if nvidia plans to double revenue next year, what other companies will ride that wave to support the revenue growth?

The strongest beneficiaries supply what the installation cannot do without

Nvidia’s growth increasingly involves an interconnected system. Its Rubin platform brings together a CPU, GPU, NVLink switch, network adapter, data-processing unit and Ethernet switch. The customer must also provide the electrical and cooling infrastructure needed to run that equipment. Following those requirements gives a more useful investment map than collecting companies with an AI label. [4][5]

The strongest links combine an essential function with evidence of participation in the deployment cycle.

Sources: company disclosures and technical documentation. Ordering expresses our business-exposure judgment; it does not rank expected stock returns. [6][7][8][9][10][11]

There are two different payment routes here. Nvidia’s manufacturing and memory requirements feed upstream suppliers. Data-center owners buy power and cooling infrastructure alongside the computing equipment. Those revenues belong to different stages of the same installation, so adding them together would overstate the amount of distinct spending.

A revenue doubling creates about $100 billion of additional product costs

Nvidia reported $96.2 billion of revenue in its latest quarter, of which $89.0 billion came from Data Center. Its company-wide gross margin was 75%. At that margin, approximately one-quarter of an additional sales dollar becomes cost of revenue; the rest remains gross profit before operating expenses. That is why a roughly $400 billion sales increase translates into a much smaller upstream cost pool. [2]

The annual denominator is still taking shape. Nvidia reported $177.8 billion in first-half fiscal-2027 revenue and guided the third quarter to $108 billion, plus or minus 2%. We use an assumed fourth quarter of $110–135 billion, spanning roughly flat revenue against the upper end of third-quarter guidance through a further sequential ramp. The upper case is the more aggressive delivery assumption. [2][3]

A doubling requires another $118–127 billion of sales beyond the outcome implied by 70% growth on the same assumed bases.

Sources: Nvidia disclosures; our calculations. Base revenue equals reported first-half revenue plus third-quarter guidance and assumed fourth-quarter revenue. Additional costs assume a constant 75% gross margin. [1][2][3]

The resulting $98–106 billion is an accounting cost envelope. Supplier orders, production and cash payments will occur on different schedules. A wider sensitivity, holding gross margins constant between 70% and 80% in both years, produces $79–127 billion of incremental costs. The conclusion survives that range: a very large supplier opportunity, with Nvidia retaining much of the economics. [2][3]

Product mix matters as much as the headline growth rate. Higher revenue per system allows Nvidia to double sales with less than double the physical output. For example, an assumed 15% increase in average revenue per equivalent system reduces the required unit increase to 74%, calculated as two divided by 1.15, less one. Suppliers benefit according to their content per system, production share and pricing—not through a uniform growth multiplier.

TSMC controls the most consequential manufacturing link

Before more systems can be installed, their advanced chips must be manufactured and packaged. Packaging connects processors and memory into a usable assembly; expanding chip fabrication alone does not remove a packaging bottleneck. TSMC’s chief executive, C.C. Wei, put the constraint plainly: “Our packaging capacity is so tight that now it limits my customers’ growth.” [6]

Our view is that TSMC offers the highest-confidence foundational participation in Nvidia’s expansion. The combination of advanced fabrication and packaging places it at a necessary production step. It also serves other chip designers, giving it broader exposure when customers change their accelerator choices. [6]

The economics are already substantial. TSMC reported $40.2 billion of second-quarter 2026 revenue, up 12% sequentially, and a company operating margin of 60.3%. It raised its full-year capital budget to $60–64 billion. Those figures show both the profitability of the existing business and the investment required to expand it. [6]

That investment is the price of participation. TSMC cited overseas-fab dilution even as better utilization and cost improvements lifted its gross margin. Our preference therefore comes with capital intensity, Taiwan concentration and execution risk. The crucial development is productive capacity arriving on schedule; a larger capital budget alone does not deliver another packaged accelerator. [6]

Relieving packaging pressure would help Nvidia ship more. It would also move attention to the next scarce input: the memory placed beside those processors.

SK hynix and Micron sell an input that consumes scarce factory capacity

High-bandwidth memory, or HBM, supplies the rapid flow of data required by AI processors. Its manufacturing demands make the opportunity more consequential than simply attaching more memory to each system: allocating production to HBM competes with other uses of memory-factory capacity. TrendForce expects HBM to account for 30% of leading suppliers’ DRAM wafer input by the end of 2027, while producing 13% of DRAM bit supply. [12]

HBM absorbs a much larger share of wafer input than its share of delivered memory bits.

Source: TrendForce estimates published June 2, 2026. Year-end shares cover the leading three suppliers; 2026 and 2027 are projections. [12]

The projected wafer-allocation share rises eight percentage points between 2026 and 2027, from 22% to 30%. That is a 36% relative increase in allocation share. It explains why this ramp matters across the memory industry, even without assuming a comparable increase in total wafer capacity. [12]

SK hynix is our preferred established HBM leader. Its multiyear Nvidia technology partnership provides a direct connection to the buildout, and it began mass HBM4 shipments in the second quarter of 2026. Counterpoint estimated its worldwide HBM revenue share at 58% in the first quarter. That combination of relationship, shipped product and market position is the basis for our preference. [7][13][14]

Micron is the most straightforward US-listed memory expression of the thesis. Its high-volume HBM4 announcement specifically identifies Nvidia Vera Rubin. Micron also expects its Singapore facility to contribute meaningfully to HBM packaging capacity in the first half of 2027. The product connection and expansion timing align with the period investors are trying to capture. [8][15]

The qualification is competition. Samsung remains a credible supplier and potential share-gain beneficiary; Counterpoint estimated its worldwide HBM revenue share at 21% in the first quarter of 2026. More competitive supply can help Nvidia deliver while reducing incumbent memory suppliers’ pricing power. We prefer SK hynix’s demonstrated leadership and Micron’s direct Rubin participation, without declaring either the better stock at an unspecified valuation. [14][16]

Vertiv, Eaton and Schneider monetize the work required to switch the racks on

Manufacturing the system gets it to the customer. Operating it requires another layer of spending. Lenovo’s GB300 NVL72 specification gives the scale: 135 kilowatts of thermal design power per rack, 155 kilowatts at peak, with approximately 90% of heat removed through liquid cooling. A thousand equivalent racks represent 135 megawatts of IT design load before facility overhead. [5]

That physical requirement supports our preference for Vertiv as the focused infrastructure beneficiary. Its 2.3-megawatt coolant-distribution unit received Nvidia DSX Ready qualification, connecting its product offering to Nvidia deployments. Vertiv reported $3.27 billion of second-quarter 2026 sales and a 22.6% adjusted operating margin. The qualification shows compatibility; subsequent shipments determine the revenue opportunity. [9][17]

Eaton supplies the electrical prerequisite. Its Electrical Americas segment generated approximately $4 billion of second-quarter sales at a 27.5% segment operating margin. Organic orders grew 41% on the company’s stated rolling-average measure. The segment serves a wider market than data centers, but the order growth is tangible evidence of spending reaching electrical suppliers. [10]

Schneider Electric is a close alternative, with broad power-management exposure. It reported €21.2 billion of first-half 2026 revenue and a 19.3% adjusted EBITA margin. We see no decisive evidence here to declare Eaton categorically superior; both provide a broader electrical route alongside Vertiv’s more focused power-and-cooling connection. [11]

The attraction is that electrical distribution and heat removal remain necessary when customers change accelerator vendors. The risk is timing: equipment can be ordered well before a site is ready. A grid connection that slips can delay the revenue opportunity across several suppliers at once.

Super Micro provides the concrete case. Explaining its fourth-quarter fiscal-2026 results, Charles Liang pointed to “short-term customer delays in power, cooling, and networking.” Management described this as a timing story. We take the constraint seriously: an installation can have demand and computing equipment yet still lack the infrastructure required to operate. [18]

Server sales show participation; cash conversion determines its value

Dell belongs on the secondary shortlist. It recognized $16.4 billion of AI-optimized-server revenue in its second quarter of fiscal 2027 and ended the period with a $95 billion AI-server backlog. It is already delivering at considerable scale. [19]

We nevertheless rank it below the core names in this investigation. A system seller records revenue that includes expensive components purchased from others. The investor’s question is how much profit and cash remain after those purchases, integration costs and working-capital requirements. Dell’s delivery figures establish direct participation, while our preference remains with the indispensable inputs and infrastructure.

Super Micro makes the cash question unavoidable. It reported a 10.9% company-wide non-GAAP gross margin and $6.8 billion of operating cash outflow in fiscal 2026. Rapid system growth can require a supplier to finance inventory and customer delivery before collecting the associated cash. Sustained improvement in profit and cash conversion would be the evidence needed to raise our ranking of the assemblers. [18]

HPE and Taiwanese system manufacturers also participate in the buildout. But large segment totals need careful interpretation: HPE’s $9 billion quarterly Cloud & AI revenue covers a business broader than Nvidia rack integration. We would apply the same retained-profit test rather than rank manufacturers by the largest AI-related revenue headline. [20][21]

Networking and grid exposure need their own underwriting

More computing creates more connectivity requirements, but Nvidia also supplies important parts of that connectivity. Rubin includes NVLink switching, network adapters and Ethernet switching. That makes the mapping from Nvidia growth to independent networking vendors more conditional than the mapping to essential memory. [4]

Arista has attractive economics: its second-quarter 2026 non-GAAP operating margin was 49.9%. We are nevertheless skeptical of treating its revenue as a proportional derivative of Nvidia’s. Customer Ethernet design wins and deployment choices determine how much of the opportunity it captures. [22][4]

We apply the same selectivity to Broadcom and Marvell rather than including them automatically in the core shortlist. Marvell’s NVLink Fusion participation provides a concrete ecosystem connection; its next-year earnings contribution still depends on commercial conversion. Lumentum and Coherent also deserve watchlist positions because their Nvidia optics partnerships are explicit. Quantified shipments and margins would justify promoting them. [23][24][25]

Further upstream, GE Vernova booked $2.7 billion of data-center orders in Electrification during the second quarter of 2026. That is a useful direct demand signal. Quanta Services reported $53.4 billion of total backlog, reflecting the broader construction opportunity. We regard both as potentially longer-duration beneficiaries because the path from equipment orders and construction backlog to energized sites can extend beyond the immediate Nvidia shipment cycle. [26][27]