Jefferies Forecasts NVIDIA Q2 Revenue at $95 Billion, Potentially the Largest Beat in History
Nashnova编辑部
Jefferies expects Nvidia's Q2 revenue to hit $95 billion, roughly $3 billion above consensus — a beat that, if confirmed, would be the largest single-quarter upside surprise in the company's history and a signal that AI compute demand is still accelerating.
Where does the $95 billion forecast come from?
Jefferies projects Nvidia's Q2 revenue (fiscal quarter ending July 2026) at $95 billion, versus the Street consensus of $92.07 billion — a gap of roughly $3 billion.
Last quarter, Q1 revenue of $81.62 billion already beat consensus by $2.5 billion, setting a record at the time. This means → if the call is right, Nvidia will have broken its own beat record two quarters running.
In plain terms = the market's expectations are already stretched, and Jefferies is saying the actual number will stretch them further.
The Q3 forecast is even bolder — and Vera Rubin isn't even in it?
Jefferies' Q3 estimate is $108 billion, roughly $4.3 billion above consensus.
Analyst Blayne Curtis explicitly noted that both the Q2 and Q3 forecasts do not include material contribution from the Vera Rubin system.
This means → the current numbers are almost entirely carried by the existing Blackwell product line. Once Rubin ramps, there is room to revise higher still.
When does Rubin actually ramp?
Jefferies estimates VR/R200 will account for 12% of Q3 GPU revenue, rising to over 40% by Q4.
Management has confirmed Rubin is in volume production and shipping; fiscal Q1 2028 will be a "very significant" inflection point — at which Rubin is expected to surpass Blackwell as the primary revenue driver.
In plain terms = Rubin is on the runway now. The real acceleration comes at year-end and into early next year.
What changed in Rubin's design, and why is deployment so much faster?
The Vera Rubin NVL72 retains the 72-GPU rack form factor, shares physical dimensions and deployment infrastructure with GB200/GB300, and its 45°C warm-water cooling plugs directly into existing liquid-cooled data centers.
The critical change: compute trays are now cable-free, hose-free, fan-free modular units, replacing the extensive hand-installed NVLink cabling in Blackwell with a central PCB. This means → assembly time drops from roughly two hours to five minutes.
In plain terms = what used to feel like hand-plumbing a machine now snaps together like building blocks — more than 20× faster.
Every frontier AI lab plans to use Rubin from day one — what does that signal?
Jefferies notes that all frontier AI labs are expected to adopt Vera Rubin from the outset, unlike the gradual transition when Blackwell launched.
Case in point: the OpenAI–SB Energy partnership announced August 17. The campus will deploy Nvidia GPUs, CPUs, and networking gear, starting at 4.25 GW of IT capacity with an additional 3.75 GW expansion option. Nvidia will invest $1.5 billion in SB Energy and provide credit support for the initial build.
This reflects a deeper shift — Nvidia is moving from "selling chips" to "using its balance sheet to lock in marquee customers."
Nvidia is putting its own money in — what should investors watch?
Curtis flagged that Nvidia's $1.5 billion investment plus credit support for SB Energy is a material change to the company's balance sheet.
This comes just one week after Nvidia's $500 billion third-party financing platform was characterized as "moving customer financing off the balance sheet." This means → it may reignite investor concerns about potential circular financing.
In plain terms = Nvidia is helping customers fund data-center builds while also writing checks itself. Whether that's a show of confidence or a shift of risk onto its own books is the question the market will answer after next week's earnings.
Content is for reference only, not financial advice.