Nvidia goes 'full steam ahead' on Blackwell chips

Just days after reports suggested its next-gen AI servers were running into serious cooling issues, Nvidia delivered an earnings report that basically told Wall Street not to worry.

In its Q3 fiscal 2025 earnings call, the company brushed aside reports from The Information about server overheating by simply not addressing them. Instead, management focused on a much louder message: Blackwell is in full production, moving full steam ahead, and already racking up orders.

Moving iron while the rumors swirl

Nvidia CFO Colette Kress revealed that the company has already shipped 13,000 Blackwell samples to customers this quarter.

CEO Jensen Huang told investors that Blackwell’s success can already be measured in the billions of dollars, adding that the hardware is in great shape as customer systems are actively stood up.

Meanwhile, the financial engine powering this momentum is staggering. Nvidia brought in $35 billion in revenue for the quarter, with $30.7 billion coming purely from AI data centers. Gaming, the business that made Nvidia famous, brought in just $2 to $3 billion.

Profitability is climbing just as fast. Net profit hit $19.3 billion this quarter, up from $16.6 billion in Q2 and $14.8 billion in Q1. For comparison, tech titans Microsoft and Apple posted profits of $24.7 billion and $21.4 billion, respectively.

What is equally remarkable is how Nvidia manages overlapping generations. While Blackwell grabs headlines, the H200 announced last year has become the fastest-growing product in the company’s history, generating multiple billions of dollars this quarter alone. Huang expects demand for the entire H-series lineup to remain strong through most of next year.

Why it matters

We think this earnings call proves how masterfully Nvidia manages both its product pipeline and market psychology.

For enterprise customers and AI labs, the anxiety around Blackwell cooling bugs was real. If next-gen deployments stalled, compute roadmaps would break. By shipping 13,000 test units while keeping H200 production firing on all cylinders, Nvidia showed buyers they do not need to pause capital expenditure. Even if Blackwell takes time to ramp, the current generation still delivers what customers need.

For the semiconductor industry, Nvidia and rival AMD have rewritten the playbook by shifting from a traditional two-year cadence to an aggressive annual release cycle. This relentless pace keeps rivals like Intel on the defensive and forces competitors to burn cash just to stay in the conversation.

However, this breakneck speed carries genuine risks. Compressing design and deployment cycles to twelve months pushes thermal and electrical limits to the edge, making engineering slip-ups far more likely. More importantly, as hyperscalers accumulate layers of H100s, H200s, and soon Blackwell racks, the pressure on software developers to show tangible software revenue against these massive infrastructure bills will only intensify.