Introduction

Ask almost anyone building AI infrastructure in the US right now which single company matters most, and the answer is nearly unanimous: NVIDIA. Not a cloud provider. Not an AI lab. A chipmaker that has quietly become the default systems vendor for the most capital-intensive technology buildout in American history.

In fiscal 2026, NVIDIA generated $215.9 billion in total revenue, up 65% from the year before. Its data center segment alone brought in $62.3 billion in the fourth quarter, growing 75% year-over-year. By the first quarter of fiscal 2027 (reported May 2026), quarterly data center revenue hit a record $75.2 billion, part of an overall $81.6 billion quarter. These aren’t just impressive numbers — they’re a signal of how deeply NVIDIA’s hardware, software, and business relationships now shape where and how AI data centers get built across the US.

This article breaks down exactly how NVIDIA is influencing the future of American AI infrastructure — from the chips themselves to the mega-projects reshaping entire states.

NVIDIA’s Market Position, in Numbers

Metric	Figure
FY2026 total revenue	$215.9 billion (+65% YoY)
FY2026 Q4 data center revenue	$62.3 billion (+75% YoY)
Q1 FY2027 data center revenue	$75.2 billion (record)
AI accelerator / data center GPU market share	Roughly 80–90%
Combined Blackwell + Rubin revenue visibility	~$500 billion (2025–2026)
Data center networking revenue growth (NVLink, InfiniBand)	+142% YoY

Sources: NVIDIA fiscal 2026/2027 earnings releases and Q1 FY2027 results

That last row matters more than it might look. NVIDIA isn’t just selling individual chips anymore — it’s selling entire rack-scale systems that combine GPUs, high-speed networking, and software, and the 142% growth in networking revenue shows how much of a US data center build today runs specifically on NVIDIA’s connective technology, not just its processors.

1. Blackwell Set the Standard Nearly Every New US Data Center Follows

NVIDIA’s current-generation Blackwell architecture has become the reference platform for AI-ready data centers being built across the US in 2026. Blackwell GPUs have reportedly been sold out with cloud capacity backordered, and CEO Jensen Huang has described demand as “off the charts.” According to earnings commentary, GB300 shipments have now crossed GB200 volumes and account for roughly two-thirds of total Blackwell revenue, showing rapid adoption of NVIDIA’s newest rack-scale configuration among US hyperscalers.

What makes Blackwell significant for US data center design isn’t just raw performance — it’s that the architecture is increasingly deployed as a full system, not a standalone chip. Hyperscalers now build data center floor plans, power delivery, and cooling infrastructure around NVIDIA’s rack specifications, which effectively gives NVIDIA influence over physical data center design decisions well beyond the processor itself.

2. Vera Rubin Is Already Reshaping 2026–2027 Construction Plans

NVIDIA’s next-generation Vera Rubin platform began full production ramp in mid-2026, and it’s already influencing how new US facilities are being planned. Rubin is explicitly designed around agentic AI workloads — AI systems that take multi-step actions rather than just answering single prompts — and NVIDIA says the platform can deliver up to 10 times lower token costs compared to Blackwell.

CFO Colette Kress has stated NVIDIA has visibility into roughly $500 billion in combined Blackwell and Rubin revenue from the start of calendar 2025 through the end of calendar 2026 — a scale of forward demand that gives US data center developers real confidence to keep building, even amid rising local opposition to new construction (a separate, growing challenge for the industry).

3. NVIDIA Is Directly Funding US Data Center Mega-Projects

Perhaps the clearest sign of NVIDIA’s influence isn’t a chip spec — it’s a balance sheet decision. In September 2025, NVIDIA and OpenAI signed a letter of intent for NVIDIA to deploy at least 10 gigawatts of NVIDIA systems for OpenAI’s infrastructure, with NVIDIA committing to invest up to $100 billion in OpenAI as each gigawatt is deployed, paid out in roughly $10 billion installments tied to construction milestones. Jensen Huang described 10 gigawatts as the equivalent of 4 to 5 million GPUs — roughly double what NVIDIA shipped the prior year.

This investment feeds directly into Project Stargate, the $500 billion US data center initiative backed by OpenAI, Oracle, SoftBank, and NVIDIA. As of mid-2026, Stargate had expanded to nearly 7 gigawatts of planned US capacity across sites including the flagship Abilene, Texas campus — expected to be fully operational by mid-2026 — plus new locations in New Mexico and Ohio. Oracle alone has committed roughly $40 billion to NVIDIA hardware for the Abilene site.

NVIDIA has struck similar large-scale commitments elsewhere in the US market: a partnership with Anthropic to adopt an initial 1 gigawatt of Grace Blackwell and Vera Rubin systems, support for xAI’s 2-gigawatt Colossus 2 data center, and involvement in an aggregate of roughly 5 million GPUs across various US AI factory projects. Very few individual companies can single-handedly influence how much AI infrastructure gets built in America — NVIDIA is now doing it on multiple fronts simultaneously.

4. The CUDA Ecosystem Locks In Long-Term Infrastructure Decisions

NVIDIA’s hardware dominance is reinforced by something harder to displace than a chip: its CUDA software platform. CUDA has been NVIDIA’s parallel computing framework for over a decade, and it has become deeply embedded in how AI developers write, train, and deploy models. Because switching away from CUDA requires significant retraining and rewriting of software stacks, US data center operators face real friction in adopting alternative chips from AMD or custom hyperscaler silicon — even when those alternatives are cheaper.

This software lock-in is a major reason analysts continue to estimate NVIDIA holds roughly 80–90% of the data center AI GPU market, despite growing competitive pressure. For US data center planning, it means infrastructure decisions made today are likely to keep favoring NVIDIA-compatible systems for years, simply because of the switching costs involved.

5. Networking Is Becoming as Strategic as Compute

One of the more overlooked shifts in NVIDIA’s 2026 growth is how much of it is coming from networking, not just GPUs. Data center networking revenue — driven by NVIDIA’s NVLink compute fabric alongside Ethernet and InfiniBand technologies — surged 142% year-over-year in fiscal 2026. As AI clusters scale into the tens of thousands of GPUs, the speed at which those chips can communicate with each other becomes just as important as the chips themselves.

This matters for US data center design because it means facilities are increasingly being engineered around NVIDIA’s full-stack approach — compute, networking, and software together — rather than treating GPUs as a component that can be mixed and matched freely with other vendors’ systems.

Risks and Limits to NVIDIA’s Influence

NVIDIA’s position isn’t without real constraints, and it’s worth being clear-eyed about them:

• China exposure. NVIDIA’s recent revenue outlooks have explicitly excluded data-center compute revenue from China in certain forecasts, reflecting US export-control restrictions that limit one of the world’s largest technology markets.
• Rising competition. AMD and custom silicon from hyperscalers themselves (including Google’s TPUs and Amazon’s Trainium chips) represent a long-term push to reduce dependence on NVIDIA, even if switching costs remain high in the near term.
• Capital spending durability. NVIDIA’s growth is directly tied to continued hyperscaler capex, projected at a combined $700 billion in 2026 across Alphabet, Amazon, Meta, and Microsoft. Any slowdown in that spending — whether from financial pressure, power constraints, or shifting AI economics — would flow directly through NVIDIA’s data center revenue.
• Physical bottlenecks beyond NVIDIA’s control. Even with unlimited chip supply, US data center growth is increasingly constrained by grid interconnection delays, water availability, and local construction opposition — factors NVIDIA’s technology can’t solve on its own.

What This Means for the Future of US AI Infrastructure

NVIDIA’s influence over US AI data centers in 2026 goes well beyond being a chip supplier. Through the combination of the Blackwell and Vera Rubin platforms, direct multi-billion-dollar investments in projects like Stargate, a networking business now core to how AI clusters are wired together, and a software ecosystem that keeps developers within its orbit, NVIDIA effectively sets the technical blueprint that much of the US data center industry is building around.

Looking ahead, expect NVIDIA’s next architecture cycles to continue driving how American data centers are physically designed — from power density per rack to cooling requirements — while its direct equity stakes in AI infrastructure projects give it a say not just in what gets built, but how fast and where. For an industry facing real headwinds from power shortages and local opposition, NVIDIA’s willingness to co-invest in infrastructure, rather than simply sell into it, may prove to be just as influential as its chip roadmap.