# NVIDIA Reports Impressive Growth Amid Rising Demand for AI and Automotive Solutions

Full stack optimizations for NVIDIA, supported by a vast ecosystem including 5.9 million developers, continue to enhance our customers’ business economics. In the fourth quarter, large cloud service providers (CSPs) accounted for about half of our data center revenue, which nearly doubled year-on-year. These major players quickly adopted the Blackwell architecture, deploying GB200 systems across Azure, GCP, AWS, and OCI to address surging AI demand.

Moreover, regional cloud hosting of NVIDIA GPUs has increased as a segment of our data center revenue. This uptick reflects the global expansion of AI infrastructure and a steady rise in demand for AI reasoning models and agents. We have successfully deployed a cluster of 100,000 GB200 units, leveraging NVLink Switch and Quantum 2 InfiniBand.

Consumer internet revenue has surged threefold compared to last year, driven by innovative generative AI and deep learning applications. These applications span diverse areas such as recommender systems, vision, language understanding, synthetic data generation, and agent-based AI. Notably, xAI has adopted the GB200 technology to train and deploy its next-generation Grok AI models.

Meta’s advanced Andromeda advertising engine operates on NVIDIA’s Grace Hopper Superchip, which efficiently serves large volumes of ads on Instagram and Facebook. This engine maximizes ad personalization and boosts inference throughput by three times, resulting in significant improvements in monetization and return on investment. Additionally, enterprise revenue nearly doubled, driven by escalating demand for model fine-tuning, RAG, agent-based AI workflows, and GPU-accelerated data processing.

The introduction of the NVIDIA Llama Nemotron model family aims to assist developers in creating and deploying AI agents across various applications, including customer support, fraud detection, and supply chain management. Leading AI agent platforms such as SAP and ServiceNow are among the early adopters of these new models. In healthcare, organizations like IQVIA, Illumina, Mayo Clinic, and the Arc Institute utilize NVIDIA AI to expedite drug discovery, enhance genomic research, and pioneer advanced healthcare services with generative and agent-based AI.

As AI applications broaden into robotics and physical AI, NVIDIA’s infrastructure and software are increasingly pivotal in these sectors. Virtually all autonomous vehicle (AV) companies rely on NVIDIA technology, whether in the data center or within vehicles. The automotive sector’s revenue is projected to reach approximately $5 billion this fiscal year. At CES, Hyundai Motor Group revealed its plans to integrate NVIDIA technologies to accelerate developments in AV, robotics, and smart factories. Innovations in vision transformers, self-supervised learning, multimodal sensor fusion, and high-fidelity simulation are anticipated to require tenfold increases in computational capacity.

At CES, we launched the NVIDIA COSMO World Foundation model platform, akin to language foundation models designed for physical AI to transform robotics. Early adopters include robotics and automotive firms, with ridesharing leader Uber among them. From a geographic standpoint, we observed the strongest sequential revenue growth in the U.S. due to the initial ramp-up of Blackwell, as countries globally build their AI ecosystems to keep pace with surging demand for computational resources.

France’s €100 billion AI investment and the European Union’s €200 billion AI initiatives signal significant global infrastructure growth in the coming years. However, data center sales in China remain significantly below pre-export control levels, and absent changes in regulations, shipments to China are expected to stabilize at current figures. The competitive landscape for data center solutions in China remains challenging.

We remain committed to complying with export controls while servicing our client base. In terms of networking, revenue experienced a 3% sequential decline, though over 75% of our networking sales were tied to GPU compute systems. We are transitioning from the smaller NVLink 8 with InfiniBand to the larger NVLink 72 with Spectrum-X. Revenue from Spectrum-X and NVLink Switch is on the rise, marking a new growth opportunity. We anticipate a rebound in networking growth in Q1, as AI necessitates advanced networking solutions.

NVIDIA offers NVLink Switch systems for scale-up computing needs, while our quantum incentive supports scale-out requirements for HPC supercomputers and Spectrum-X enhances Ethernet performance for AI applications. This innovation has enjoyed significant success, with Microsoft Azure, OCI, CoreWeave, and other companies constructing expansive AI factories with Spectrum-X. Furthermore, Cisco recently announced the integration of Spectrum-X into its networking portfolio, aiding enterprises in building AI infrastructure across various sectors.

Turning to the gaming segment, gaming revenue reached $2.5 billion but fell by 22% sequentially and 11% year on year. Nevertheless, full-year revenue of $11.4 billion increased by 9% annually, bolstered by strong demand during the holiday season. Q4 shipments were affected by supply constraints, but we anticipate strong sequential growth in Q1 due to improved supply conditions. Our new GeForce RTX 50 Series desktop and laptop GPUs are now available, designed for gamers, creators, and developers.

These GPUs, powered by the Blackwell architecture and incorporating fifth-generation Tensor cores and fourth-generation RT cores, show remarkable performance improvements. Notably, the new DLSS 4 technology can enhance frame rates up to eightfold through AI-driven frame generation. These GPUs also represent the industry’s first real-time application of transformer models, offering two times the parameters and quadrupling compute capabilities for enhanced visual fidelity.

Additionally, we unveiled a new line of GeForce Blackwell laptop GPUs featuring NVIDIA Max-Q technology, which extends battery life by up to 40%. These laptops are expected to hit the market in March, supplied by leading manufacturers. In our professional visualization segment, revenue reached $511 million, marking a 5% sequential increase and a 10% annual rise. Full-year revenue of $1.9 billion signifies a 21% year-on-year growth, fueled primarily by demand from the automotive and healthcare industries.

NVIDIA technologies, particularly those utilizing generative AI, are transforming design, engineering, and simulation workloads. Major software platforms from companies like ANSYS, Cadence, and Siemens are increasingly leveraging our RTX workstations. Turning to automotive, revenue has reached a record $570 million, up 27% sequentially and 103% year on year, with a full-year revenue of $1.7 billion showing a 5% increase.

At CES, we announced that Toyota, the world’s largest automaker, will utilize NVIDIA Orin to build its next-generation vehicles, running on the safety-certified NVIDIA DriveOS. We also revealed plans with Aurora and Continental to deploy driverless trucks at scale, powered by NVIDIA Drive Thor. Lastly, our comprehensive autonomous vehicle platform, NVIDIA Drive Hyperion, has successfully undergone safety assessments by industry leaders like TUV SUD and TUV Rheinland, making it the first AV platform to receive a complete set of third-party evaluations.

NVIDIA Reports Strong Demand and Outlook for Blackwell Architecture

Let’s examine the details of NVIDIA’s Profit & Loss statement (P&L). During the recent quarter, GAAP gross margin landed at 73%, while non-GAAP gross margin was slightly higher at 73.5%. This figures declined sequentially as anticipated, attributed to the initial deliveries of the Blackwell architecture. As noted in the previous quarter, Blackwell represents an adaptable AI infrastructure featuring various NVIDIA-built chips and networking options, suitable for both air and liquid-cooled data centers.

NVIDIA exceeded expectations in Q4 by ramping up the Blackwell system, boosting system availability, and offering multiple configurations. As Blackwell continues to scale, gross margins are projected to stabilize in the low 70s. In the early stages, the focus is on expediting Blackwell manufacturing to meet robust customer demand. Once fully operational, there will be numerous opportunities to reduce costs and enhance gross margins, which may recover to the mid-70s by late this fiscal year.

Operating Expenses and Shareholder Returns

Sequentially, GAAP operating expenses rose by 9%, while non-GAAP operating expenses upticked by 11%. These increases are due to heightened engineering development costs and expenses related to new product introductions. In Q4, NVIDIA returned $8.1 billion to shareholders through share buybacks and cash dividends. Looking ahead to the first fiscal quarter, total revenue is projected at approximately $43 billion, with a variance of plus or minus 2%.

Given continued strong demand, a substantial ramp-up of Blackwell is expected in Q1. Anticipated sequential growth is expected in both the Data Center and Gaming sectors. In the Data Center segment, growth should come from both computing and networking aspects. Gross margins are estimated to be around 70.6% for GAAP and 71% for non-GAAP, with a potential variance of 50 basis points.

Future Financial Outlook

Projected GAAP operating expenses and non-GAAP operating expenses are around $5.2 billion and $3.6 billion, respectively. For the full fiscal year 2026, operating expenses are expected to climb into the mid-30s. Meanwhile, GAAP and non-GAAP other income is anticipated to be about $400 million, not accounting for gains and losses tied to nonmarketable and publicly held equity securities.

In terms of taxation, GAAP and non-GAAP tax rates are expected to be around 17%, with a margin of plus or minus 1%, excluding discrete items. Comprehensive financial insights are available in the CFO commentary, as well as on the Investor Relations website, which now features a new financial information AI agent.

Upcoming Events for Financial Community

Looking ahead, NVIDIA has several key events planned for the financial community. The company will present at the TD Cowen Health Care Conference in Boston on March 3, followed by the Morgan Stanley Technology, Media, and Telecom Conference in San Francisco on March 5.

Furthermore, the Annual GTC conference is set to commence on Monday, March 17 in San Jose, California. CEO Jensen Huang will deliver a keynote address packed with news on March 18. A Q&A session for financial analysts will take place the following day, March 19. The earnings call to discuss results for the first quarter of fiscal 2026 is scheduled for May 28, 2025.

Now, we will open the call to questions.

Questions & Answers:

Operator

Thank you. [Operator instructions] The first question comes from C.J. Muse with Cantor Fitzgerald. Please go ahead.

C.J. Muse — Analyst

Good afternoon. Thank you for taking my question. Jensen, with the advancements in test-time compute and reinforcement learning, we see the lines between training and inference becoming increasingly blurred. How might this impact the future of inference-dedicated clusters, and what implications do you foresee for NVIDIA and your clients? Thank you.

Jensen Huang — President and Chief Executive Officer

I appreciate the question, C.J. Multiple scaling laws are in play. The pre-training scaling law continues to evolve, bolstered by multimodal data integrated into pretraining. Following that, we have the post-training scaling, where the computational demand could exceed that of pretraining due to reinforcement learning and large synthetic data generation.

The third consideration is test-time compute concerning reasoning and inference scaling. The required computational capacity for inference has already expanded significantly and could amplify even further in the future. This leads us to the challenge of constructing an architecture capable of accommodating various configurations, whether that’s auto-regressive models, diffusion-based models, or varied data center configurations.

We’re currently focusing our efforts on Blackwell to address these needs. Today’s inference computation outpaces our earlier capabilities, and Blackwell introduces a new level of performance for long-term reasoning tasks. Its architecture is versatile enough to adapt based on the workload—be it pretraining, post-training, or scaling out inference—making NVIDIA’s architecture desirable.

The trend toward a unified architecture is stronger than ever.

Operator

Your next question is from Joe Moore with Morgan Stanley. Please proceed.

Joe Moore — JPMorgan Chase and Company — Analyst

Thank you. Jensen, regarding the GB200 showcased at CES, you discussed the intricate nature of rack-level systems. How is the ramping process going? Are there still system-level bottlenecks beyond the chip production? Also, do you still have strong support for the NVL72 platforms?

Jensen Huang — President and Chief Executive Officer

My enthusiasm has actually grown since CES, especially as we have shipped considerably more units. We have around 350 plants manufacturing the 1.5 million components needed for each Blackwell rack, and the complexity is notable. We successfully ramped up Grace Blackwell, generating $11 billion in revenue last quarter. We are committed to scaling further as customer demand continues to rise.

Grace Blackwell Systems Gain Traction Amid Growing Demand for AI Infrastructure

Recent discussions have highlighted the rapid deployment of Grace Blackwell systems. Our organization has a significant installation of these systems supporting our engineering, design, and software teams. Other major players like CoreWeave, Microsoft, and OpenAI are also advancing their Grace Blackwell implementations, indicating a robust trend in the industry. Although the technology landscape is challenging, we feel confident about our progress and that of our partners.

Analyst Inquiry: Gross Margins and Future Demand

Operator

Your next question comes from Vivek Arya with Bank of America Securities. Please go ahead.

Vivek Arya — Analyst

Thank you for taking my question. Colette, could you confirm if Q1 represents the bottom for gross margins? Jensen, I’d like to know what indicators give you confidence that strong demand will persist into next year? Has DeepSeek or any innovations impacted that perspective? Thank you.

Focus on Gross Margins and Manufacturing Efficiency

Colette M. Kress — Chief Financial Officer, Executive Vice President

Let’s first address the gross margin aspect. Throughout the Blackwell ramp, we expect margins to settle in the low 70s. Our priority now is to expedite manufacturing processes, ensuring timely delivery to customers. Once we achieve full operational capacity with Blackwell, we anticipate improvement in both costs and gross margins, likely reaching the mid-70s later this year. As Jensen highlighted, the systems’ complexity allows for customization, including various networking options and cooling methods, which presents opportunities for enhancing these gross margins moving forward.

Demand Outlook and Future Software Landscape

Jensen Huang — President and Chief Executive Officer

We have a solid understanding of the capital investments being made in data centers. The shift toward machine learning is reshaping software architecture, making accelerated computing essential. We’re receiving positive feedback from our top partners alongside innovative startups that require significant computing infrastructure, indicating sustained demand. Demand indicators such as purchase orders and capital expenditures offer insights into near- and mid-term prospects. Long-term signals indicate a substantial shift from traditional coding on CPUs to machine learning and AI software running on GPUs.

Currently, consumer AI and generative AI applications are just scratching the surface. The upcoming wave involves agentic AI for enterprises, robotics with physical AI, and region-specific sovereign AI structures. All these developments are just beginning, and we see substantial activity across various sectors indicating a bright future for our industry.

Blackwell Ultra: Meeting Market Demand

Operator

Your next question comes from Harlan Sur with JPMorgan. Please go ahead.

Harlan Sur — Analyst

Good afternoon. Thank you for taking my question. Blackwell Ultra is expected to launch in the latter half of this year, consistent with your product release cycle. Jensen, can you elaborate on demand dynamics for Blackwell Ultra, especially as the current Blackwell solutions continue to ramp? How are your customers and the supply chain managing these simultaneous launches? Are you on schedule for Blackwell Ultra’s introduction?

Jensen Huang — President and Chief Executive Officer

Yes, Blackwell Ultra is on track for the second half of the year. The initial launch of Blackwell faced some delays, but we’ve made significant progress in our recovery efforts. As we ramp production for Blackwell, planning for Blackwell Ultra is also in motion. All necessary information has been communicated to our partners and customers for a smooth transition.

The architecture remains consistent between Blackwell and Blackwell Ultra, simplifying the transition compared to past shifts. We’ve also been working closely with partners on the development of future solutions, including the upcoming Vera Rubin. Expect substantial advancements and clarity on these developments at GTC.

Custom ASICs vs. Merchant GPUs

Operator

Your next question comes from Timothy Arcuri with UBS. Please proceed.

Timothy Arcuri — Analyst

Thanks! Jensen, it seems there is a lot of talk around custom ASICs. Could you discuss the balance between customer ASICs and merchant GPUs? Are customers planning to create heterogeneous superclusters utilizing both, or will these infrastructures largely remain separate?

Jensen Huang — President and Chief Executive Officer

We build different technologies compared to ASICs; there are significant distinctions. However, there are intersections in certain areas where our offerings complement one another.

NVIDIA Discusses Unique Architecture and AI’s Growing Market Impact

NVIDIA differentiates itself in various ways from competitors in the AI landscape. The company boasts a versatile architecture that caters to a wide array of models—be it diffusion-based, vision-based, multimodal, or text-focused. NVIDIA excels across these segments due to its robust software stack and architecture, positioning it as a key player in the latest innovations and algorithms. This broad capability sets NVIDIA apart as a more adaptable choice compared to narrow-focused competitors.

Furthermore, NVIDIA offers an end-to-end solution that encompasses the full spectrum of data processing—from the curation of training data to training itself, and even reinforcement learning post-training to inference with challenging time scaling. This comprehensive approach ensures that NVIDIA’s offerings are accessible across different environments, whether it be in the cloud, on-premises, or integrated into robotics. Because of this flexibility, NVIDIA becomes an attractive option for startups venturing into the AI space.

Another significant advantage is NVIDIA’s remarkable performance. The company highlights that data centers operate within fixed constraints of size and power. With performance per watt potentially increasing by 2x to 8x, the revenue implications can be substantial. For instance, in a 100-megawatt data center, achieving higher throughput translates directly to increased revenues. This model differs from traditional data centers, as revenue generation in AI factories is directly linked to the tokens generated, making NVIDIA’s architecture essential for companies focusing on profitability and quick returns on investment.

NVIDIA’s fast performance coupled with a complex software ecosystem makes it a leader in AI technology. The advancement of the ecosystem surpasses that of just two years ago and is pivotal as AI rapidly evolves. Effectively managing this expansive software framework which sits atop the hardware is a demanding task that fuels NVIDIA’s competitive edge.

However, it’s essential to note that designing chips does not guarantee their deployment. The industry has witnessed numerous instances where innovations in chip design have faltered during rollout due to business decision-making regarding their integration into size- and power-constrained AI factories. NVIDIA’s technology excels not only in performance but also boasts enhanced software capabilities and fast deployment speeds, which are crucial in this evolving market dynamics.

Operator

Your next question comes from Ben Reitzes at Melius Research. Please proceed.

Ben Reitzes — Analyst

Hi, Ben Reitzes here. Thank you for the opportunity. Jensen, my question concerns geography. You did a commendable job detailing the demand factors, particularly noting that the U.S. increased by roughly $5 billion sequentially. There’s some concern about whether the U.S. can adequately compensate if regulations affect other markets. Can you elaborate on the sustainability of this surge in the U.S. and its implications for your growth rate? It seems like China remains steady as well; could you clarify that dynamic?

Jensen Huang — President and Chief Executive Officer

China’s contribution remains consistent with prior quarters, approximately at half of pre-export control levels. In terms of geography, it’s vital to recognize that AI is fundamentally rooted in software.

This modern software has permeated various sectors, including delivery and shopping services. From purchasing daily goods like milk to educational tools, AI plays a critical role across industries like healthcare, finance, and environmental technology. As a result, it is reasonable to conclude that AI has become mainstream and integrates into numerous applications. We hope this technology progresses in a safe and constructive manner for society. The current shift indicates that we are at the onset of a new phase in technology.

This transformation is not just a minor enhancement; it marks a departure from decades of development focused on conventional computing architectures like CPUs. Moving forward, AI will form the backbone of software and services, emphasizing machine learning. The convergence of data and software will drive advancements within this landscape, marking a significant transition that has already begun.

Finally, no other technology has previously had the potential to impact the global economy significantly, similar to AI’s capabilities today. This software tool targets a larger segment of the global GDP than ever before. Therefore, our perspective on growth must be framed within this context, emphasizing that we are merely at the beginning of this expansive journey.

Operator

Your next question comes from Aaron Rakers with Wells Fargo. Please go ahead. Aaron, your line is open.

Your next question comes from Mark Lipacis with Evercore ISI. Please proceed.

Mark Lipacis — Analyst

Hi, it’s Mark. Thank you for taking my question. I need clarification on a point. Did you mention that enterprise growth within the data center doubled year-over-year for the January quarter? If so, does this indicate that it is growing faster than hyperscalers? And, Jensen, considering that hyperscalers are the primary consumers of your solutions, do they procure equipment for both internal and external workloads? Specifically, I’m interested in how their external workflows serve enterprise customers.

Understanding Hyperscaler Spending Trends and AI’s Role in Enterprise Growth

Colette M. Kress — Chief Financial Officer, Executive Vice President

Thank you for your inquiry regarding our enterprise business. It has doubled in size, mirroring trends observed with our large Cloud Service Providers (CSPs). Both areas are crucial for understanding current dynamics.

Collaboration with CSPs often involves projects related to large language models and inference. However, enterprises are also leveraging these resources, either through CSPs or by developing their own capabilities. It’s noteworthy that both segments are experiencing robust growth.

Jensen Huang — President and Chief Executive Officer

CSPs constitute about half of our business, encompassing both internal and external consumption. Our collaborations aim to optimize internal workloads since CSPs utilize significant infrastructures built with NVIDIA technology.

Our technology serves diverse applications, from AI to video processing and data analytics. This flexibility extends the useful life of our infrastructure, which in turn decreases total cost of ownership (TCO). As for future growth, I believe that the enterprise segment will ultimately be larger than CSPs. The current computer industry leaves a considerable portion of industrial needs unaddressed.

For context, let’s consider the automotive industry. Car manufacturers engage in both physical and software production. In the realm of enterprise, our solutions are designed to enhance employee productivity through agentic AI—systems that empower workers to plan, market, and operate more efficiently.

Conversely, the vehicles produced necessitate a sophisticated AI system for fleet management and training. Currently, there are around 1 billion vehicles on the road, and we anticipate a future where all of these will be autonomous, continuously collecting data to improve operations through what we term an AI factory.

In the evolution of manufacturing, traditional car factories will coexist with AI factories. The cars themselves are equipped with complex systems requiring advanced AI, which we refer to as physical AI. This type of AI goes beyond understanding words; it must grasp physical concepts such as friction, inertia, and cause-effect relationships. This is a new frontier, and we see great potential in the intersection of agentic AI and physical AI within enterprises.

The world’s economic output is largely driven by industrial and heavy industries, highlighting the significance of this shift.

Operator

Your next question comes from Aaron Rakers with Wells Fargo. Please proceed.

Aaron Rakers — Analyst

Thank you for the opportunity, Jensen. As we approach two years since the Hopper milestone in 2023 and consider your strategic roadmap, how do you view the existing infrastructure in terms of its replacement cycle, especially relating to models like GB300 or any cycles you foresee for the Rubin framework?

Jensen Huang — President and Chief Executive Officer

I appreciate your question. Many clients continue to operate on generations like Voltas and Pascals due to the adaptability of CUDA technology. For instance, one of the current major applications for the Blackwell architecture is data processing and curation.

When you identify situations where an AI model is faltering, you can utilize a vision language model to analyze these situations. This model can provide insights into the failure, which can then be used to enhance your overall AI training process. The process involves gathering data from these instances, then training new models on enhanced datasets, facilitating renewals in your tech stack.

Every architecture is CUDA-compatible, allowing for workload distribution across installed infrastructures. Thus, while newer systems handle intensive tasks, existing architectures remain effective for less demanding workloads.

Operator

We have time for one last question. Atif Malik from Citi, please go ahead.

Atif Malik — Analyst

Thank you for taking my question, Colette. I’d like to follow up regarding gross margins. Given the various dynamics involving Blackwell yields and NVLink, you suggested that the April quarter might represent a low point. Can you elaborate on the expected trajectory to reach mid-70s gross margins by the end of the fiscal year?

Colette M. Kress — Chief Financial Officer, Executive Vice President

Thank you for the question. Regarding our gross margins…

NVIDIA CEO Jensen Huang Discusses Blackwell’s Potential and AI Trends

The conversation around Blackwell’s system highlights significant opportunities for enhancing gross margins over time. NVIDIA’s diverse configurations within the Blackwell platform are poised to support this initiative effectively.

The company is currently focused on ramping up production for customers. Following this phase, NVIDIA anticipates making substantial improvements as soon as possible. If short-term enhancements are feasible, they will be implemented promptly. Currently, uncertainty surrounds tariffs, depending on the U.S. government’s forthcoming plans regarding timing and scope. NVIDIA remains committed to following export controls and any related tariffs.

Operator

Ladies and gentlemen, that concludes our question-and-answer session. Thank you.

Jensen Huang — President and Chief Executive Officer

I appreciate it. Thank you, Colette. The demand for Blackwell remains exceptional. Artificial Intelligence (AI) is advancing rapidly from basic perception and generative AI into reasoning capabilities.

With the emergence of reasoning AI, another scaling law is taking shape: inference time scaling. Increased computation leads to smarter outcomes. Models like OpenAI’s, Grok 3, and DeepSeek-R1 utilize reasoning that significantly amplifies their processing needs, requiring up to 100 times more compute resources.

Looking ahead, future reasoning models are expected to demand even greater computational power. DeepSeek-R1 has generated global interest, representing a significant innovation in AI. Its open-sourced framework allows AI developers to leverage reasoning models, improving their applications using techniques such as reinforcement learning and the chain of thought model.

Three key scaling laws drive the demand for AI computing. While traditional AI scaling remains relevant, enhancements like multimodality in foundation models are on the rise. However, the landscape is evolving past pre-training scaling alone. There are now two new dimensions: post-training scaling, which involves reinforcement learning, fine-tuning, and model distillation, and inference time scaling, which illustrates an immense increase in computational needs for a single query.

Blackwell is designed to meet these evolving demands. It provides a seamless transition between pre-training, post-training, and test time scaling. Its FP4 transformer engine, NVLink 72 scale-up fabric, and advanced software technologies enable Blackwell to process reasoning AI models 25 times faster than the previous Hopper architecture.

In terms of production, the Blackwell configuration is fully operational. Each Grace Blackwell NVLink 72 rack is a feat of engineering, comprising 1.5 million components manufactured across 350 sites by roughly 100,000 operators. AI is progressing at an unprecedented pace.

We have entered the early stages of reasoning AI and inference time scaling, with multimodal AIs, enterprise AI, sovereign AI, and physical AI on the horizon. As we look forward, data centers are forecasted to allocate most of their capital expenditures toward accelerated computing and AI.

In the future, data centers will evolve into AI production facilities, which will either be rented or managed internally by companies. I want to thank everyone for joining us today. I invite you to attend GTC in a few weeks, where we will discuss Blackwell Ultra, Rubin, and numerous developments in computing, networking, reasoning AI, and physical AI products.

Thank you.

Operator

[Operator signoff]

Call Participants:

Stewart Stecker — Senior Director, Investor Relations

Colette M. Kress — Chief Financial Officer, Executive Vice President

C.J. Muse — Analyst

Jensen Huang — President and Chief Executive Officer

Joe Moore — JPMorgan Chase and Company Analyst

Vivek Arya — Analyst

Harlan Sur — Analyst

Timothy Arcuri — Analyst

Ben Reitzes — Analyst

Mark Lipacis — Analyst

Aaron Rakers — Analyst

Atif Malik — Analyst

More NVDA analysis available.

All earnings call transcripts can be found here.

This transcript is produced for The Motley Fool. We strive for accuracy, but errors may occur. Users should independently verify content before making decisions and review the company’s SEC filings. Please see our Terms and Conditions for disclaimers and additional details.

The Motley Fool recommends Nvidia and has positions in it.

The views expressed here are those of the author and do not necessarily reflect the views of Nasdaq, Inc.