Research · News and analysis
The NVIDIA effect: how AI reshaped the IT estate
Four years, thirteenfold · The money · Why NVIDIA leads · Timeline · Alternatives · Your IT estate · Where most organizations are · Questions · Readiness check · Sources
In four years a graphics-chip company became the center of corporate IT spending. Most organizations will never buy an AI supercomputer, but almost all of them now feel its effects: in cloud bills, power and cooling, network design, hardware prices and the questions their boards ask. Here is what happened, what the alternatives are, and what it means for your IT estate.
The figures and sources could not be loaded. Please reload the page.
Four years, thirteenfold
Chart values as a table
| Period | Data center revenue | Kind |
|---|
Sources: · · ·
Single quarters: ·
NVIDIA's data center revenue grew from about $15 billion in fiscal 2023 to about $194 billion in fiscal 2026, and its latest quarter alone brought in $89 billion.
The money behind it
worldwide AI spending forecast for 2026, up 49.5%
AI infrastructure spending forecast for 2026
data center capital spending forecast for 2026
of 2026 AI server shipments built on custom chips rather than GPUs
Why NVIDIA leads
1. Software
Over 7.5 million developers use CUDA and NVIDIA's tools. Switching platforms means rewriting or re-testing that software.
2. Networking
Networking revenue reached $14.8B in one quarter, up 199% on the year before. NVIDIA sells the network that ties GPUs together, not just the GPUs.
3. A yearly cadence
A new platform nearly every year: Hopper H100 in full production in Sept 2022, Blackwell announced Mar 2024, Blackwell Ultra Mar 2025, Rubin unveiled Jan 2026, and Vera Rubin described as in full production by Aug 2026.
H100 · Blackwell · Blackwell Ultra · Rubin · Vera Rubin
A timeline: NVIDIA and the alternatives
Select any point to see what happened and the source. On narrow screens the timeline becomes a list.
Select a point on the timeline.
The alternatives, side by side
| Option | Who makes it | How you get it | Latest milestone |
|---|---|---|---|
| AMD Instinct | AMD | Servers from major makers and some clouds | MI400 Series launched Jul 23, 2026 |
| Intel Gaudi | Intel | Servers | Gaudi 3 launched Sep 24, 2024 |
| Google TPU | Google Cloud only | Ironwood generally available Mar 31, 2026; eighth generation announced Apr 22, 2026, no availability date | |
| AWS Trainium and Inferentia | Amazon Web Services | AWS only | Trainium3 UltraServers generally available Dec 2, 2025 |
| Microsoft Maia | Microsoft | Inside Microsoft's own services | Maia 200 announced Jan 26, 2026 |
| Open standards: Ultra Ethernet and UALink | Ultra Ethernet Consortium and UALink Consortium | Multi-vendor networking specifications | Ultra Ethernet 1.0 released Jun 11, 2025; UALink 1.0 released Apr 8, 2025 |
On software: AMD describes ROCm as its open-source AI software stack , and NVIDIA's CUDA and tools are covered under “Why NVIDIA leads” above.
The practical difference for most organizations: NVIDIA and AMD sell chips that end up in servers you can buy or rent in many clouds; Google, AWS and Microsoft build chips that you only reach through their own cloud services.
AMD's data center segment brought in $6.7B in the second quarter of 2026, up 107% on the year; AMD does not report a separate AI accelerator figure.
What it does to your IT estate
This is King Tech's ground: power, cooling, network, monitoring, cost and buying decisions.
Power
US data centers used about 176 TWh in 2023, 4.4% of US electricity, and are projected to reach 6.7% to 12% by 2028.
What it means: AI load competes with everything else for power and grid capacity; budget and facilities plans need to account for it.
Rack density
In Uptime Institute's 2026 survey of 800+ operators, the typical (modal) rack density was above 11 kW and 24% reported peak racks of 30 kW or more (19% in 2025).
What it means: a single AI server can draw more than an older rack was designed for. Check power and cooling before any purchase.
Cooling
TrendForce expects liquid cooling to be used with about 33% of AI chips in 2025, rising to 53% in 2026.
What it means: high-end AI hardware increasingly assumes liquid cooling, which most office and school server rooms don't have.
Network
NVIDIA's networking revenue reached $14.8B in one quarter, up 199% on the year. The Ultra Ethernet Consortium released a multi-vendor Ethernet specification for AI and HPC in June 2025.
NVIDIA, Q1 FY2027 results · Ultra Ethernet Consortium, Specification 1.0
What it means: AI clusters need fast, lossless networks; even cloud-based AI raises the bar on internet and WAN capacity.
Hardware prices
Gartner expects DRAM and SSD prices to rise 130% by the end of 2026, lifting PC prices 17% and cutting 2026 PC shipments by 10.4%.
What it means: AI demand for memory is pushing up the price of ordinary PCs and servers, so refresh budgets need headroom.
Cloud spend
Gartner forecasts AI-optimized IaaS spending of $42.3B in 2026, up 96.4%, with inference ($23.3B) overtaking training ($19B) and reaching 55% of that spend.
What it means: most organizations will rent AI capacity, and inference is now the bigger bill, so FinOps for AI matters.
Where most organizations really are
of US businesses used AI in May 2026
of AI use cases in IT operations fully met their ROI goals
of school districts report established AI guidelines
of core-subject teachers used AI for school in 2024-25
For most mid-sized organizations, school districts and cities, AI arrives through cloud services, software subscriptions and new PCs, not through racks of GPUs. The decisions that matter are about data, security, cost and readiness.
Questions to ask before you buy AI infrastructure
- What will it run, and could a cloud service or software subscription do the same job without new hardware?
- Do we have the power, cooling and rack space for it, measured, not assumed?
- Can our network and internet links carry the data it needs?
- Who will monitor it, and do our monitoring tools understand GPUs?
- What does it cost per month to run, and who owns that budget?
- How hard would it be to switch platforms in three years?
Is your estate ready for AI?
Eight questions, about two minutes. You get a scorecard with steps to fix each gap. Your answers stay in your browser unless you choose to send a project request.
The check needs JavaScript.
Planning for this in your organization?
King Tech collaborates on custom solution architecture: we work alongside your team to design an approach that fits your systems, your constraints and your timeline. Contact King Tech to start the conversation.
Sources
Facts checked Oct 12, 2026.
Company names are trademarks of their owners. King Tech is independent and isn't affiliated with any company named on this page. Figures are from each company's own filings and announcements or from the analysts named, as checked on Oct 12, 2026.