Nebius and IREN: How Two AI Infrastructure Highfliers Fell 40–60% From Their Highs

Back in February, we called Nebius a hypergrowth capex machine that Wall Street was underrating, and separately covered IREN's metamorphosis from Bitcoin miner to MSCI USA Index member on the back of a $9.7 billion Microsoft contract. Both stocks have had an extraordinary — and extraordinarily painful — six months since. Both are down sharply from their 2026 highs. Both got hit by the same specific catalyst in July. And both are still signing genuinely large new contracts even as their stock prices tell a story of investors losing patience.

The Drawdowns, Side by Side

Both stocks are trading well below where fundamentals-focused analysts think they should be — IREN's own $81 consensus target implies roughly 175% upside from its Wednesday close, a gap its covering analysts themselves describe as unusually wide. Whether that reflects genuine mispricing or a market that's simply stopped trusting neocloud valuation models is the live question for both names right now.

The Shared Shock: Meta's Own Cloud Ambitions

The single biggest catalyst hitting both stocks this cycle wasn't company-specific — it was Meta. On July 1, Bloomberg reported Meta plans to launch "Meta Compute," a new unit to sell its surplus AI compute capacity and hosted models directly to other customers — effectively entering the cloud market Nebius, IREN, and CoreWeave all compete in. Nebius alone lost roughly $12 billion in market value in a single day on the news; CoreWeave and other neoclouds fell in sympathy the same session. Amazon responded within the same window by raising reserved GPU prices roughly 20%, in what one outlet read as AWS's own answer to Meta's move.

This is directly connected to what we covered in our own Meta earnings analysis three weeks later: Zuckerberg confirmed on Meta's Q2 call that a cloud-compute offering is coming, while framing it as a smaller side opportunity next to "selling intelligence." For Nebius and IREN, that framing is exactly backwards from what matters to their stock prices — even a "side opportunity" version of Meta Compute represents a hyperscaler-scale potential competitor entering their market, and the mere announcement was enough to wipe billions off both names before Meta has shipped a single contract under the new unit.

The tension is sharper for Nebius specifically, because Meta is simultaneously one of Nebius's own reported customers (contracted revenue north of $40 billion across Microsoft and Meta combined, per Nebius's own disclosures) — meaning Meta is, at once, a customer buying Nebius's compute today and a company building the infrastructure to potentially stop needing to buy it tomorrow. That's not a hypothetical conflict of interest; it's the exact dynamic playing out in real time in Nebius's own contract book.

What Made July Worse: The Macro Backdrop

Both stocks' late-July declines were amplified by the same macro shock we've covered across this earnings season: on July 29, escalating Iran-Israel hostilities pushed oil above $84/barrel, the Federal Reserve delivered its most hawkish dissent in nearly a decade (raising September rate-hike odds to roughly 80%), and the Dow fell 1,153 points — its worst single day since April 2025. For a stock with IREN's 4.28 beta, that's not noise; it's a mechanical amplifier. Higher oil pushes Treasury yields up, higher yields raise the discount rate applied to future cash flows, and both Nebius and IREN are, fundamentally, bets on years of future revenue from contracts that are barely generating cash today. A higher discount rate compresses the present value of that future revenue immediately, with zero change to either company's actual contract book. Comparable pivots — TeraWulf and Applied Digital among them — fell 36–38% over the same stretch, while a diversified data-center infrastructure ETF fell a comparatively contained 13%, isolating how much of the July damage was beta amplification versus company-specific news.

Company-Specific Wounds Layered on Top

Nebius's own July was rockier still, independent of Meta and macro. On July 28, shares fell 12% to $165.30 as investors scrutinized the math directly: Q1 AI-cloud adjusted EBITDA margin of 45% doesn't account for depreciation, and group D&A ran to 53% of revenue — meaning GPU ownership costs are eating deeply into reported profitability. 2026 capex guidance of $20–25 billion against $3.0–3.4 billion of revenue guidance works out to 5.9–8.3 times revenue in spending — even by AI-infrastructure standards, a striking ratio. Nebius announced a new "asset-light" data-center partnership model on July 15 (partners fund and own equipment; Nebius supplies architecture, software, and customer relationships) as a partial answer to that funding math, but disclosed no named partners, capacity commitments, or timelines — leaving the market to take the strategy on faith for now. A $775 million secured debt facility on July 17, backed by deployed GPUs and contracted customer cash flows, helped ease acute dilution fears and drove an 8% bounce, evidence the market does respond positively once Nebius shows concrete funding progress rather than just directional strategy.

IREN's July wound was more technical than strategic: a large stock-compensation award to co-CEOs Daniel Roberts and William Roberts in early July triggered nine consecutive losing sessions, with retail investors explicitly citing the dilution as the trigger — separate from and compounding the macro and Meta-driven pressure. IREN's short float sat near 25% at the time of its July 20 contract news, a substantial overhang that amplifies moves in both directions.

What Each Is Actually Signing — the "Rumored Contracts" Question

Both companies spent July doing the opposite of retreating: signing real, disclosed contracts even as their stock prices fell.

IREN's July 20 announcement is the clearer of the two. $2.8 billion in new multi-year AI cloud contracts, raising the year-end 2026 ARR target from $3.7 billion to over $4 billion, with roughly 85% of that revised target now under signed contract and customer prepayments covering about 45% of associated GPU capex. Crucially, the customer list has diversified well beyond IREN's original Microsoft anchor: Microsoft and NVIDIA as hyperscaler-tier customers, joined by Perplexity, Figure AI, Together AI, Fluidstack, Fireworks AI, Fal AI, Hume AI, and one undisclosed additional developer. Goldman's Michael Ng specifically flagged this as reducing customer-concentration risk — a real structural improvement versus the single-hyperscaler dependency that defined IREN a year ago, even though Goldman still rates the stock Neutral on execution-risk grounds.

Nebius's forward pipeline is a mix of a real disclosed deal and undisclosed preliminary ones. The $1 billion Reflection AI compute agreement (an $8 billion AI startup) pushed Nebius's contracted backlog toward roughly $50 billion when combined with its existing Microsoft and Meta commitments — Microsoft's original deal alone was reported at $19 billion. Nebius's new asset-light model has reportedly reached "preliminary agreements" with unnamed partners, per CEO Arkady Volozh's own comments, but no partner names, capacity figures, or timelines have been disclosed — the single biggest piece of "rumored, not yet confirmed" upside sitting in Nebius's story right now, and the thing most likely to move the stock meaningfully whenever it's finally specified.

What to Watch

• IREN's next earnings report, expected September 16, 2026 — the first hard checkpoint on whether the 480-megawatt Childress capacity additions and the $4B ARR target are tracking on schedule.
• Whether Nebius names actual partners under its asset-light model — turning a strategy announcement into disclosed commitments would likely be a bigger catalyst than any single new customer contract at this point.
• Meta Compute's actual launch and early customer list, whenever it materializes — the difference between "a side project" and "a real hyperscaler-scale competitor" will only become clear once Meta ships something concrete.
• Whether September CPI prints and the Fed's actual September decision validate or deflate the roughly 80% rate-hike probability currently priced in — both stocks' valuations are unusually sensitive to the discount-rate assumptions behind their multi-year contract backlogs.

The Bottom Line

Nebius and IREN are living through nearly the same story from two different starting points: real, growing, increasingly diversified contract books, undercut by extreme capex-to-revenue ratios, funding structures that still lean on debt and dilution, and a market that's currently unwilling to look past near-term financing risk to price in the multi-year backlog either company has actually signed. Both got hit by the same Meta-shaped shock to the neocloud thesis, and both got hit harder than the fundamentals alone would suggest by a geopolitically rattled, hawkish-Fed macro tape that punishes long-duration growth stories disproportionately. The gap between where analysts think these stocks should trade and where they actually trade is currently about as wide as it's been all year for both names — which makes the next hard data points (IREN's September earnings, whatever Nebius eventually names under its asset-light model) unusually consequential for settling which read is right.

This is a follow-up to our February 2026 Nebius analysis and February 2026 IREN analysis, incorporating both companies' July 2026 contract announcements and stock drawdowns.