The Most Valuable Pickaxe in the AI Gold Rush

In 1Q26, NVIDIA lost China, wrote off $4.5B, and still posted record growth.

Wondering why the market is up today? Nvidia is the Titan holding up the SPX. Surging 4.76% to $3.43T in market cap.

Last night, NVIDIA reported Q1 FY26 earnings. They lost access to the $50 billion China market, wrote off $4.5 billion in unsellable chips, and still grew revenue 69% year-over-year to $44 billion. You’d think something would slow them down eventually, but it turns out no: not China, not trade policy, not inventory write-downs. Not even physics, apparently.

Let’s do the numbers first, because this is still technically an earnings story:

🧾 What happened

• Revenue: $44.1B (↑69% YoY), beating expectations

• Data Center: $39B (↑73%), now ~90% of total revenue

• $4.5B write-off: from H20 chips blocked by U.S. export bans to China

• Q2 Guide: $45B revenue (↑50% YoY), despite an $8B headwind from China

The stock surged. Jensen Huang smiled. Somewhere on Wall Street an analyst revised their AI TAM model again. Somewhere in D.C., a trade advisor quietly Googled "what is a reasoning agent."

NVIDIA didn’t just grow, it reallocated, absorbed geopolitical shocks, ramped the fastest product launch in its history, pivoted from selling to hyperscalers to selling to nations, and told Washington (politely) that export restrictions might be helping Beijing more than D.C.

This was less a quarterly update, more a State of the AI Union - delivered by a man whose GPUs are now literally powering civilization.

Let’s get into it:

🪓 The China Decoupling Is Real- and NVIDIA Is Already Moving On

Mid-quarter, the U.S. banned NVIDIA from selling its Hopper chips to China. Again. No grace period, no warning. One day they’re legal, the next they’re geopolitical contraband. NVIDIA shrugged and said: “Okay. Guess we’ll sell them to, uh, everyone else.”

The H20 export ban ended our Hopper Data Center business in China... We are taking a multibillion-dollar write-off.

The company ate a $4.5B loss like it was a light afternoon snack - and still posted record growth. They’re not banking on a return to China anytime soon. And they’re not panicking.

Instead, they’re pivoting to the UAE, Saudi Arabia, Taiwan, Sweden - places with less regulatory drama and more petrodollars. NVIDIA’s non-China demand engine is strong enough to sustain growth even with one of its largest markets cut off.

In one of the most interesting moments on the earnings call, Jensen critiqued the American trade policy.

China's AI moves on with or without U.S. chips. It has the compute to train and deploy advanced models. The question is not whether China will have AI, it already does. The question is whether one of the world's largest AI markets will run on American platforms.

The U.S. has based its policy on the assumption that China cannot make AI chips. That assumption was always questionable, and now it's clearly wrong. China has enormous manufacturing capability. In the end, the platform that wins the AI developers win AI . Export controls should strengthen U.S. platforms, not drive half of the world's AI talent to rivals.

The subtext: Export controls aren’t limiting China. They’re limiting America.

---

🧠 Inference Is the New Goldmine

If training was the big AI story last year, inference is this year’s breakout sequel-and it's a box office hit. Jensen explained:

Inference is exploding. Reasoning AI agents require orders of magnitude more compute.

In fact, Microsoft processed over 100 trillion tokens in Q1, a fivefold increase YoY.

Models are no longer just parroting back one-shot answers. They’re reasoning, planning, and - if you believe the hype - thinking. That requires significantly more compute, and NVIDIA’s new Blackwell chips - 40x faster than Hopper for reasoning workloads - are the only machines fast enough to keep up.

Blackwell chips (B200, GB200 NVL72) contributed ~70% of Data Center compute revenue. Hyperscalers like Microsoft are deploying 72,000 GPUs/week. We’re entering the "inference phase" of AI deployment. Agents that think, not just respond, are creating new workloads that justify data center-scale rollouts.

---

🏭 Sovereign AI = New Growth Engine

Jensen underlined the importance of sovereign AI:

Every nation now sees AI as core to the next industrial revolution... Countries are racing to build national AI platforms.

Nvidia has a line of sight to projects requiring “tens of gigawatts of NVIDIA infrastructure” with “nearly 100 AI factories in flight, 2x YoY”.

This is infrastructure-scale computing, typically reserved for national utilities - now driven by AI demand. NVIDIA is no longer selling GPUs. It’s selling sovereign AI infrastructure. The comparison to power grids isn’t metaphorical, it’s literal. These deployments now sit in the same conversations as national energy policy and foreign diplomacy.

In the past, you’d ask who has oil. Today, you ask: who’s on the GB200 rack list?

---

🌐 Networking and Full-Stack Control Is the Moat

Spectrum-X is now annualizing at $8B. NVLink shipments exceeded $1B this quarter.

NVIDIA’s networking stack - NVLink for scale-up, Spectrum-X for scale-out - is how it locks in the entire AI lifecycle.

“Every major IT provider is partnering with us… Enterprise AI is just taking off.”

RTX Pro. DGX Station. Blackwell. NIMs. NeMo. Spectrum-X. NVLink. Omniverse. Isaac. Nvidia has built a full-stack digital deity complex, quietly absorbing workloads across consumer, enterprise, industrial, and national layers.

Once you sell the chip, you sell the thing that connects the chip to other chips. Then you sell the software that runs on top. Then the orchestration layer. Then the security layer. Then you sell the customer on the idea that no other stack will ever work again.

From training to inference to orchestration, they’re becoming the platform, not just the hardware.

---

To wrap up the call, Jensen Huang listed four “positive surprises” that are now basically secular tailwinds propelling the stock:

First, inference is no longer just a postscript to training. Thanks to reasoning models that “think” instead of regurgitate, token volumes are exploding. It turns out that giving an AI a plan, a memory, and access to tools makes it compute-hungry and expensive, like a teenager with a credit card.

Second, the U.S. quietly revoked the AI Diffusion Rule - a policy that restricted the export of advanced AI systems to certain countries under the logic that if fewer people had the tech, fewer bad things might happen. That logic has now been replaced by a more pragmatic one: if we don’t sell it, someone else will. The timing couldn’t have been better for NVIDIA, as half the world suddenly realized they need AI infrastructure the way they once needed broadband.

Third, enterprise AI is finally real. Not “real” as in slideware. Real as in: companies are running AI agents that don’t just autocomplete emails, they make decisions, analyze PDFs, and tolerate ambiguity. The company leaned in: “We envision AI agents as a new digital workforce capable of handling tasks ranging from customer service to complex decision-making processes.” Enterprise IT stacks (compute, storage, networking) now come pre-bundled with existential questions about machine agency. Progress.

Fourth, industrial AI is showing up alongside the re-shoring of manufacturing. The era of AI factories is here - factories that build AI for the robots that run the factories that make the chips that train the AI. If that sounds circular, it is. But it’s also lucrative.

Put it all together and you have a company building the infrastructure for a future where intelligence is industrialized, distributed, and - if NVIDIA gets its way- billed in 90-day increments. If Huang is right (he usually is), this is just the start of the infrastructure age of intelligence.