NBIS · Nebius Group N.V. · Stock research

Current Thesis

The GPU-neocloud trade has changed character. On 7/1 Bloomberg reported Meta is standing up Meta Compute developer access to models on Meta's own GPUs plus raw hourly GPU rental, the exact product neoclouds sell. NBIS and CoreWeave fell ~15% that session; Meta rose over 10%. The de-rating has not stopped: NBIS closed 6/18 at an all-time high of $286.69 and traded $177.93 on 7/18, roughly -38% in a month, with the 7/18 range $164.31–$186.77. That happened through good company news a $1B+ Reflection AI compute agreement (7/14), an asset-light partner model (7/15), and a $775M non-dilutive secured facility (7/17). When a name absorbs three positive catalysts in five sessions and still makes lower lows, the marginal buyer is gone and the market is repricing the terminal value of the business model, not the quarter. The operating ramp is intact; the narrative that was worth 230% in H1 is not. This is a stand-aside until the tape stops making lower lows and reclaims the high $190s–$200 zone on expanding volume, or until the 8/05 print forces a resolution.

Bullish and bearish views on Nebius Group N.V.

The model's bull view on Nebius Group N.V. (NBIS), in brief: 7/17 $775M senior secured facility, priced SOFR+250, maturing 2030-10-31, led by MUFG with ABN AMRO, BofA, Deutsche Bank and HSBC as lead arrangers. The bear view: Meta Compute is a demand-base attack, not a headline. Both cases follow in full.

Bull Case

• 7/17 $775M senior secured facility, priced SOFR+250, maturing 2030-10-31, led by MUFG with ABN AMRO, BofA, Deutsche Bank and HSBC as lead arrangers. Collateralized by deployed GPUs and contracted cash flows from an investment-grade customer; the company states it covers >100% of the capex required to deploy that infrastructure. Shares jumped ~8% into a broadly red tape.
• 7/14 Reflection AI agreement: >$1B of compute sold through 2029, on Nvidia GB300 silicon. Backlog builds from a customer that is neither Meta nor Microsoft, which directly answers the concentration critique.
• 7/15 asset-light partner model: infrastructure partners deploy and operate the full Nebius stack in their own data centers. Capex per dollar of revenue falls; the power/land constraint stops being solely a Nebius balance-sheet problem.
• 5/14 Q1'26 print: revenue $399M (+684% YoY, +75% QoQ), Nebius AI revenue $390M (+841% YoY, ~98% of group). Core AI-cloud ARR $1.92B, +54% QoQ. Group adj EBITDA $130M (32% margin); Nebius AI adj EBITDA margin 45% versus 24% in Q4'25.
• Funding structure remains the differentiator. Q1 operating cash flow of $2.3B on customer prepayments, ~$2B post-Yandex cash, and now GPU-backed senior debt the 2026 capex program is funded without an equity raise, against debt-funded CoreWeave and equity-funded APLD/IREN.
• Contracted anchors: 3/16 Meta $27B/5-yr and the prior $17B Microsoft contract; 5/20 Bloom Energy agreement for ~250MW guaranteed power (up to $2.6B in service fees) with the contracted-power guide raised 3GW→4GW; 6/8 UK expansion of £1.7B across four Nvidia-powered sites.
• Ownership base: Nvidia's 13F (5/15) shows 1,190,476 shares; Situational Awareness disclosed 5.6% on 5/27; Nasdaq-100 inclusion effective ~6/22 added passive demand. Consensus PT sits near $244 against a high-$170s tape.

Bear Case

• Meta Compute is a demand-base attack, not a headline. Meta spends $125–145B on AI infrastructure in 2026, including a 2,250-acre Louisiana campus and a 1GW Midwest build. The 7/18 report that Meta is negotiating a $10B, two-year compute deal with Anthropic is the confirming datapoint: Meta is already selling capacity to frontier labs. The neocloud pitch was scarce GPUs at premium rates; a hyperscaler with excess capacity and no cost of customer acquisition compresses that rate.
• The $27B Meta contract is now a liability in narrative terms. Concentration risk with a counterparty that is building the competing product is the structural bear, and CoreWeave's $21B Meta exposure means the group re-rates together on any Meta commentary.
• The tape rejects good news. Down 13% on 7/16 and extending 7/17 despite the debt facility and the Reflection AI win. IREN is -37% and NBIS -36% over the trailing month. Failure to rally on catalysts is the cleanest evidence of distribution.
• Sentiment flipped inside two weeks. NBIS was a CNBC Final Trade on 7/10 and 7/13; by 7/17 Cramer called it "at the nexus of the craziness" and "not done going down." Mainstream capitulation commentary at a -38% drawdown means the retail bid that carried the June leg is being unwound, not replenished.
• Leverage arrives at the wrong moment. The $775M facility solves dilution but adds fixed obligations into a period where GPU rental pricing faces its first credible competitive pressure. Secured debt against depreciating GPU assets is fine at full utilization and unforgiving below it.
• Power-chain read-through: the 7/8 Hunterbrook short report on Bloom Energy's China-free scandium supply claims puts a question mark on the 250MW anchor that underwrites part of the buildout.
• Sell-side is now stranded above the price. A ~$244 consensus target against a $178 market invites a cut cycle; PT reductions into a broken chart tend to arrive in clusters and mark the second leg down, not the bottom.

Setup & Price Structure

Structure is broken on every timeframe that matters. The 6/18 all-time closing high of $286.69 was followed by a 7/1 gap-down on the Meta report, a failed stabilization attempt in mid-July, and fresh lows on 7/16–7/17. The 20-EMA rolled over in early July and the 50-day is now overhead resistance rather than support. Price sits roughly 38% below the June high with no completed higher low.

The working zones: $164.31 (7/18 session low) is the immediate shelf losing it on a weekly close removes the last reference under the June breakout and opens the $140s, where the pre-breakout base sits. $186.77 (7/18 high) is the first thing bulls must reclaim to argue the selling is exhausted. The real re-arm level is the $196–$203 band, which contains the round number, the broken July support, and the low end of the current-price-target cluster; a close above it on expanding volume with CRWV and IREN moving in sympathy would be the first evidence of a genuine base rather than a bounce. $223.62 (5/28 signal level) is the level that would confirm a trend repair.

The behavioral read: this is what a crowded momentum name looks like on the way out, with short interest that had been building across NBIS/IREN/CoreWeave since late June now working against longs instead of fueling squeezes. Buying the -38% drawdown before a higher low exists is averaging into a broken structure. There is no entry here that isn't a knife-catch, and the 8/05 print is close enough that anything bought now is a bet on an earnings reaction rather than a trend.

Catalyst Calendar (next 30 days)

• 2026-08-05 Q2'26 earnings (confirmed). Consensus revenue ~$593.6M, EPS -$0.68. The number that matters is core AI-cloud ARR against the $1.92B Q1 base; anything short of ~50% QoQ growth validates the competition thesis. Second: management's direct commentary on Meta as a competitor and on the durability of the $27B contract. Avoid fresh entries within three trading days of the print.
• Late July (est.) Meta Q2 earnings and any Meta Compute commercial detail. Confirmation of Meta Compute pricing, GA timing, or the Anthropic deal is the single largest exogenous risk to the whole group over this window.
• Ongoing, 7/19–8/05 sell-side price-target revisions. A cluster of cuts from the ~$244 consensus toward the tape would confirm the de-rating; a defended target on the debt facility and Reflection AI win would be the first contrarian datapoint.
• ~2026-08 (undated) CoreWeave Q2 print. Read-through on GPU rental pricing and utilization for the entire cohort; often the more honest tell on rate compression than NBIS's own guide.
• Ongoing asset-light partner announcements. The 7/15 model needs named partners with signed capacity to become a thesis rather than a press release.

What Would Change Our Mind

• Bullish re-arm: a weekly close back above the $196–$203 band on above-average volume, with CRWV and IREN confirming that would mark a completed higher low and re-establish a tradable trend.
• Bullish, fundamental: the 8/05 print showing core AI-cloud ARR sustaining ~50% QoQ growth plus explicit contract-durability language on Meta. A second non-hyperscaler contract in the Reflection AI size range would break the concentration argument outright.
• Bullish, competitive: credible reporting that Meta Compute is limited to internal and partner-lab use rather than a general merchant GPU offering would refute the 7/1 thesis and force a violent re-rating higher.
• Bearish confirmation: a weekly close below $164 opens the $140s pre-breakout base. Add to that any Meta disclosure of merchant GPU pricing, a reduction or restructuring of the $27B contract, or a cluster of price-target cuts landing below $200.
• Bearish, structural: an ARR ramp below ~40% QoQ on 8/05, or gross-margin compression on the AI-cloud segment from the 45% Q1 level, would mean rate pressure is already in the numbers and the drawdown is a re-rating rather than a scare.

Correlation Notes

• Primary comp: CRWV. Same customer (Meta, $21B), same product, more leverage. Moves first and harder on Meta headlines; a CRWV bounce that NBIS fails to match would signal name-specific selling.
• Higher-beta proxies: IREN, APLD. IREN -37% over the trailing month tracks NBIS almost tick-for-tick on group news, with a bitcoin-mining conversion story layered on top that adds noise.
• Inverse leg: META. Up over 10% on 7/1 while the group fell 15%. Long META / short the neocloud basket is the expression the market has chosen for this theme; that pair moving further apart is the bear case being confirmed in real time.
• Upstream: NVDA. Nvidia is both supplier and shareholder here. NBIS decoupled from NVDA in July the group is trading on rental-rate risk, not chip-demand risk, so NVDA strength no longer lifts it.
• Power chain: BE, VRT, PWR. The 5/20 Bloom agreement links NBIS to power-equipment sentiment; the 7/8 Hunterbrook report on BE is a live secondary exposure.
• Theme state: MATURING, with the burden of proof now on the bulls. The GPU-neocloud narrative is no longer in its accelerating one-way phase. It has not gone dead contracted backlog and the debt market both say the buildout continues but it has become a two-sided trade where the marginal news flow determines direction. Fresh long exposure in this cohort requires a completed base and peer confirmation, not a valuation argument.