IREN's AI cloud ambitions face pressure as GPU lease rates tumble 30% amid intensifying competition from Meta's custom silicon and slowing demand.

Leasing rates for Nvidia's B200 accelerators have plunged roughly 30% in just three weeks, a decline that directly threatens the business model of specialized AI cloud operators such as IREN. The sell-off comes as Meta signaled it may begin renting out spare computing capacity—a move that could flood an already price-sensitive market.

IREN shares fell 2.67% on Friday to €35.60, extending a retreat that has left the stock nearly 48% below its November 2025 record of €68.61. While shares show a 144.4% gain over twelve months, the recent pullback has erased almost all year-to-date gains; the stock is down 2.39% since the start of 2026.

Meta's aggressive infrastructure push is driving the current weakness. The social media giant plans to invest between $125 billion and $145 billion in data centers during 2026 alone, much of it dedicated to its own large language models such as the recently released Muse Spark 1.1. Analysts increasingly worry that resulting capacity surplus will be offered to third parties, creating direct competition with IREN and other neocloud companies that lease GPU time.

For IREN, the timing is particularly challenging. The company is mid-expansion, growing its GPU fleet from 23,000 to 140,000 units by end of 2026. While the company anchors revenue through a $9.7 billion contract with Microsoft, its broader customer base remains thin compared with certain rivals.

The transformation from pure-play Bitcoin miner to AI infrastructure provider has added to volatility. The annualized price swing now stands at 91%, with a relative strength index around 38 indicating oversold conditions. Market capitalization stands at €12.44 billion.

The GPU expansion requires aggressive use of convertible bonds and debt—a capital-intensive strategy that depends on strong future margins. Investors are watching whether the company can achieve its $3.4 billion annual recurring revenue target once expansion completes. That goal assumes rental rates remain elevated, a bet growing increasingly fragile as hyperscalers like Meta and Amazon invest in proprietary chip development.

Meta plans mass production of its own AI chip, code-named Iris, beginning September 2026. This move would reduce dependence on external hardware and potentially erode the premium currently commanded by specialized GPU providers. IREN reports quarterly earnings on September 16—a key date for investors assessing whether the company's margins can withstand sustained price pressure from both existing competitors and new entrants.