Inside Nvidia’s acquisition paperwork, Mellanox Technologies appears as a company purchased for approximately $7 billion, its shareholders paid $125 a share and its business absorbed into an American corporation. Six years later, the networking organisation built from that purchase employs thousands in Israel, supplies technology on which Nvidia’s expanding data-centre business depends, and has given the company its chief technology officer. Jensen Huang bought Mellanox while preparing for a future in which computing would require much larger networks of processors; with Huang now out of office and that future generating enormous revenues, the Israeli organisation he acquired has become one of the institutions from which his successor could come. The SEC transaction record is a useful place to begin because the succession has an industrial history that predates the scandal.
Huang resigned as president, chief executive and director on November 12, after reporting about his relationship with a Megaspeed executive whom American authorities suspect of working for Chinese intelligence. Nvidia appointed finance chief Colette Kress interim CEO and Harvey C. Jones its first board chair, while announcing a search across internal and external candidates. The company’s statement promised continuity in customer commitments and the product roadmap. Behind that promise sit engineering organisations that a board can neither replace nor move as quickly as it can appoint a chief executive. Nvidia’s announcement transferred executive authority immediately; the company’s dependence on those organisations continues through the transition.
Michael Kagan, the Mellanox co-founder who became Nvidia’s chief technology officer, features in succession reporting alongside Apple hardware chief Johny Srouji, while Ynet has also discussed Mellanox co-founder Eyal Waldman. No board shortlist has been disclosed. Their presence in the discussion nevertheless connects the collapse of Huang’s leadership to a longer consolidation of Israeli engineering inside American technology companies, one already visible at Apple, where Srouji gained authority over both hardware engineering and hardware technologies earlier this year. An Israeli executive succeeding Huang would bring the chief executive’s authority closer to an engineering base whose importance had been growing throughout his tenure.
Washington has treated access to advanced computing as a national-security question when the customer is Chinese, and it has celebrated the corporate expansion that makes Israeli engineering increasingly difficult to replace, and yet the implications of that second relationship receive much less scrutiny. Nvidia’s acquisition strategy has made Israel important to the development of AI infrastructure and to the revenues it generates. The evidence supports an argument about structural dependence, technical authority and the political protection those relationships can create. Establishing Israeli state direction of Nvidia, or a personal commitment by Srouji to act for that state, would require evidence beyond the corporate record examined here.
Reuters reported that American authorities had concluded Alice Huang was a Chinese intelligence asset, placing the founder’s private conduct inside an investigation with consequences for Nvidia’s security arrangements and commercial relationships. Its November 12 account also described the immediate leadership transition and Donald Trump’s public proposal that Tim Cook take over. The espionage finding remains an attributed assessment by American authorities, rather than a criminal conviction, and Huang’s resignation does not by itself establish which information changed hands or who authorised any unlawful shipment. Reuters’ reporting leaves the next chief executive inheriting an investigation as well as a business.
Senators Elizabeth Warren and Chris Coons addressed their November 11 preservation demand to Commerce Secretary Howard Lutnick, Secretary of State Marco Rubio and White House chief of staff Stephen Miller. They asked the three offices to retain records covering advanced-chip exports, the suspected intelligence operation and the government’s handling of Megaspeed. Citing published reporting, their letter put Megaspeed’s purchases since June 2024 at more than $4.8 billion and raised questions about changes to a Commerce investigation following communications involving Nvidia representatives. Their demand extends the inquiry beyond Huang’s conduct to the conduct of officials responsible for enforcing the restrictions. The senators’ letter asks whether recommendations were overruled and requires preservation of the records needed to answer.
The recipients matter. Commerce administers the export rules, State assesses foreign-policy consequences, and the White House can arbitrate disputes between departments. A record of who received an intelligence warning, who sought restrictions and who opposed them could explain how commercial access continued while national-security concerns circulated. The senators’ account remains an oversight allegation drawing on reporting, but it identifies a testable sequence of decisions. Investigators can compare memoranda with meeting records and subsequent licensing or enforcement action, rather than infer official behaviour from the founder’s private relationship.
Megaspeed has disputed illegal transfers. That denial belongs beside the allegations because the company’s purchases and Nvidia’s relationship with its executives establish exposure, while the legality of individual transactions requires their records. Huang’s acknowledged infidelity can explain the immediate governance crisis; it cannot answer whether American officials allowed shipments they had reason to stop. Folding those questions into a single story of personal recklessness would spare the institutions that were supposed to manage the risk.
Between a chip designer and the machine that eventually runs a model, hardware can move through server manufacturers, distributors, cloud operators, financing vehicles and data-centre tenants. An invoice names a buyer; an export declaration identifies a destination; neither, by itself, settles who will use the resulting computing capacity. A server can remain in the country to which it was lawfully exported while customers access it from elsewhere, which makes the contractual customer and the effective user separate questions for investigators.
The Bureau of Industry and Security built its advanced-computing restrictions around precisely these problems. Its October 2022 rule covered specified chips and supercomputer end uses, and subsequent revisions expanded the perimeter. These controls operate through technical thresholds, destinations, end users and licensing requirements, rather than a prohibition attached to the Nvidia logo. Establishing a violation requires the applicable rule, the product configuration and the transaction history. The original federal rule gives investigators a legal framework; it cannot substitute for evidence about a particular shipment.
Kress therefore inherits two demands that can pull against each other. Customers want deliveries and stable technical support, while investigators need records that may expose weaknesses in how those same customer relationships were monitored. Cutting off a distributor can interrupt revenue, but continuing a relationship under scrutiny can deepen the company’s exposure. The board’s promise of continuity has to pass through these decisions, made by employees whose work attracts little of the attention once reserved for Huang’s presentations.
Mellanox entered Nvidia through a transaction announced in March 2019 and completed in April 2020. Nvidia’s closing filing with the Securities and Exchange Commission recorded a purchase of approximately $7 billion, with Mellanox shareholders receiving $125 per ordinary share. The agreement brought an Israeli networking company under Nvidia’s ownership, along with an engineering organisation built around moving data quickly between computing systems. The closing filing identifies the legal direction of control: Nvidia acquired Mellanox.
Huang’s letter to employees at completion described Israel as one of Nvidia’s most important technology centres and put the workforce there at 2,500. Mellanox would become Nvidia’s networking business, with the acquired organisation incorporated into the parent’s plans rather than dispersed after its patents had been purchased. Nvidia was buying the ability to develop successive generations of products, which meant retaining the people who knew why earlier designs worked and where they failed. Huang’s employee letter placed that organisation inside the company’s future.
For a training cluster, adding processors also adds communication. Different processors perform different pieces of the work and exchange results, and delays in those exchanges leave expensive hardware waiting. The buyer paying for a larger cluster needs useful computing output from the whole installation; processor specifications alone cannot guarantee it. Networks therefore influence the return on the investment, even when the accelerator receives the public attention.
Nvidia’s own architecture separates several kinds of communication. NVLink connects GPUs within the scale-up domain, while InfiniBand and Ethernet provide scale-out connections across systems. The DGX networking guide describes these layers. The acquired networking business became valuable because it helped Nvidia supply a larger share of the working installation around its processors.
A purchaser can select an accelerator from a catalogue, but putting a large cluster into service requires software, network configuration, cooling and technical support to work together. Once an installation is built around a supplier’s integrated design, replacing one component can require changes elsewhere. Nvidia gains influence through these dependencies, and the engineering teams maintaining them gain importance inside Nvidia. The arrangement rewards continuity even when the board is replacing the person who negotiated it.
Israel-1 supplied a concrete example of this integration. Nvidia announced the supercomputer as a blueprint and testbed for its Spectrum-X Ethernet platform, specifying Dell servers, H100-based systems, BlueField-3 data-processing units and Spectrum-4 switches. The installation joined development infrastructure to a product that Nvidia intended to sell to other customers. Israel housed part of the environment in which the company tested how the components would work together. The company’s 2023 announcement described the hardware and its intended use.
Nvidia’s second-quarter fiscal 2027 filing puts its Israel workforce at approximately 6,200 across research and development, operations, sales and marketing. Compared with the acquisition-era figure, the expansion shows a commitment extending well beyond the retention of a purchased office. It also places regional conflict inside the company’s operating disclosures. Nvidia reported that its networking supply chain, including the Israel operation, had not been significantly affected by the Middle East conflict; the assurance acknowledges a geographic exposure that its customers must accept along with the technology. The quarterly filing provides the workforce count and the risk statement.
Calcalist’s account of Nvidia’s networking business supplies a financial measure of what the acquisition has become. Networking revenue reached $17.5 billion in the reported quarter, an 18 percent increase from the preceding quarter, while the company’s fiscal 2026 annual report showed $1.287 billion in Israeli corporate tax payments. The revenue figure covers a business with substantial Israeli development activity; describing every dollar as a sale produced inside Israel would confuse a corporate product category with a geographic account. The tax payment, however, directly connects Nvidia’s growth to the Israeli treasury. Calcalist’s financial report puts a measurable fiscal relationship beside the engineering one.
Israeli economists were examining that relationship months before Huang resigned. Calcalist reported in June that Leader chief economist Jonathan Katz attributed a large part of unexpectedly strong receipts to Mellanox-related activity, including goods sold abroad that never physically crossed Israel’s borders. Meitav chief economist Alex Zabezhinsky argued that headline growth figures had become less representative of the wider economy because so much activity was concentrated in one company or a small group. Their analysis describes the effect of intellectual property, management and development registered in Israel on the country’s economic accounts, even where production takes place elsewhere. Calcalist’s economic analysistraces the route from a multinational’s business to national growth and revenue statistics.
An acquired engineering organisation can consequently serve several interests at once. Nvidia needs its products and expertise; the Israeli government receives taxes; the local industry gains experienced employees whose next jobs or companies can remain within the same economy. None of these benefits requires the parent corporation to become Israeli-owned. The arrangement allows American shareholders to capture profits while Israel retains valuable work and fiscal receipts, creating constituencies on both sides for keeping the relationship intact.
A diplomatic rupture would therefore impose costs on American commercial interests as well as Israeli ones. Relocating a development programme requires retaining engineers or replacing them, transferring work between teams and preserving the knowledge accumulated during earlier product cycles. The company could undertake that expense, but the commercial incentive is to continue expanding an organisation that already delivers. American reliance deepens through ordinary decisions about staffing and product deadlines, without requiring an administration to announce a new strategic doctrine.
Calcalist’s November reporting adds a further field of work. Israeli engineers are developing BlueField-4, which Nvidia describes as providing a separate layer through which the activity of autonomous AI agents can be monitored and constrained. Senior vice president Gilad Shainer described the intention as keeping agents within their permissions. The proposed role extends the Israeli organisation’s work into infrastructure for governing how increasingly autonomous systems use computing resources. This is a company claim about a developing capability, rather than proof that the problem of controlling agents has been solved. Calcalist’s BlueField report places that programme inside the same development base inherited from Mellanox.
For Nvidia’s customers, this would put more functions in the surrounding infrastructure. Organisations deploying agents need to restrict access to data and prevent unauthorised actions, and an external enforcement layer could become another component around which installations are designed. The inference for the succession is practical: the acquired organisation is still gaining responsibilities. A future chief executive would be inheriting active development programmes as well as a successful historical acquisition.
Kagan brings the acquired business into the succession discussion through his position as Nvidia’s chief technology officer. His relevance rests on his work inside the organisation and familiarity with its technical direction. An executive who understands the interaction between processors and networks can offer continuity at a moment when the company must convince customers that its product plans remain credible. His elevation would also give the former Mellanox organisation a direct route from the acquired company’s technical leadership to the leadership of its purchaser.
Calcalist’s CTech discussion includes Kagan and Srouji alongside Ian Buck, Kress, Arm chief Rene Haas and AMD chief Lisa Su. Buck’s role in CUDA gives the board another kind of institutional continuity, rooted in the software ecosystem that helped make Nvidia hardware useful to developers. Kress brings financial and customer knowledge accumulated since joining in 2013. CTech’s succession report explicitly presents potential candidates.
Waldman’s name adds the history of building Mellanox as an independent business. Ynet discussed him with Kagan and Srouji among Israeli executives who could enter the conversation, but his inclusion in a newspaper’s assessment does not establish a recruitment process. The significance lies in the pool of experience from which commentators are drawing: the founder of the networking company, its former technical leader, and the executive responsible for Apple’s hardware. Their careers cross the same corporate institutions through which engineering expertise accumulated. Ynet’s report places them within a broader field of possible successors.
Srouji would arrive with a different mandate from Kagan. Apple develops custom silicon for its own products, while Nvidia sells into an ecosystem of customers that can also become competitors. Running the latter requires managing buyers who want its current hardware while financing alternatives to reduce their future dependence. Technical standing can make an executive attractive, but the commercial conflict remains part of the job whichever passport the executive holds.
Apple expanded Srouji’s authority in April 2026 by combining hardware engineering with his existing hardware technologies organisation under the new title of chief hardware officer. The company’s announcement says he joined in 2008 to lead development of the A4 and had previously worked in processor development at Intel and IBM. It also records his bachelor’s and master’s degrees from the Technion. Apple’s announcement defines his expanded responsibilities.
John Ternus’s appointment as chief executive, effective September 1, and Tim Cook’s move to executive chairman placed that hardware reorganisation inside a larger succession. Ternus retained the chief executive’s authority over the corporation, while Srouji gained a wider hardware organisation. The published structure gives Srouji substantial operational responsibility without making Ternus a ceremonial figure. Apple’s board chose to concentrate hardware development under one executive as it transferred the chief executive role to another. Apple’s leadership announcement sets out that division.
Srouji’s relationship to the Israeli development operation can be traced to its establishment. In an interview published by Ynet, he described a centre that opened in March 2012 after he and Bob Mansfield proposed the project to Apple’s leadership. The report identified Aharon Aharon, formerly Srouji’s lecturer at the Technion, as the centre’s first chief executive, and put employment at 1,800 at the time of the interview. Srouji spoke approvingly of Israeli engineering and of building a world-class team. This is an active institutional commitment, evidenced by an organisation he helped bring into existence and continued to champion. The Ynet interview also describes his upbringing in Haifa’s Christian Arab community, a background that makes easy assumptions about political identity particularly careless.
Apple Israel manager Rony Friedman offered a more current account at Calcalist’s Tech TLV conference in February. He described local teams contributing across Apple’s core products, including work on its C1 cellular modem and N1 connectivity technology in collaboration with teams abroad. Apple’s development centres in Jerusalem, Herzliya and Haifa were participating in products sold throughout the world. Friedman also discussed the acquisition of Q.ai and its expected integration into the Israeli R&D operation. Calcalist’s interviewgives the relationship a product history beyond the reputation of a single executive.
Moving from processors into modems and connectivity changes the scale of that contribution. Apple’s custom silicon strategy gives it more authority over the performance and power consumption of its devices, and the teams helping develop those components participate in decisions that shape the finished product. Srouji’s wider hardware remit brings more of that work into one management structure. The consequence is visible in Apple’s organisation chart and in the projects its Israeli managers describe, without claiming that every component originates in Israel or that an engineer’s employer is taking instructions from a government.
Nvidia and Apple have arrived at related dependencies through different purchases and organisational choices. Mellanox supplied Nvidia with an established networking business; Apple assembled teams and acquired companies while extending its internal hardware capabilities. Both approaches retained work inside Israel and connected it to products whose customers are dispersed internationally. Srouji’s appearance in Nvidia succession reporting joins those histories through a possible appointment, while their existing engineering relationships continue regardless of whether he ever takes the call.
Put beside Mellanox, Apple’s arrangement shows how technical organisations persist through leadership changes. A custom processor takes coordinated work across design, software and manufacturing partners, and the experience gained in one generation feeds the next. Hiring an executive can move a person quickly; reproducing the organisation beneath that executive takes longer. The attraction of Srouji as a possible Nvidia successor comes partly from his experience managing that accumulated work, rather than a belief that a chief executive can arrive carrying an entire national industry.
A government buying a computing cluster acquires equipment inside this commercial structure. Even where it owns the building and controls access to the servers, it depends on the supplier’s product support and future development. Its expansion plans must accommodate the hardware available for sale and the legal conditions attached to it. For states outside the principal design and manufacturing centres, a promise of sovereign AI can therefore begin with a purchase order that leaves important decisions abroad.
Washington’s export regime adds a separate authority over the transaction. A willing supplier and a funded customer may still need a licence, and permission can depend on the destination, end use and applicable restrictions. Nvidia’s fiscal 2027 second-quarter filing describes licences for limited H200 shipments to specified China-based customers alongside Chinese restrictions that prevented it from selling all licensed products. The company also reported a $400 million charge associated with excess H200 inventory and purchase obligations in the first half. Conflicting state decisions reached the business through stock it could not sell as planned, with a measurable cost to the supplier. Nvidia’s filing supplies that example.
A government in the Gulf or South Asia that purchases an Nvidia installation enters these relationships from the customer’s end. It can finance the building and electricity supply, recruit operators and control physical access, while remaining dependent on an overseas supplier for the design and development of the underlying system. Buying more capacity increases what that government can do, but it also increases the investment committed to the supplier’s technology. Claims of sovereign AI need to be measured against those continuing dependencies, including the ability to maintain an installation when commercial or diplomatic conditions change.
Nvidia’s next earnings report, scheduled for November 17, will give Kress an early opportunity to explain how the transition affects customers and the company’s plans. The independent inquiry must address a different set of records, including access to sensitive information and the handling of relationships now under scrutiny. How much of what American officials knew reached Nvidia’s directors before the public disclosures? Answering that question would establish whether the company’s oversight failed through ignorance, deference to its founder or decisions taken despite warnings.
Kagan’s appointment, should the board choose him, would carry the Mellanox story from acquisition into corporate succession. Srouji’s would bring an executive who helped build Apple’s Israeli R&D operation and now oversees its combined hardware organisations into Nvidia’s leadership. A different appointment would leave the existing Israeli organisation in place, together with the networking products and new development programmes that have made it valuable. The chief executive choice can strengthen that position, but the accumulated dependence does not wait for the appointment.
Washington’s inquiry into China should follow the evidence through Megaspeed and the departments responsible for export enforcement. Alongside it sits a relationship that American policy has encouraged openly: a strategic computing supplier expanding engineering work in Israel, an Israeli treasury receiving substantial corporate taxes, and Apple assigning broader hardware authority to an executive who helped establish its development operation there. The corporate records support scrutiny of that concentration and its consequences for American decision-making. Removing Huang answers none of the questions about how those dependencies should be governed. His successor inherits them.