Nvidia's Big Beat, Cooler Read-Through: Why the Stock Slipped

$NVIDIA(NVDA)$   reported FY26 Q2 and guided Q3 above Wall Street expectations, yet the shares fell ~3% after hours—classic "great numbers, tougher bar." Here's the cheat sheet.

Q2 Key Financial Highlights

~Revenue: $46.7B, +56% YoY and up sequentially, above consensus.

~GAAP Gross Margin: 72.4% (–270 bps YoY, +1,190 bps QoQ), above prior guidance. Non-GAAP: 72.7% (–300 bps YoY, +140 bps QoQ), also above guidance.

~GAAP Net Income: $26.4B (+59% YoY, +41% QoQ), above consensus. Non-GAAP Net Income: $25.8B (+52% YoY, +30% QoQ), above consensus.

Q2 Revenue Breakdown by Platform

~Data Center (AI compute + networking): $41.1B, +56% YoY, +5% QoQ, 88% of total.

~AI Compute: $33.8B, +50% YoY, –1% QoQ, mainly reflecting the H20 halt and the Hopper→Blackwell transition; Blackwell products rose +17% QoQ in Q2.

~Networking: $7.3B, +98% YoY, +46% QoQ.

~Gaming (desktop/laptop GPUs, Switch SOCs): $4.3B, +49% YoY, +14% QoQ, 9% of total.

~Professional Visualization (workstation GPUs, AI software): $0.6B, +32% YoY, +18% QoQ, ~1%.

~Automotive (ADAS/robotics edge AI): $0.6B, +70% YoY, +3% QoQ, ~1%.

Three Things to Watch

1. Networking opens a second growth curve; compute's product swap clipped Q2, Blackwell Ultra ships in Q3

The surprise inside Data Center was the split between compute and networking.

~Compute revenue was $33.8B, +50% YoY but –1% QoQ, triggering concern about "AI-chip momentum." Management attributes this to the H20 restriction and a Hopper→Blackwell hand-off; Blackwell still ramped fast, with Q2 Blackwell revenue +15% QoQ.

~Nvidia's Q3 revenue guide is up $7.3B QoQ, signaling large-scale Blackwell shipments, with Blackwell Ultra (GB300) also set to ship.

Q2 Networking revenue hit $7.3B (+98% YoY, +46% QoQ)—a quarterly level that surpasses any standalone networking-chip vendor, including $Broadcom (AVGO.US)$ . Spectrum-X Ethernet delivered double-digit QoQ and YoY growth with annualized revenue >$10B, while InfiniBand revenue nearly doubled QoQ. The driver: GB200 NVL72 rack deployments that expand cross-sell of network silicon. As AI chips move to rack-level systems, networking becomes Nvidia's second growth engine in Data Center.

2. China is still outside the plan

China revenue fell QoQ to 6% of invoiced revenue. Singapore accounted for 22% of invoiced revenue, >99% serving U.S. Customers.

Because geopolitical issues persist, Nvidia did not include any H20 in its Q3 outlook. If policy clears, Q3 H20 revenue could reach $2–$5B; Nvidia says it can produce more if orders arrive and continues to advocate U.S. approval to sell Blackwell into China.

3. Blackwell Ultra and Rubin stay on schedule

Because GB300 (Blackwell Ultra) shares architecture, software, and physical form factor with GB200, the shift is seamless. Production lines switched in late July/early August to support the GB300 ramp, at ~1,000 racks per week. With more capacity coming online, Q3 output should accelerate.

Next-gen Rubin chips have entered the fabs—including Vera CPU, Rubin GPU, CX9 SuperNIC, NVLink 144 scale-up switch, Spectrum-X scale-out/scale-across switches, and silicon-photonics processors. Rubin remains on schedule for mass production next year.

Q3 Guidance

~Revenue: $54.0B (above prior ~$53.5B consensus).

~Non-GAAP gross margin: 73.5%.

~Non-GAAP net income: $30.1B.Importantly, no H20 to China is assumed, yet revenue is up $7.3B QoQ, a company record, driven by Blackwell's ramp. Management continues to target ~75% Non-GAAP GM in Q4.

Summary

This is another historic print. If you're nit-picking, Q2 compute was down QoQ, but the Q3 setup points to a strong re-acceleration as Blackwell scales and networking compounds. Near-term, the stock will trade the expectations game; long-term, management pegs 2025 AI infra spend around ~$600B, rising to $3–4T by 2030, leaving a vast runway.

@TigerWire  @Daily_Discussion  @TigerWire  @Daily_Discussion  @Tiger_chat  @Tiger_comments  @MillionaireTiger