AIToday
Large Language ModelsAI Business & IndustryYahoo Finance AIPublished: Sep 15, 2026, 01:00 JST

IREN CEO Daniel Roberts: AI demand outruns supply by years

IREN CEO Daniel Roberts: AI demand outruns supply by years

3 Key Points

  1. What happened

    IREN fell almost 5% in premarket trade Monday even as CEO Daniel Roberts said AI processing demand still far outstrips supply and would take years of extra infrastructure to deploy today's capabilities.

  2. Why it matters

    Roberts cited Google processing almost seven times more tokens than a year ago, plus Anthropic's development and OpenAI's constraints. GPUs aren't the only bottleneck — memory bandwidth is scarce and new capacity takes years.

  3. What to watch

    The test is whether this supply shortfall turns into greater pricing, client commitments and returns on IREN's growing infrastructure. Roberts predicted roughly half of U.S. data-center capacity installed over the next two years may arrive on time.

WHO IT HITSInvestors holding IREN and other AI infrastructure stocks face a disconnect: shares soften with the broad AI trade, yet the company's thesis rests on supply scarcity rather than falling demand. Data-center operators and their customers are watching whether power hookups and building schedules delay capacity.

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Context & Analysis

IREN, a data-center and AI infrastructure provider, saw its shares drop almost 5% in premarket trade Monday. That decline came alongside CEO Daniel Roberts' argument that demand for processing capacity remains far ahead of supply. Roberts drew a distinction between concerns about slowing frontier AI development and demand for today's AI tasks. He said even if model improvements ceased today, it would still take years of extra infrastructure to deploy today's AI capabilities.

Roberts pointed to Anthropic's significant development, OpenAI's continuous computational constraints, and Google processing almost seven times more tokens than a year ago. He also noted that GPUs are not the only bottlenecks. Memory bandwidth is still scarce, and it takes years to produce new memory capacity. On top of that, Roberts predicted that roughly half of U.S. data-center capacity to be installed over the next two years may be delivered on time, held back by power hookups and building schedules.

That leaves IREN in an unusual spot: its shares are softening with the general AI trade, but its primary thesis rests on scarcity rather than falling demand. The next test for investors is whether that supply shortfall translates into greater pricing, client commitments and returns on IREN's growing infrastructure. Much hinges on whether delayed capacity actually tightens the market enough to lift those commercial terms.

FAQ
Why did IREN fall despite the CEO's bullish comments?
The article says shares softened with the general AI trade, but the company's primary thesis is based on scarcity, not falling demand.
What bottlenecks does Roberts cite besides GPUs?
Memory bandwidth is still scarce, and it takes years to produce new memory capacity. Power hookups and building schedules also make delivery harder.
What prediction did Roberts make about U.S. data-center capacity?
He predicted that roughly half of U.S. data-center capacity to be installed over the next two years may be delivered on time.
Yahoo Finance AIRead Original Article

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • Huang names Land, Power, Shell as AI's bottleneckExponential Industry · 1h ago
  • Anthropic eyes Nasdaq IPO at $2 trillion or moreTHE DECODER · 1h ago
  • Microsoft's MAI code: human control first, no inner lifeTHE DECODER · 1h ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. Q&As are published on this page for other readers too.

Related Articles

Next articleUAE weighs moving 5-gigawatt AI data center underground