The NVIDIA Mellanox acquisition in brief
In March 2019, NVIDIA agreed to buy Mellanox Technologies, an Israeli maker of high-speed data center networking, for $125 a share in cash, an enterprise value of about $6.9 billion. The NVIDIA Mellanox acquisition was the company’s largest deal at the time, and it closed on April 27, 2020.
The deal paid off on a scale few acquirers ever see. In fiscal 2026, NVIDIA’s networking revenue was $31.4 billion, about 4.4 times what it paid. But the approval China gave in 2020 came with conditions, and in 2025 Beijing said NVIDIA had broken them.
This case covers both halves: why buying a bottleneck worked, and why the promises made to close the deal are still on the books six years later.
What NVIDIA bought: a bottleneck, not market share
NVIDIA made graphics processors. As AI models grew, the work moved from one chip to thousands of them, and the network linking the chips became the limit on speed. Mellanox’s InfiniBand and Ethernet products were that network. In 2018, Mellanox had about $1.1 billion of revenue.
Jensen Huang framed the logic in the announcement: AI demand would require “holistic architectures that connect vast numbers of fast computing nodes.” The two companies already sold to every major cloud provider and computer maker, and together their technology ran in over 250 of the world’s 500 fastest supercomputers.
At roughly 6.3 times Mellanox’s 2018 revenue, the price was full for a hardware supplier. NVIDIA paid it in cash and said the deal would be immediately accretive to gross margin, earnings per share and free cash flow. Acquisition strategy that starts from where the constraint will be, not where the revenue is today, explains the premium.
What the purchase accounting shows
By closing, the NVIDIA Mellanox acquisition cost $7.134 billion. NVIDIA booked $3.431 billion as goodwill and $2.970 billion as identified intangibles, including $630 million of in-process research and development. Only about $0.7 billion was everything else.
Figure 1. About 90% of the price was goodwill and intangibles: people, technology and the growth NVIDIA expected. Source: NVIDIA 10-Q, July 2020.
NVIDIA funded the deal partly with $5 billion of new notes and still expected it to be accretive including the extra interest. It also took a $161 million inventory step-up charge and $26 million of acquisition costs.
When 90% of a price sits in goodwill and intangibles, the value is in people and know-how. That is why HR due diligence on key engineers, and technology due diligence on what the products can do next, matter more than the balance sheet. Financial due diligence should then test whether the target’s growth can carry that much goodwill.
Thirteen months and a Chinese approval with strings
The merger agreement set an outside date of December 10, 2019, which could be extended twice by three months while regulators decided. NVIDIA agreed to pay Mellanox a $350 million termination fee if approvals failed. The NVIDIA Mellanox acquisition needed sign-off in the US, China and Israel.
China’s State Administration for Market Regulation (SAMR) approved it on April 16, 2020, with conditions. According to an analysis of the published decision, NVIDIA could not bundle GPUs with networking, had to supply on fair and non-discriminatory terms, and had to keep its products interoperable with rivals’ products.
It also had to keep confidential information from third-party chipmakers walled off between its GPU and networking teams. The conditions run for six years. The deal closed 11 days later, about six weeks before the last extension would have expired.
Figure 2. The conditions that cleared the deal outlived the integration. Source: NVIDIA 8-K; SAMR decision via Legal 500; JURIST. The April 2026 end date is six years from approval.
How the NVIDIA Mellanox acquisition was integrated
NVIDIA integrated for growth, not cost. In its first quarter inside NVIDIA, Mellanox contributed 14% of revenue, and data center revenue hit a record $1.75 billion, up 167% from a year earlier.
Founder Eyal Waldman left six months after closing, saying “you don’t want to be number two.” NVIDIA had pledged at signing to keep investing in its Israeli talent base.
The China conditions shaped the integration too. Keeping GPU and networking teams separate on third-party information, and selling each product on its own, are requirements for go-to-market integration, not just for the legal team. Post-merger integration plans have to build them in from Day 1.
Retaining people while losing the founder is the usual pattern in technology deals. HR integration and a clear 100-day plan decide whether the engineers who follow the founder out are the few or the many.
NVIDIA bought the part of the system that would become the limit, before everyone else saw it.
The payoff: networking became a $31 billion business
In fiscal 2026, ended January 2026, NVIDIA’s networking revenue was $31.4 billion, up 142% from the year before. In the fourth quarter alone, it was $11.0 billion, driven by NVLink, Ethernet and InfiniBand.
Figure 3. One year of networking revenue now covers the purchase price more than four times over. Source: The Register; NVIDIA filings.
Not all of that is Mellanox’s original products, and NVIDIA no longer reports Mellanox separately. But the NVIDIA Mellanox acquisition gave it the base to build on.
Dominance drew a response. In July 2023, AMD, Broadcom, Cisco, Intel, Meta, Microsoft and others formed the Ultra Ethernet Consortium to build an open Ethernet alternative for AI networks. NVIDIA now sells Ethernet as well as InfiniBand.
The bill that came later: China
In December 2024, SAMR opened a review of whether NVIDIA had kept its 2020 commitments. On September 15, 2025, it said a preliminary investigation found NVIDIA had violated China’s anti-monopoly law and that it would investigate further.
Under Chinese law, a fine could reach 1% to 5% of the previous year’s China sales, Fortune reported. The finding came as US export controls limited what NVIDIA could sell in China, and on the eve of US-China trade talks.
As of late May 2026, no closure notice had been published, even though the six-year term would have ended in April 2026. For anyone studying the NVIDIA Mellanox acquisition, the lesson is that a regulatory commitment is a liability with no fixed maturity, as Carrier’s Viessmann covenants also show.
What the NVIDIA Mellanox acquisition teaches add-on buyers
The core lesson: buy the constraint your customers will hit next, and treat every promise made to close the deal as a cost you will carry after it.
Most sponsors will never buy a chip company. The same logic applies to a platform buying the scheduling software, the test lab or the specialist installer its customers cannot get enough of. Run this pre-mortem before signing.
| If the deal fails in five years, why? | What NVIDIA faced | Question for your add-on |
|---|---|---|
| The bottleneck moved | It did not: networking grew into a $31.4B business | Will customers still hit this limit in five years? |
| The people left | The founder left within six months | Who must stay, and what keeps them? |
| Bundling was blocked | China barred bundling for six years | Does the plan depend on selling the two together? |
| Rivals united against it | The Ultra Ethernet Consortium formed in 2023 | Could customers fund an open alternative? |
| A promise came due later | A 2025 finding of violation, five years after close | Which commitments outlive the deal team? |
Each row belongs in buy-side diligence, and the commercial ones in commercial due diligence. A buyer that can see the next constraint, and price the strings attached, is building an end-to-end M&A capability.
The pattern of buying a capability early and letting it run has worked before: Microsoft kept LinkedIn largely independent and let it grow.
The promises that close a deal are liabilities with no fixed maturity date.
For emerging managers: The Charter Program
NVIDIA had the balance sheet to pay up for a bottleneck and the legal team to manage the conditions for years. A Fund I, II or III manager making the same kind of bet on an add-on rarely has a corporate development or integration team to test it.
The Charter Program is built for that gap: build the capabilities of a larger private equity firm without building the cost structure of one.
Origination carries no monthly retainer; Xpertegic earns a success fee at close and reinvests 10-20% of that fee into the platform. Diligence and integration are scoped and billed separately. Three managers are admitted in year one, starting with a pre-application.
Sources
- The Register, Mellanox fourth quarter and full-year 2018 results, January 31, 2019
- NVIDIA, NVIDIA to acquire Mellanox for $6.9 billion, March 11, 2019
- NVIDIA, Form 8-K on the Mellanox merger agreement, March 2019
- Legal 500 (Akin Gump), Nvidia/Mellanox: China’s close scrutiny of semiconductor deals continues, 2020
- NVIDIA, NVIDIA completes acquisition of Mellanox, April 27, 2020
- NVIDIA, Form 10-Q for the quarter ended July 26, 2020
- NVIDIA, Financial results for the second quarter of fiscal 2021, August 19, 2020
- Globes, Waldman quits Mellanox: I don’t want to be number two, November 4, 2020
- Linux Foundation, Ultra Ethernet Consortium formed, July 19, 2023
- JURIST, China finds Nvidia violated antimonopoly law in preliminary probe, September 16, 2025
- Fortune, China says Nvidia violated antitrust law, September 15, 2025
- NVIDIA, CFO commentary on fourth quarter and fiscal 2026 results, February 25, 2026
- Tech Times, China-Russia antitrust alliance adds enforcement pressure on Nvidia, Qualcomm, May 27, 2026
