NVIDIA's Groq 3 LPX Lands First at Nebius, Capping a Quarter That Rewrote the Neocloud Calculus

Nebius's reported selection as NVIDIA's Groq 3 LPX launch customer — arriving alongside a 454% revenue surge, a maiden positive EBITDA, and a $5 billion capital raise — marks a decisive inflection for Europe's AI infrastructure buildout, though mounting leverage and potential hyperscaler competition keep the outlook double-edged.

A Moomoo report dated August 25 states that NVIDIA has begun full-scale mass production of the Groq 3 LPX, a chip targeting the low-latency AI inference market, with the first systems confirmed for deployment at Nebius (NBIS.US). The claim, as yet uncorroborated by independent sources, carries significant weight if accurate: designation as launch customer for a new NVIDIA inference platform would place the Amsterdam-headquartered neocloud at the frontier of the inference hardware cycle before rivals can access equivalent silicon. The chip's ultimate impact on inference economics remains to be demonstrated, but the timing is telling — the report arrives days after Nebius posted the strongest quarterly results in its short life as an independent AI-infrastructure company, capping a period of relentless capital deployment and geographic expansion that has transformed it from a post-Yandex restructuring curiosity into one of the most closely watched names in European tech.

The financial results themselves require little embellishment. Multiple sources, including SEC 6-K filings and a Business Wire release, confirm that Nebius posted Q2 2026 AI cloud revenue of approximately $582 million, up 454% year-over-year. More consequential than the topline was the margin pivot: EBITDA swung from a $21 million loss in the prior-year period to a positive $236 million in a single quarter, implying a margin of roughly 41%. Shares surged as much as 34% intraday on the report. The EBITDA inflection suggests that utilization rates and contract pricing have stabilized at levels capable of absorbing the company's substantial depreciation base — a key threshold for a business whose model depends on high fixed-asset leverage. Earlier in the quarter, Nebius had validated NVIDIA's Vera Rubin NVL72 architecture at its Finland facility, reportedly the first cloud provider globally to do so, demonstrating that its early-access hardware relationships span multiple NVIDIA product generations simultaneously.

The post-earnings euphoria was short-lived. Within days of the Q2 report, Nebius announced a $4.5 billion convertible senior notes offering that was subsequently upsized to $5.0 billion by August 24, with multiple SEC 6-K filings documenting the process. The market reacted sharply: shares fell approximately 14% on the announcement, layering onto a week already marked by a broader neocloud selloff driven by interest rate pressure — CoreWeave fell roughly 7% in that macro-driven move, while Nebius itself had already shed around 5%. The stock now trades at $210.91, roughly 6% below the $224 target set by Piper Sandler at its Neutral initiation in early August. The $5 billion raise is a structural statement: a bet that AI infrastructure demand will remain robust well into the decade, and that GPU-backed capacity can generate returns sufficient to service the debt. The question the market is implicitly pricing is what happens to that calculus if Meta — reportedly planning an independent 'Meta Compute' cloud service — converts one of the sector's largest demand drivers into a direct competitor.

Nebius's present form cannot be understood without its unusual corporate genesis. The company emerged from the restructuring of Yandex NV — the Dutch parent of Russia's dominant technology group — following the divestiture of Russian assets in the aftermath of the 2022 invasion of Ukraine. Rebranded and reoriented around GPU cloud infrastructure, Nebius has since constructed a European-first data center footprint spanning two facilities in Mäntsälä, Finland, a new deployment in Estonia, and a leased capacity deal at Vantage Data Centers' CLW1 site in Newport, Wales, situated within the UK government's South Wales AI Growth Zone. A New Jersey facility brought the company into the US market, where it partnered with Bloom Energy for on-site fuel-cell power — a deal that sent Bloom's own shares up 12-13% on announcement. These choices reflect a coherent positioning logic: European data-sovereignty regulation creates structural demand for locally operated cloud alternatives, while low-carbon energy sourcing addresses ESG procurement requirements increasingly embedded in enterprise contracts.

The risks are as concrete as the opportunity. Michael Burry disclosed a short position at $211.77 in August, characterizing AI infrastructure as an overcrowded trade — a framing that, whatever one's view of the specific call, maps onto a real sector-wide concern: GPU-backed loan structures across neoclouds including CoreWeave, Nebius, and Lambda are concentrated in a 2026–2028 maturity window, creating a refinancing overhang precisely as rates remain elevated. The $5 billion convertible offering addresses part of this exposure while simultaneously increasing balance-sheet leverage heading into a period of macroeconomic uncertainty. The August 21 neocloud selloff — driven entirely by macro rate dynamics rather than any Nebius-specific news — illustrated how quickly the sector reprices when the cost of capital shifts. Customer concentration compounds the picture: the neocloud model depends on a small number of hyperscalers or frontier AI labs signing multi-year take-or-pay agreements, meaning any contraction in that demand base flows directly and rapidly through utilization into revenue.

Three signals will define the next chapter. First, the rate at which Groq 3 LPX systems translate into billable inference revenue: if low-latency inference pricing proves durable, these platforms could extend Nebius's margin expansion beyond its training-compute base and diversify its revenue mix in ways the current model does not yet reflect. Second, the sequential revenue trajectory through H2 2026 — the Q2 print of $582 million represents a strong absolute level, but sustaining comparable growth at scale will determine whether 454% year-over-year growth reflects genuine run-rate demand or a steep ramp still being absorbed by the market. Third, any concrete development in Meta's cloud ambitions: formal confirmation of a 'Meta Compute' offering would trigger immediate reassessment of revenue visibility across the entire neocloud sector. Nebius's fundamentals have, for now, outpaced both the skeptics and the macro headwinds. NVIDIA's reported hardware selection represents a meaningful signal of confidence from the dominant infrastructure supplier. But at $210.91 — trading below where the post-earnings surge began — the stock already prices in a level of execution that leaves limited room for downside surprises.