Nvidia’s Investment Boom

A handful of tech giants are rapidly consolidating control over the future of AI. While public attention often focuses on OpenAI, Amazon, or Microsoft, one of the most powerful actors in the AI boom flies under the radar of regulatory scrutiny: Nvidia.

Nvidia sells the specialised chips, Graphics Processing Units (GPUs), that power AI data centres and has therefore become central to the industry. By maintaining more than 80-90 per cent share of the data centre chip market and facing little competition 1 , Nvidia has become a gatekeeper of the AI boom.

The scale of Nvidia’s rise is unprecedented. Since the launch of ChatGPT in late 2022, the company’s market value has increased more than tenfold. As of May 2026, Nvidia is the world’s most valuable company, surpassing a USD 5 trillion(opens in new window) market valuation – roughly comparable to Germany’s annual GDP.(opens in new window) Nvidia is also extremely profitable. It currently enjoys(opens in new window) gross profit margins of over 70 per cent and, between the fiscal years of 2022 and 2026, just four years, Nvidia’s revenues increased(opens in new window) by more than 700 per cent.

While Nvidia has become one of the biggest profiteers of the global AI boom, this growth has created an unusual problem: it has more cash than it can easily spend.

^{Note: Nvidia’s fıscal reports end on 31 January of each year}.

Nvidia has addressed this “problem” by deploying its growing cash reserves into investments, partnerships, and acquisitions. These moves not only help reinforce its position in AI infrastructure but also support continued demand for the chips it produces. Nvidia’s investments span the entire AI industry: from startups that build models and apps, to companies in the AI supply chain, including some of Nvidia’s own actual or potential competitors.

According to S&P Global, Nvidia ranked as the fourth-largest corporate venture investor(opens in new window) in 2023, also investing in sectors such as healthcare and biotechnology. In May 2026, the Financial Times (opens in new window) estimated that Nvidia had committed roughly USD 90 billion over the previous 16 months in investments and partnerships.

For Europe, Nvidia’s rise raises a strategic question. Today, the EU is almost entirely dependent(opens in new window) on other countries for most advanced chips, including AI chips. Even if Europe succeeds in developing its own AI models and cloud services, can it achieve technological sovereignty if the infrastructure underpinning those systems remains controlled by a single US company? Nvidia’s growing concentration of power not only risks weakening competition in the AI industry but also shaping AI’s infrastructure and development by its commercial interests, with limited democratic accountability.

From gaming chips to shaping the AI industry

AI systems rely on a combination of hardware, including processors, memory, storage, and networking equipment. Among these components, GPUs have become the most critical. They are now used both to train AI models and to run them once they are deployed.

Founded(opens in new window) in 1993, Nvidia became a leading player in the graphics industry thanks to the success of its GPUs, chips originally designed for video gaming and design work.

That changed in 2012, when researchers(opens in new window) discovered that GPUs were also extremely effective at handling the kinds of repeated calculations used in training AI systems. Instead of processing one task at a time, GPUs can perform many operations in parallel, making them well-suited for modern AI tasks such as inference 2 and training 3 . Every major AI company – from model developers to cloud providers – depends on access to large numbers of these chips. The explosive growth of AI has therefore also fuelled an unprecedented surge in demand for Nvidia’s chips.

The European Commission’s investigations into Nvidia’s acquisitions of Mellanox(opens in new window) and Run:ai(opens in new window) show that Nvidia has been controlling more than 80 to 90 per cent of the discrete data centre GPU market for a long time. Estimates(opens in new window) suggest that as of 2026, Nvidia still controls more than 90 per cent of the market while competitors such as AMD and Intel are falling behind.

Nvidia’s chip empire is protected by a software moat

Nvidia’s dominance in AI chips is not based on hardware alone. It is reinforced by a software system that sits on top of its GPUs.

This system is called CUDA (opens in new window) (Compute Unified Device Architecture), which enables developers to write code that can run on Nvidia’s GPUs. Nvidia itself has emphasised(opens in new window) its importance, stating in 2020: “Almost every deep learning framework today uses CUDA/GPU computing to accelerate deep learning training and inference.” In other words, most modern AI systems are built to rely on Nvidia’s software. Because of this, CUDA has become what a senior analyst, Janakiram MSV, labelled(opens in new window) the “lingua franca” of GPU computing.

While CUDA is free to download, it has an important catch: code written on CUDA is not directly compatible with non-Nvidia GPUs. This creates dependency. Once developers build their systems around CUDA, switching to alternative hardware can become difficult and costly.

A report on generative AI(opens in new window) from the French Competition Authority highlights the dependence on CUDA as a risk of abuse. Tech writer Sheon Han identifies(opens in new window) CUDA software as the “forbidding moat” that surrounds Nvidia. In business terms, a “moat” refers to a company’s competitive advantage that helps protect it from competitors. For Nvidia, that moat is software-driven: even if rival GPUs improve, the cost and complexity of moving away from CUDA can keep developers locked into Nvidia’s ecosystem.

More than chips: Nvidia’s investment strategy for AI factories

While GPUs power Nvidia’s main business, the company has started moving beyond designing and selling GPUs and offering software systems, to actively shaping the broader AI infrastructure – meaning the building and supporting of data centres where AI models are trained and run. Something it calls “AI factories(opens in new window) ”.

A key part of this strategy shift was Nvidia’s acquisition(opens in new window) of the compute networking company Mellanox for USD 6.9 billion in 2019. Mellanox produces networking equipment that helps data to move efficiently between servers, storage systems, and other parts of data centres. This is critical for AI, where thousands of chips often need to work together simultaneously.

In March 2026, Nvidia’s senior vice president of networking, Kevin Deierling, explained(opens in new window) to TechCrunch the rationale behind the Mellanox acquisition, where having a networking business alongside its GPU business allowed the company to sell its chips with the tech that they work best with. He added: “I can’t think of other companies that have [the] full-stack capabilities that we have.” This strategy has also allowed Nvidia to solidify its position in the data centre market. According to its 2026 annual report(opens in new window) , Nvidia’s networking business generated more than USD 31 billion in revenue. Nvidia’s CEO reportedly said(opens in new window) in March 2026: “We’re now the largest networking company in the world.”

Less than six months after completing(opens in new window) the acquisition of Mellanox, Nvidia took another step with a far larger acquisition. In September 2020, it announced(opens in new window) a USD 40 billion deal to acquire Arm. Although not a direct competitor, Arm licenses Central Processing Unit (CPU) architectures and core designs that other companies use to build chips across sectors, from data centres to automotive.

The proposed acquisition quickly drew regulatory scrutiny on both sides of the Atlantic. In October 2021, the European Commission opened(opens in new window) an in-depth investigation into the deal, citing concerns about the restriction of competition. This was followed by a lawsuit(opens in new window) from the US Federal Trade Commission seeking to block the merger in December 2021. Faced with this opposition, Nvidia and Arm ultimately abandoned(opens in new window) the deal in February 2022. But this didn’t prevent Nvidia from investing in other companies.

^{Note: Negative value indicates profit from Nvidia’s investing activities. Positive values indicate net spending} 4 .

Nvidia’s investment machine is expanding its AI ecosystem

Nvidia has publicly explained that to “fuel the AI revolution” it has implemented a three-pronged strategy(opens in new window) . First, the company makes direct investments through its standard corporate investment programmes. Second, Nvidia invests in startups through its venture capital arm NVentures(opens in new window) . Third, it runs Nvidia Inception(opens in new window) , a programme that offers startups training, developer tools, investor access, and market reach.

Based on Crunchbase(opens in new window) data 5 , SOMO found that Nvidia participated in 283 funding rounds 6 between the start of 2021 and the end of 2025 through its three investment vehicles. While the majority of these funding rounds were handled through its corporate investment arm, the role of NVentures and Nvidia Inception has grown significantly in recent years, reflecting Nvidia’s more strategic approach to investing. By 2025, almost 85 per cent of its investments were in AI startups 7 .

Most of these investments also involve equity stakes. Around 250 out of the 283 of these funding rounds have been equity-only funding 8 , meaning that Nvidia and other investors acquired partial ownership in the invested companies.

Taken together, the data shows a rapid acceleration in Nvidia’s investment activity after 2022 following the collapse of Arm acquisition. As part of this strategy, Nvidia invested in some of the most promising AI startups, sometimes more than once. Between 2021 and 2025, it invested in 241 unique companies, and six received three or more investments, including leading AI startups such as Cohere (four times), Mistral AI (three times) and Perplexity (four times). The scale of Nvidia’s web of investment is vast.

For example, Forbes’s AI50(opens in new window) list of the world’s most promising privately held AI companies, published in April 2026, includes companies ranging from model makers, app builders, and chipmakers, to data centre builders. Nine out of the ten topmost highly funded AI startups on the list received investment from Nvidia, including rivals such as OpenAI 9 , Anthropic 10 , and MistralAI 11 .

Yet, Nvidia’s investment strategy extends across the wider AI industry supply chain. In September 2025, Nvidia invested USD 5 billion(opens in new window) in Intel, acquiring a roughly 4 per cent stake(opens in new window) in a company that has long been one of its closest rivals in the chip industry and another powerful player in the data centre infrastructure. For years, Intel has been trying to catch up with Nvidia’s success. In December 2025, Nvidia also invested USD 2 billion(opens in new window) in Synopsys, a leading provider of semiconductor software that plays a key role in chip development.

In 2026, Nvidia announced two major investments outside of the scope of this research period: a USD 2 billion investment(opens in new window) in Marvell, a company that produces compute, networking, and storage solutions, particularly for data centres; and up to USD 2.1 billion in IREN(opens in new window) , a data centre infrastructure company. These two deals show that Nvidia is expanding its reach deeper into the infrastructural layer of the AI industry.

Nvidia’s growth hinges on the AI boom fuelling demand for its chips

Access to computing power, mainly GPUs, is extremely expensive and out of reach for startups, which typically have limited resources. This creates a structural tension for Nvidia: if AI startups cannot access enough computing power, AI development would slow or start developing less compute-heavy models, which would also reduce demand for its chips. By funding and investing in AI companies, Nvidia helps sustain the very demand that drives its core business, while embedding itself more deeply in the broader AI ecosystem. As TechCrunch highlighted(opens in new window) , this created a strategic advantage for Nvidia: “the earlier it builds relationships with promising AI startups, the more likely those companies are to rely on its computing infrastructure as they scale.”

As a result, capital invested in AI companies largely contributes to demand for Nvidia’s products. For example, Nvidia announced(opens in new window) up to a USD 100 billion investment plan into OpenAI in 2025, which included agreements for OpenAI to buy Nvidia systems. Later, in March 2026, Nvidia reportedly(opens in new window) reduced its investment in OpenAI to USD 30 billion. Similarly, MistralAI, which received multiple rounds of investment from Nvidia, reportedly raised(opens in new window) USD 830 million in new ​debt to buy over 13,000 Nvidia chips.

These investment cycles raise questions(opens in new window) about whether AI growth reflects genuine market demand or whether it is partially sustained by a self-reinforcing industry that overstates underlying profitability and demand. An analyst from wealth management firm Wedbush Securities interprets(opens in new window) Nvidia’s investments as fitting “squarely into the circular investment theme”.

Circumventing merger regulation

Some of Nvidia’s most recent deals also illustrate how companies can expand their influence through structures that fall outside traditional merger control. Some of these partnerships, instead of full acquisitions, take the form of technology licensing and selective hiring of top talent. Venture Capital partner Kevin Kwok called(opens in new window) these “Hire and License Out (HALO)” deals. According to the Financial Times(opens in new window) , such deals have emerged “as a defensive manoeuvre against the antitrust policy of the Joe Biden era [… as they help] companies to sidestep the lengthy reviews that accompany traditional M&A.”

Take Nvidia’s investment in Groq, a startup that has developed a specialised AI chip known as the LPU(opens in new window) (Language Processing Unit). This chip offers an alternative approach(opens in new window) to AI inference and is designed to be power-efficient(opens in new window) , potentially reducing the energy use of AI workloads. In December 2025, Nvidia and Groq announced(opens in new window) a non-exclusive licensing agreement that gives Nvidia access to Groq’s inference technology, alongside the hiring(opens in new window) of key employees, including its founder and president.

This agreement was not a traditional acquisition of a company. As the CEO of Nvidia reportedly said(opens in new window) , “While we are adding talented employees to our ranks and licensing Groq’s IP, we are not acquiring Groq as a company.”

Yet, Nvidia paid quite a substantial amount for the deal. Nvidia recorded(opens in new window) the deal at around USD 17 billion in its annual records, of which around USD 14 billion is allocated as goodwill 12 . Given that Groq’s most recent valuation was USD 6.9 billion(opens in new window) , it is particularly striking that Nvidia paid nearly twice that amount for only a part of the company.

Some analysts have suggested that, in practice, such structures may function similarly to acquisitions. Yahoo Finance reported(opens in new window) that Hedgeye Risk Management analysts described this transaction as “essentially an acquisition of Groq without being labelled one (to avoid the regulators’ scrutiny).” A tech writer even described it as “Nvidia’s $20B Antitrust Loophole(opens in new window) ”.

Lawmakers have also raised concerns. US Senators Elizabeth Warren and Richard Blumenthal questioned(opens in new window) the deal, stating: “[B]y licensing its technology and hiring its most important employees, NVIDIA has effectively acquired Groq in all but name.”

From a competition perspective, these deals raise an important question: how meaningful is the distinction between these deals and official acquisitions? Avoiding regulatory scrutiny has ultimately allowed Nvidia to close the deal far more quickly than a traditional merger. By comparison, its acquisition(opens in new window) of Mellanox took over a year to complete(opens in new window) and was reviewed(opens in new window) by the European Commission.

In January 2026, Bloomberg reported(opens in new window) that the US Federal Trade Commission had begun scrutinising large tech firms for hiring employees rather than acquiring companies outright. The US Department of Justice antitrust head also reportedly warned(opens in new window) that deals in which Big Tech snaps up talent from AI startups are a red flag. “When I see conduct that appears aimed to circumvent that process, as a litigator, as an enforcer, that’s more of a red flag to ​me than if ​you had ⁠just participated and complied.” Despite these signals of regulatory unease, competition authorities and policymakers have so far remained largely silent.

Tech sovereignty starts at the infrastructure level, and policymakers must act accordingly

Nvidia has built a leading position in the AI industry through its dominance in hardware, software and infrastructure across the technology stack. Today, it is difficult to imagine AI without Nvidia.

As UCL professor and researcher Cecilia Rikap notes(opens in new window) , corporate power in AI goes beyond straightforward ownership. Big Tech firms can deepen their influence not only by acquiring companies but also through investment strategies and partnerships that shape the broader ecosystem and, ultimately, how the entire sector develops. Moreover, the dynamics of some of Nvidia’s investments risk creating a circular investment cycle that artificially shapes the future of the AI economy.

Against this backdrop, policymakers in Europe and elsewhere bear a special responsibility to protect the public from growing economic and technological dependencies on Big Tech’s shadow market power through investments.(opens in new window)

Existing regulatory and competition tools must be used to scrutinise deal structures that may attempt to avoid merger control or lock startups into the ecosystems of dominant tech companies. Without such oversight, there is a risk that innovation, talent, and critical technologies become increasingly concentrated in the hands of a few major tech firms, limiting startups’ ability to compete and challenge tech giants. The intervention in the failed Nvidia/Arm deal and Arm’s subsequent growth suggest that regulatory scrutiny can allow innovation to develop.

However, competition enforcement alone is unlikely to be sufficient. Policymakers must also use existing frameworks, including public investment, to support the emergence of credible alternatives. The EU’s forthcoming Technological Sovereignty Package(opens in new window) , including Chips Act 2.0(opens in new window) , could be an important step in this direction. The package aims to strengthen Europe’s position in semiconductors, AI, cloud computing, and open-source technologies.

Yet, initiatives such as Deutsche Telekom’s collaboration with Nvidia(opens in new window) in Germany and Mistral AI’s joint venture with Nvidia and other organisations(opens in new window) in France highlight a persistent risk of deeper dependency: the hardware that powers them. Even as Europe may develop its own AI models, it may remain dependent on Nvidia’s underlying infrastructure.

The underlying problem is that Nvidia’s AI investments steer AI development into one specific direction that primarily serves Nvidia’s interest in selling more chips. However, there are other ways to develop AI, many of which would require fewer chips. EU policy-makers must not simply accept the trajectory established by US tech giants like Nvidia.

Public funding should therefore be designed to reduce dependencies, not lock them in. If public money is used to shape the future of AI, it must be accompanied by democratic oversight of the direction of that investment and the public-interest goals it serves. Otherwise, decisions that shape Europe’s technological future risk remaining concentrated in the hands of a few powerful corporations.

A different trajectory is possible, one in which AI development prioritises social well-being, environmental sustainability, and the broader public interest. Without stronger competition policy interventions and industrial coordination at the infrastructure layer, Europe may develop AI capabilities while remaining structurally dependent on Nvidia’s systems and private interests.