"Demand is going parabolic. Utterly parabolic."
That was Jensen Huang's opening line at Dell Technologies World 2026, repeating "parabolic" twice — not for rhyme, but because he had Dell's data to back it up.
Michael Dell followed: "The rate of change has gone parabolic, and it's not slowing down."
$64B Orders + $43B Backlog: The Scale of Enterprise AI Hardware
Dell disclosed:
- $64B: Cumulative AI-related orders for FY2026
- $43B: Backlog (orders not yet delivered)
- 5,000 enterprises: Total Dell AI Factory customers to date
- +1,000: New customers added this quarter
- 2x: Dell's rack-scale server shipments are double those of competitors
The last point is striking. AI data centers compete on delivery capacity — servers waiting for tape-outs, HBM, and liquid cooling lines. Whoever can ship thousands of complete rack-scale systems monthly captures the market.
Dell's customer list is not fictional:
- Eli Lilly: AI pipeline for drug discovery
- Samsung Electronics: Chip design and semiconductor fabs
- Honeywell: Industrial AI and digital twins
- Hudson River Trading: Algorithmic trading research — this high-frequency market maker has overtaken Citadel Securities in market share this year
67% of Agent Workloads Have Left the Cloud
Another number from Huang:
- 67%: AI workloads running outside the cloud (on-premises or edge)
- 88%: Share of enterprises running at least one local AI workload
The "AI is all in the cloud" narrative, pitched to capital markets over the past three years, has flipped. The majority has moved into enterprise data centers. The reason: agent workloads are long-tail, high-frequency, and persistent — cloud API costs based on token consumption become unsustainable.
Huang added: "What took months now takes weeks. What took weeks now takes days. And what takes days now takes hours." This is not hyperbole; it's the core argument Dell and NVIDIA are making: AI work speed has compressed, making hardware throughput the bottleneck.
Looking to 2030: A $3-4 Trillion Compute Bill
Huang also offered a long-term view:
- 3,400%: Projected token consumption growth by 2030
- $3-4 trillion: Global AI infrastructure spending by 2030
For context, Saudi Aramco's annual revenue is about $450 billion; OpenAI's recent compute contracts over a decade total several hundred billion. $3-4 trillion represents the ceiling of a unified global AI infrastructure market. Whether the numbers hold is debatable — NVIDIA has incentive to inflate the pie, Dell wants clients to order servers now. But one thing is clear: today's capital expenditure is already scaling to that level. Meta is spending $40 billion on AI this year, Microsoft $80 billion, and OpenAI projects $850 billion by 2030. Dell's $64 billion in orders is a slice of that.
Good Times for Hardware Companies
Dell's stock reacted mildly — an initial pop, then a slight decline. The market has heard "AI demand parabolic" before and doesn't always buy it. But the numbers don't lie:
- 5,000 new AI Factory customers in a year
- 1,000 more in a single quarter
- $64 billion order book
- $43 billion backlog
Even if the global macro environment deteriorates sharply, this backlog provides a year's worth of revenue. A year ago, Michael Dell faced market skepticism — thin AI server margins, declining PC business, and a depressed stock price. Now, the story has shifted from "anxious PC maker in transition" to "AI blacksmith overwhelmed by orders."
One side of the parabola is Dell's capacity curve; the other is CFOs' budget limits. How these two lines intersect in the second half of 2026, Huang didn't say. But the 1,000 new customer contracts signed at the event tell you: clients aren't waiting.
Sources: NVIDIA CEO Jensen Huang at Dell Technologies World: "Demand Is Going Parabolic, Utterly Parabolic" (NVIDIA Blog); Dell Adds 1,000 Clients for AI Servers, Targets Corporate Users (Bloomberg); Dell Technologies Closes the Gap Between AI Ambition and AI Outcomes (Dell Press Release); CocoLoop