Key takeaways

• NVIDIA's dominance in the AI space is a result of its first-mover advantage, having released its proprietary CUDA architecture, suitable for deep learning applications, back in 2006. The company differentiated itself via its full-stack approach, creating a host of software libraries tailored to different AI applications, alongside its hardware-based offering for AI training and inference. 

• NVIDIA's early activities in the space were concentrated on creating chips to power autonomous vehicles for leading automakers including Mercedes-Benz, Volvo, and Audi. This however failed to attract significant traction given the limited revenue potential of the industry relative to costs, alongside safety concerns associated with driverless vehicles.

• Since 2015, the company has consistently launched deep learning-focused graphics processing units (GPUs) every year. These GPUs encompass those used in its data center operations, specifically for Edge Computing and Machine Learning (ML) Infrastructure. Notable in-house developments include the H100 AI chip, used in its DGX Cloud servers. Additionally, the company supplied its AI chips to major cloud providers including Microsoft, Amazon, and Google. As a result of these initiatives, the company’s data center revenues as a % of total revenue grew from ~20% in FY2018 to 56% in FY2023.

• The company also gained exposure within the Manufacturing sector via the launch of NVIDIA Omniverse, which allowed for the creation of AI-enabled Digital Twin simulations. This complemented the company’s activities in other industries, such as in Auto Tech, with manufacturers using NVIDIA's hardware stack for autonomous vehicles, alongside its simulations to improve driver safety.

• Partnerships in the Data Infrastructure & Analytics industry included IBM, Oracle, and HP to launch a platform to streamline GPU use for ML models, alongside the launch of generative AI cloud services, which were leveraged by Shutterstock and Getty Images to create custom text-to-image LLMs.

• However, NVIDIA's market position is being challenged by the likes of AMD, which is set to debut its MI300X AI training chip in Q4 2023, with Amazon, Intel, and Google trying to secure additional market share with the launch of similar chips. Other risk factors include the company’s sole reliance on TSMC to manufacture its chips, resulting in GPU supply bottlenecks, and a potential tightening of AI chip export rules to China—one of the company’s major markets.

NVIDIA’s evolution: From graphics giant to AI infrastructure titan

Not a comprehensive list of NVIDIA AI products and services

Source: Compiled by SPEEDA Edge based on multiple sources

NVIDIA AI-related in-house developments

NVIDIA’s AI-related partnerships, investments, and M&As

• The company’s AI activities across emerging industries witnessed a sharp increase from 2021, reaching its peak in 2023 (47). While the Transportation & Logistics hub held sway over activity levels from 2017 to 2020, this recent uptick can be attributed to the Information Technology hub, which saw growing engagement due to heightened interest in ML hardware. Meanwhile, the noticeable slowdown in activity in 2020 was likely due to Covid-19-related supply-chain constraints, which resulted in a widespread shortage of consumer-grade GPUs. This, coupled with the added demand from the cryptocurrency boom, pushed the company to focus on bridging the supply gap. 
• Interest in acquisitions was heavily concentrated in 2020, with NVIDIA expanding its operations through the acquisitions of startups in the ML infrastructure and edge computing spaces. In the investment realm, 2021 saw the company establish the Inception VC Alliance. This initiative aimed to link AI startups with venture capital, leading to an upswing in investment-related activity that year. Additionally, the heightened activity in 2023 is likely related to the increasing venture capital interest in training and inference for LLMs. 

NVIDIA’s key AI focus areas in detail

• A major revenue driver for NVIDIA was selling its AI chips and infrastructure in bulk to Big Tech firms running cloud hosting services. This included Microsoft, Amazon, and Google, alongside startups such as CoreWeave and Snowflake. Additionally, the company made its DGX Cloud available to customers of third-party companies, including Hugging Face, while other partnerships, such as the ones with Weights & Biases and Run:ai, primarily focused on optimizing its GPUs for AI training and inference.
• Together with its partnerships with Dell and Microsoft, this complemented the company’s activities in the Edge Computing space by bringing AI inference and training to the edge. 
• For Generative AI Applications, NVIDIA partnered with Shutterstock and Getty Images to create custom versions of the company’s Picasso cloud-based text-to-image service, trained using its proprietary data. Additionally, the company partnered with Adobe to co-develop GenAI models, which will be brought to market via Adobe’s Creative Cloud and NVIDIA’s Picasso. Given recent product launches, the company will also likely seek future partnerships for the provision and development of NVIDIA NeMo and BioNeMo.
• NVIDIA also partnered with leading automotive companies such as Uber, Mercedes-Benz, BMW, Audi, Volvo, Bosch, and Jaguar Land Rover to power autonomous vehicles or for infotainment purposes. 
• Other partnerships include those with Siemens and Kroger for digital twin simulations, alongside AI-drug-discovery-related partnerships with GlaxoSmithKline, AstraZeneca, and Recursion.

1. IT: Powering the AI revolution at the edge

• NVIDIA’s activities in this space are primarily driven by the ML Infrastructure industry, which accounted for ~84% of activities within the IT vertical.

Note: Companies with multiple partnerships are indicated numerically

Source: Compiled by SPEEDA Edge

• NVIDIA had a notable head start in the AI training and inference space compared with competitors such as AMD, which launched its CUDA-alternative, the ROCm deep-learning framework, in 2017. During this time, NVIDIA already had a few generations’ worth of products in play and AMD had to play catch-up while also focusing on the CPU market. Meanwhile, NVIDIA sought to improve its market position by expressing interest in acquiring chip manufacturer Arm in 2021 for USD 40 billion. However, the deal failed, given the significant regulatory challenges.

Source: Compiled by SPEEDA Edge

• Notable acquisitions to augment its AI infrastructure capabilities included that of data center chip company Mellanox Technologies for ~USD 6.9 billion in 2020. This was done to support its data center infrastructure at a time when efficient multi-node GPU communication was required to address the growing complexity of AI models. Mellanox’s Infiniband networking tech was also leveraged to provide connectivity to its A100 GPUs and DGX systems, with the NVIDIA CEO stating that the acquisition was “one of the greatest strategic decisions” the company had ever made. 
• Other acquisitions that complemented its AI strategy include 1) SwiftStack (2020) to leverage its tech and accelerate performance in training ML models on GPU clusters running across cloud environments, 2) Excelero (2022) to integrate high-performance storage solutions with AI and ML, and 3) OmniML (2023) to offer scalable deep learning models.
• The company’s data center business reaped the rewards of these initiatives, making up over 50% of NVIDIA’s revenue from FY2023 onwards (~20% in FY2018) on the back of strong demand for cloud adoption. 
• This fits the company’s partnership trend relating to ML infrastructure, which gained momentum in 2021. It also participated in the USD 100 million funding rounds of integrated MLOps solutions providers H2O.ai (November 2021) and Domino Data Lab (October 2021).

2. Transportation & Logistics: Autonomous driving hits speed bumps

• NVIDIA’s activities in the space are primarily driven by the Auto Tech sector, which accounts for ~90% of activities within this vertical. The larger amount of activity in this field compared with other sectors can be attributed to NVIDIA's initial emphasis on it, which led to the introduction of the Tegra K1 system-on-a-chip (SoC) in 2014, designed to power autonomous vehicles. 
• Tegra processors were initially used to power mobile phones, with NVIDIA having also acquired Icera, a wireless modem provider for mobile phones, in June 2011. However, given the stiff competition from Qualcomm’s Snapdragon processor, NVIDIA pivoted its focus to autonomous vehicles, discontinuing the use of Tegra processors for mobile phones, and subsequently wrote off its acquisition of Icera in 2015.

Note: Companies with multiple partnerships are indicated numerically

Source: Compiled by SPEEDA Edge

• The change in strategy proved fruitful for the company at the time, with 19 car manufacturers including BMW, Mercedes-Benz, Google, and Toyota having announced milestones or plans in 2015/2016 in terms of driverless car tests. 
• To capitalize on this demand and expand the company’s ecosystem, NVIDIA introduced the DRIVE PX platform in 2015, a developer kit for training self-driving vehicles. At the time, the company claimed that this was 3,000x faster than DAVE, the autonomous vehicle technology developed by DARPA. In 2016, the company launched the DRIVE PX 2, which had improved processing power, and DriveNet, the company’s own neural network for vehicle-based object recognition.
• Although the company enjoyed its early-mover position in the space, it faced hurdles when a self-driving Uber using NVIDIA’s tech was involved in a fatal car crash in 2018. The company, alongside Uber and Toyota, temporarily suspended testing of self-driving cars, with Auto Tech partnerships seeing a noticeable drop-off in the years that followed. 
• The notable developments that followed to bolster the safety of driverless vehicles included 1) the launch of Drive Constellation (2019), a simulation platform for testing and developing autonomous vehicles, which became a part of the company’s Omniverse platform; 2) the acquisition of mapping startup DeepMap (2021), providing the company’s autonomous stack with greater abilities to locate itself on roads; and 3) the company’s investment in Foretellix’s (a safety and driving assist tools provider for autonomous vehicles) USD 43 million funding round (2023). 
• However, despite these initiatives, the road became rockier for NVIDIA, as interest in the Auto Tech space waned following insufficient revenue generation from autonomous driving. This included Ford and Volkswagen shutting down their multi-billion-dollar self-driving initiatives in 2022, with other prominent auto manufacturers deprioritizing development efforts. As such, revenue from the sector remained constrained for NVIDIA, accounting for just ~3% of total revenue in FY2023.

3. Pharma & Life Sciences: Accelerating AI drug discovery

Source: Compiled by SPEEDA Edge

• NVIDIA’s activities in this space are primarily driven by its medical imaging and genomics platform, Clara, having increasingly been used during the Covid-19 pandemic to accelerate vaccine development.
• Notable feature additions to Clara include 1) Clara Guardian, an application framework and multimodal AI used for patient care (May 2020) and 2) Clara Holoscan MGX, which allows medical sensors to train AI algorithms and visualize biology in real time (March 2022). Additionally, NVIDIA acquired sequencing analysis software developer Parabricks in 2020 to double down on providing accurate genome analysis tools. This software stack complements the company’s AI infrastructure, running on its EGX edge computing platform and the DGX Cloud to provide a full-stack solution for the healthcare industry. 
• To place additional bets on the sector, NVIDIA invested USD 50 million to speed up training of Recursion's AI models for drug discovery. The company stands to benefit from potentially licensing Recurison’s AI models on its cloud service for GenAI-based drug discovery, BioNeMo

4. Manufacturing: Creating the Matrix with realistic digital twin simulations

• NVIDIA’s activities in the space are primarily driven by the Digital Twin sector, which accounted for ~79% of activities within this vertical. The company notably had three partnerships with Siemens, which integrated its industrial design and development technology into NVIDIA Omniverse and created digital twin turbines and defect detection models.

Note: Companies with more than one partnership are indicated numerically

Source: Compiled by SPEEDA Edge

• Activity in the space was limited compared with other verticals, given NVIDIA’s late entry into the sector with the launch of NVIDIA Omniverse in April 2021. Posited as a metaverse, the platform was launched when Big Tech and investor interest in metaverse platforms was gathering steam, although its AI-enabled full-fidelity live digital twins solution differed from what users typically associated with metaverse platforms. 
• Omniverse extends the company’s AI activities in the Auto Tech space, facilitating the creation of digital twin environments specifically designed for the development and testing of autonomous driving systems. This allowed NVIDIA to provide an ecosystem for auto manufacturers, which not only powered autonomous vehicles, but also tested them on a virtual platform. For example, Mercedes-Benz tested driving capabilities using NVIDIA’s Drive Sim on Omniverse while creating a digital twin of its factory to help mitigate waste and energy consumption. 
• Additionally, BMW partnered with NVIDIA to establish a smart factory, leveraging the AI capabilities of the company’s Isaac platform to create logistics robots for its automotive factories. Other use cases include the creation of digital twins of 5G networks, power plants, and climate research projects.

5. Work: Infusing AI for modeling and analytics  

• NVIDIA’s activities in the space were driven by the Data Infrastructure & Analytics and GenAI Applications segments, which collectively accounted for 72% of activities within this vertical.

Source: Compiled by SPEEDA Edge

• Although the company lacks specific products geared toward the Data Infrastructure & Analytics industry, it gained exposure via the use of its GPUs for analytical and modeling tool applications. 
• Given the buzz around GenAI, the company launched a series of products, powered by its DGX Cloud (NeMo, BioNeMo, and Picasso) in March 2023. The company also launched NeMo Guardrails in April 2023, an open-source toolkit designed to enhance the safety conversational systems, including ChatGPT. This fits the company’s investment into foundation model creator Inflection AI, which prioritizes AI safeguards. The company also announced FlexiCubes in August 2023, which uses GenAI to create realistic 3D meshes of objects, enabling the use of robust physics simulations.
• Seeking to address the time-consuming nature of developing and deploying foundation models, the company announced NVIDIA AI Workbench in August 2023, a toolkit allowing developers to create, test, and customize pre-trained GenAI models on a PC or workstation before scaling them to the public cloud or data center. 
• A majority of its activities in the GenAI Applications space relate to content creation and marketing-related applications, including collaborations with Adobe to use GenAI to streamline content workflows. In addition, NVIDIA invested in text-to-video generation company Runway AI, which leverages NVIDIA’s GPUs to run its tech, alongside Luma AI and Cohere.

Key risks: Supply chain worries and intensifying competition

• Supply chain blockages impacting its reliability: Despite CEO Jenson Huang’s claims of an ample supply of AI capacity through 2023, the company is facing bottlenecks in supplying the massive demand it is seeing for AI chips from enterprises, particularly its top-of-the-line H100 chip. These concerns are compounded by the roughly one-year lead time NVIDIA needs to fulfill a production order and the company’s overreliance on TSMC to manufacture its chips. TSMC also pushed back the launch of its AI chip facility in Arizona to 2025 due to labor shortages.
• Competition in the sector: AMD has wanted a larger slice of the AI pie with CEO Lisa Su stating that the company saw a 7x increase in demand for its chips during Q2 2023 backed by multiple cloud providers engaging with the company. This coincides with AMD’s MI300X AI chip announcement (set to ship to customers in Q4 2023). Other competitors developing AI chips include Amazon, Intel, Google, and Microsoft. However, strong competition from startups is less likely unless backed by significant funding, given the capital-intensive nature of scaling ML hardware/training LLMs (NVIDIA spent USD 7.3 billion on R&D or ~27% of its revenue in FY2023).

Competitor AI chips positioning relative to NVIDIA

• US-China trade relations restricting AI chip exports: The company’s macroeconomic woes include US AI chip export rules imposed on China in October 2022, which resulted in NVIDIA supplying downgraded versions of its H100 and A100 catered to the Chinese market. Additionally, US lawmakers have been pushing for tighter export restrictions as of late, with NVIDIA stating that restricting sales of its AI chips to China "would result in a permanent loss of opportunities.” If this materializes, it could have a widespread impact on the company, having derived ~21% of its FY2023 revenue from China. 

The road ahead: Toward an AI-powered future

Appendix 1: Industry potential of NVIDIA’s AI focus industries

Appendix 2: List of NVIDIA AI M&As

Appendix 3: List of NVIDIA AI Investments

Appendix 4: List of AI partnerships by NVIDIA