capacity may be only about half of what was originally planned, flagging construction-start data as the key metric to watch.
This means → if enterprise-side AI utilization cannot support the current spending pace, overbuild risk will surface gradually.
JPMorgan's bull case for Nvidia rests on structural demand from chip replacement cycles — but whether that logic holds ultimately depends on whether the AI application layer can generate enough commercial returns to sustain continued investment.
Content is for reference only, not financial advice.