NVIDIA unveils Blackwell Ultra for next-gen AI

SAN JOSE, Calif. - NVIDIA (NVDA), the semiconductor giant with a market capitalization of $2.85 trillion and impressive revenue growth of 114% over the last twelve months, has announced the Blackwell Ultra, the latest advancement in its AI factory platform, designed to enhance AI reasoning and scaling inference capabilities for a variety of applications. According to InvestingPro, NVIDIA maintains a "GREAT" financial health score, positioning it strongly to continue leading the AI revolution. The Blackwell Ultra, building on the existing Blackwell architecture, is expected to deliver 1.5 times the AI performance of its predecessor and significantly increase revenue opportunities for AI factories.

The new platform includes the NVIDIA GB300 NVL72 rack-scale solution and the NVIDIA HGX B300 NVL16 system, both aimed at improving the quality of AI services by providing more computing power for inference. The GB300 NVL72 connects 72 Blackwell Ultra GPUs with 36 ArmNeoverse-based NVIDIA Grace CPUs, effectively acting as a massive GPU to support complex problem-solving and enhance response quality.

NVIDIA’s founder and CEO, Jensen Huang, emphasized the leap in computing performance required for reasoning and agentic AI, stating that Blackwell Ultra is designed for these demands, capable of handling pretraining, post-training, and reasoning AI inference efficiently.

The Blackwell Ultra platform is also optimized for agentic AI, which involves sophisticated reasoning and planning, as well as physical AI, which can generate synthetic, photorealistic videos in real-time for training robots and autonomous vehicles at scale. With an industry-leading gross profit margin of 75% and strong cash flows that easily cover its obligations, NVIDIA continues to demonstrate its ability to deliver high-value innovations while maintaining robust profitability.

In terms of networking, Blackwell Ultra systems will integrate with NVIDIA Spectrum-X Ethernet and NVIDIA Quantum-X800 InfiniBand platforms, providing 800 Gb/s data throughput for each GPU and reducing latency and jitter.

The announcement also introduced NVIDIA Dynamo, an open-source inference framework that aims to scale up reasoning AI services, improve throughput, and reduce response times and model serving costs.

The Blackwell Ultra-based products are expected to become available from various partners, including Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro, starting from the second half of 2025. Major cloud service providers like Amazon Web Services, Google Cloud, Microsoft Azure, and Oracle Cloud Infrastructure, along with GPU cloud providers, are set to offer Blackwell Ultra-powered instances.

This news is based on a press release statement from NVIDIA, and it marks a significant step in the company’s efforts to advance AI infrastructure and performance. For investors seeking deeper insights into NVIDIA’s valuation and growth prospects, InvestingPro offers comprehensive analysis through its Pro Research Report, available among 1,400+ top stocks, along with 20+ additional exclusive ProTips that can help inform investment decisions.

In other recent news, NVIDIA has unveiled NVIDIA Dynamo, an open-source software aimed at enhancing the efficiency of AI inference in AI factories. This new platform is designed to optimize AI inference requests across extensive GPU networks, reportedly doubling the performance and revenue of AI factories using the same number of GPUs. NVIDIA’s collaboration with telecom giants like T-Mobile and Cisco to develop AI-driven 6G networks further underscores its commitment to integrating AI into next-generation wireless infrastructure. These partnerships focus on creating AI-native wireless networks that promise improved spectral efficiency and new revenue opportunities for telecom companies.

In terms of analyst ratings, Truist Securities has maintained its Buy rating on NVIDIA with a price target of $205, expecting the company’s GTC event to potentially boost investor confidence. Similarly, UBS has also reaffirmed a Buy rating with a $185 target, noting that NVIDIA’s GPU shipment projections remain steady despite changes in product mix. UBS adjusted its revenue estimates for NVIDIA’s first fiscal quarter to approximately $46 billion, with earnings per share projected at $5.27 for 2025 and $6.22 for 2026.

Looking ahead, NVIDIA’s GTC 2025 conference is expected to focus on AI servers, with potential announcements about the B300 AI chip and data center networking solutions. Analyst Ming-Chi Kuo suggests that the conference could be a catalyst for a short-term rebound in AI stocks. The B300 chip is anticipated to offer significant performance improvements, with trial production slated for the second quarter of 2025.

This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.