Without a high-speed, low-latency, and scalable network, GPUs are nothing more than isolated computing islands.

As computing has shifted from standalone machines to data centers, GPU performance is no longer the sole bottleneck. What truly determines the upper limit is how data flows between thousands of chips.

Mellanox’s leading position in InfiniBand and high-speed Ethernet filled exactly the most critical and hardest-to-develop piece of the puzzle for NVIDIA. In hindsight, without this acquisition, it would have been extremely difficult for NVIDIA to transform GPUs from mere high-performance computing cards into genuine data center infrastructure amid the subsequent AI boom.

This strategy of making bold, high-stakes moves was further extended the following year—when NVIDIA attempted to acquire ARM.

At the time, SoftBank was looking to exit, and NVIDIA moved swiftly to take over the company that holds the lifeline of the world’s CPU architecture. Had the deal succeeded, NVIDIA would have simultaneously seized the key gateways to the GPU, networking, and CPU ecosystems, laying bare its unmistakable ambition.

But this time, reality delivered a clear verdict: the transaction faced intense antitrust scrutiny across multiple global jurisdictions and was ultimately abandoned. The failure of the ARM deal also put NVIDIA squarely in the crosshairs of regulators worldwide, making it a prime target for antitrust investigations.

Since then, NVIDIA’s acquisition pace has undergone a distinct shift. Gone are the mega-mergers on the scale of Mellanox or ARM. Instead, they have been replaced by a series of smaller, more targeted deals, focusing largely on software, scheduling, efficiency tools, and development platforms.

Whether through direct acquisitions or “acquisition-like” approaches—such as asset purchases, team integrations, licensing agreements coupled with talent recruitment—NVIDIA has deliberately kept these transactions low-profile to minimize regulatory risks.

Jensen Huang has no need to “prove his presence” through mergers and acquisitions. Instead, under regulatory pressure, he has adjusted his strategy in a timely manner, patiently seeking out startups that perfectly complement NVIDIA’s existing ecosystem.

These companies are often small in size but pivotal in position—flying under the radar yet holding the keys to future technological evolution.

Rather than swallowing entire companies in one go, Jensen Huang cares more about precisely integrating their capabilities, technological roadmaps, and talent into NVIDIA’s system.

The Mellanox acquisition was a high-stakes gamble NVIDIA had to win; the ARM bid was an overly aggressive attempt. After these two experiences, NVIDIA has clearly learned to operate within boundaries and call it quits at the right moment.

V

With this context in mind, it is easy to understand why NVIDIA was quick to deny the “acquisition” rumors surrounding Groq and repeatedly emphasized that it was merely a non-exclusive technology licensing deal.

In the current regulatory climate, any acquisition rumors involving potential competitors will be quickly amplified. For NVIDIA, even a deal that “looks like an acquisition” could trigger unnecessary regulatory scrutiny. Therefore, in cases like Groq, downplaying the deal and drawing clear boundaries is almost an inevitable choice.

The broader context is that NVIDIA’s position in the market has undergone a fundamental transformation.

When it attempted to acquire ARM in 2020, NVIDIA’s market capitalization hovered around $300 billion. Today, it has rapidly grown into a tech behemoth with a market cap exceeding $4 trillion.

At this scale, it is almost impossible for NVIDIA to casually launch another “largest acquisition in history.”

Real-world choices have confirmed this trend: since 2025, NVIDIA has announced only a handful of acquisitions, all of which are startups, with most transaction amounts undisclosed. Among the few cases where prices were reported by the media, a typical example is the acquisition of Gretel, a data privacy startup, for approximately $320 million.

In terms of both scale and influence, NVIDIA has deliberately steered clear of the sensitive zone of “large-scale mergers and acquisitions.”

This does not mean, however, that regulatory pressures on NVIDIA are easing.

Its impressive market capitalization, near-monopolistic position in the AI chip sector, and the fact that its acquisition pace has not truly slowed down mean that the shadow of antitrust scrutiny looms overhead at all times. Even though NVIDIA has clearly shifted toward smaller, low-profile acquisitions and acquisition-like deals, regulators are no longer focusing solely on transaction amounts. Instead, they are increasingly concerned with whether these moves continue to consolidate NVIDIA’s control over the ecosystem.

As previously mentioned, U.S. regulators have launched an investigation into Microsoft’s “disguised acquisition” of Inflection AI through talent and asset poaching, and have begun collecting information and conducting evaluations of such AI partnerships as a whole. UK regulators have even directly initiated a full merger review under their existing framework. Although the deal was ultimately approved, it served as a stark warning to tech giants.

Regulators have made their stance clear: they are no longer just looking at whether a company has been “bought outright.” As long as entire teams of talent are poached, core capabilities are transferred, or potential competition is weakened, even deals that are not labeled as acquisitions may be scrutinized as such.

NVIDIA is not the company it once was—it has ballooned into a global giant in just a few years. Jensen Huang now faces a tangible dilemma: how, under the current circumstances, to gracefully integrate the assets and capabilities he must acquire?

NVIDIA has plunged headfirst into a cat-and-mouse game with regulators. The $20 billion “non-acquisition” deal may well reflect Jensen Huang’s shrewd judgment, but whether it will attract regulatory scrutiny remains an open question.

References

Silicon Research Lab: Crazy NVIDIA, Crazy Spending

IT Home: NVIDIA Shakes Up the AI Landscape: Participated in 50 Funding Rounds, Invested Over $1 Billion, and Covered 20 Companies in 2024

C114 Communications Network: NVIDIA’s $14 Billion Gamble: The Value of ARM

Qubit: Is Jensen Huang the Savior of the Switch?

Tae Kim: The NVIDIA Way

Without a high-speed, low-latency, and scalable network, GPUs are nothing more than isolated computing islands.

As computing has shifted from standalone machines to data centers, GPU performance is no longer the sole bottleneck. What truly determines the upper limit is how data flows between thousands of chips.

Mellanox’s leading position in InfiniBand and high-speed Ethernet filled exactly the most critical and hardest-to-develop piece of the puzzle for NVIDIA. In hindsight, without this acquisition, it would have been extremely difficult for NVIDIA to transform GPUs from mere high-performance computing cards into genuine data center infrastructure amid the subsequent AI boom.

This strategy of making bold, high-stakes moves was further extended the following year—when NVIDIA attempted to acquire ARM.

At the time, SoftBank was looking to exit, and NVIDIA moved swiftly to take over the company that holds the lifeline of the world’s CPU architecture. Had the deal succeeded, NVIDIA would have simultaneously seized the key gateways to the GPU, networking, and CPU ecosystems, laying bare its unmistakable ambition.

But this time, reality delivered a clear verdict: the transaction faced intense antitrust scrutiny across multiple global jurisdictions and was ultimately abandoned. The failure of the ARM deal also put NVIDIA squarely in the crosshairs of regulators worldwide, making it a prime target for antitrust investigations.

Since then, NVIDIA’s acquisition pace has undergone a distinct shift. Gone are the mega-mergers on the scale of Mellanox or ARM. Instead, they have been replaced by a series of smaller, more targeted deals, focusing largely on software, scheduling, efficiency tools, and development platforms.

Whether through direct acquisitions or “acquisition-like” approaches—such as asset purchases, team integrations, licensing agreements coupled with talent recruitment—NVIDIA has deliberately kept these transactions low-profile to minimize regulatory risks.

Jensen Huang has no need to “prove his presence” through mergers and acquisitions. Instead, under regulatory pressure, he has adjusted his strategy in a timely manner, patiently seeking out startups that perfectly complement NVIDIA’s existing ecosystem.

These companies are often small in size but pivotal in position—flying under the radar yet holding the keys to future technological evolution.

Rather than swallowing entire companies in one go, Jensen Huang cares more about precisely integrating their capabilities, technological roadmaps, and talent into NVIDIA’s system.

The Mellanox acquisition was a high-stakes gamble NVIDIA had to win; the ARM bid was an overly aggressive attempt. After these two experiences, NVIDIA has clearly learned to operate within boundaries and call it quits at the right moment.

V

With this context in mind, it is easy to understand why NVIDIA was quick to deny the “acquisition” rumors surrounding Groq and repeatedly emphasized that it was merely a non-exclusive technology licensing deal.

In the current regulatory climate, any acquisition rumors involving potential competitors will be quickly amplified. For NVIDIA, even a deal that “looks like an acquisition” could trigger unnecessary regulatory scrutiny. Therefore, in cases like Groq, downplaying the deal and drawing clear boundaries is almost an inevitable choice.

The broader context is that NVIDIA’s position in the market has undergone a fundamental transformation.

When it attempted to acquire ARM in 2020, NVIDIA’s market capitalization hovered around $300 billion. Today, it has rapidly grown into a tech behemoth with a market cap exceeding $4 trillion.

At this scale, it is almost impossible for NVIDIA to casually launch another “largest acquisition in history.”

Real-world choices have confirmed this trend: since 2025, NVIDIA has announced only a handful of acquisitions, all of which are startups, with most transaction amounts undisclosed. Among the few cases where prices were reported by the media, a typical example is the acquisition of Gretel, a data privacy startup, for approximately $320 million.

In terms of both scale and influence, NVIDIA has deliberately steered clear of the sensitive zone of “large-scale mergers and acquisitions.”

This does not mean, however, that regulatory pressures on NVIDIA are easing.

Its impressive market capitalization, near-monopolistic position in the AI chip sector, and the fact that its acquisition pace has not truly slowed down mean that the shadow of antitrust scrutiny looms overhead at all times. Even though NVIDIA has clearly shifted toward smaller, low-profile acquisitions and acquisition-like deals, regulators are no longer focusing solely on transaction amounts. Instead, they are increasingly concerned with whether these moves continue to consolidate NVIDIA’s control over the ecosystem.

As previously mentioned, U.S. regulators have launched an investigation into Microsoft’s “disguised acquisition” of Inflection AI through talent and asset poaching, and have begun collecting information and conducting evaluations of such AI partnerships as a whole. UK regulators have even directly initiated a full merger review under their existing framework. Although the deal was ultimately approved, it served as a stark warning to tech giants.

Regulators have made their stance clear: they are no longer just looking at whether a company has been “bought outright.” As long as entire teams of talent are poached, core capabilities are transferred, or potential competition is weakened, even deals that are not labeled as acquisitions may be scrutinized as such.

NVIDIA is not the company it once was—it has ballooned into a global giant in just a few years. Jensen Huang now faces a tangible dilemma: how, under the current circumstances, to gracefully integrate the assets and capabilities he must acquire?

NVIDIA has plunged headfirst into a cat-and-mouse game with regulators. The $20 billion “non-acquisition” deal may well reflect Jensen Huang’s shrewd judgment, but whether it will attract regulatory scrutiny remains an open question.

References

Silicon Research Lab: Crazy NVIDIA, Crazy Spending

IT Home: NVIDIA Shakes Up the AI Landscape: Participated in 50 Funding Rounds, Invested Over $1 Billion, and Covered 20 Companies in 2024

C114 Communications Network: NVIDIA’s $14 Billion Gamble: The Value of ARM

Qubit: Is Jensen Huang the Savior of the Switch?

Tae Kim: The NVIDIA Way