🥛 The most important AI number 👀

GM. This is Milk Road AI, your daily download on the companies, infrastructure, and power players shaping the future of AI.

Here’s what we’ve got for you today:

✍️ Why NVIDIA can’t be stopped.
🎙️ The Milk Road AI Show: What The Economy Looks Like When Machines Can Pay Each Other.
🍪 Anthropic just landed a superstar.

Kalshi is a regulated prediction market where you can trade on real-world events. Special offer for Milk Road Readers: Trade $10 on Kalshi → Claim $10 for free.

Prices as of 10:00 a.m. ET.

WHY NVIDIA KEEPS WINNING

In 1953, the New York State Thruway opened 559 miles of highway connecting Buffalo to New York City.

The engineers who built it became forgotten, while the politicians who funded it got their names on plaques.

The people who got rich? The ones who owned the tollbooths.

It didn't matter who was driving, where they were going, or what they were hauling in the back, a family station wagon paid the same as an eighteen-wheeler.

Source: Pinterest

Jensen Huang built the only tollbooth on the highway to AI, and last week, he told us how much traffic is on the road.

$81.6B flowed through that tollbooth in a single quarter.

What actually happened

Let's run the numbers fast so we can get to the interesting part.

Revenue was $81.6B, up 85% year-over-year and 20% from last quarter, beating consensus by roughly $3B.

Source: Fiscal

Data Center alone, which is what this company has become, generated $75.2B, up 92% year-over-year.

For context, that's more than the annual revenue of most Fortune 100 companies from a single business segment in a single quarter.

Their net income alone was $58.3B, Wall Street had modeled $42.2B, and the beat was 38%.

And that gap is so large it raises a legitimate question about whether sell-side analysts are structurally incapable of modeling NVIDIA's operating leverage, or whether they just keep undershooting on purpose so they can keep beating.

Their free cash flow hit a record $49-50B in a single quarter, they raised the quarterly dividend 25 times over from $0.01 to $0.25 per share, and authorized an $80B share buyback with no expiration date.

This means management is so confident in the forward revenue picture that they are comfortable promising to return capital indefinitely.

Then they guided Q2 at $91B.

Goldman Sachs had penciled in $87.7B, and the broader Street was at $85-87B.

NVIDIA guided $4-6B above the most optimistic credible estimate on Wall Street, and they did it while explicitly saying they assumed zero revenue from China in that number.

We will come back to that because it is the most important sentence in the entire earnings release.

The $91B guide with zero China assumed is the equivalent of promising to run a 4:00 mile while carrying a 20-pound weight, the real question is what happens when the weight gets removed.

And the stock went down 3% after hours, before settling roughly flat which tells you everything about the situation NVIDIA now finds itself in.

Expectations have become so stratospheric that beating Wall Street by billions isn't enough anymore.

The number everyone missed

Every headline fixated on $81.6B, but almost nobody wrote about the number buried three lines down in the Data Center breakdown.

NVIDIA's networking revenue hit $14.8B in Q1, up 199% year-over-year.

Pause and actually sit with that. $14.8B in a single quarter, from networking.

That one line item, not the GPUs, just the cables, switches, and connectors, would rank as a top-10 chip company on earth by quarterly revenue if it were its own standalone business.

This is every switch, every NIC, every InfiniBand connector needed to stitch tens of thousands of GPU chips together into a single system large enough to run a frontier AI model.

Here is the thing about AI clusters that doesn't get enough attention: as they scale from 10,000 GPUs to 100,000 GPUs, the networking spend doesn't grow at the same rate as compute, it grows faster.

More GPUs mean exponentially more connections, which means exponentially more networking infrastructure.

NVIDIA doesn’t just own the tollbooth; they own the entire highway system around it and charge for every part of it.

The 199% growth in networking is the clearest signal in the whole report that the AI buildout is accelerating.

The AI clusters are getting larger and more complex, and NVIDIA is the only company selling both the compute and networking fabric connecting them at scale.

CRYPTO’S THIRD PMF MOMENT IS HERE

PMF = Product Market Fit

Until now, crypto has really had two major PMF moments: 24/7 trading and stablecoins.

Now a third one may be emerging with Kalshi: Prediction Markets.

Kalshi is a regulated prediction market exchange where people trade on the outcomes of real-world events (elections, inflation, sports, crypto prices and more).

And their growth has been massive:

Recently raised $1B at a $22B valuation
Institutional trading volume is up 800% in six months.
Annualized volume is somewhere around $178B.

Kalshi is giving Milk Road readers $10 free when you trade $10.

Trade $10 → Claim $10 on Kalshi.

WHY NVIDIA KEEPS WINNING (P2)

Here is what most people missed while they were arguing about whether $81B was good enough guidance.

Jensen Huang announced that NVIDIA is entering the CPU market.

For its entire existence, thirty years, give or take, NVIDIA has been a GPU company.

The GPU is the thinking part of AI, it crunches through the matrix math that turns a pile of text into a coherent response, a blurry image into a sharp one, or a sensor reading into a self-driving decision.

NVIDIA controls 92% of the GPU market, and its CUDA software is so deeply embedded in AI that switching away is far easier said than done.

But here is the problem with GPUs that nobody really talks about: they are brilliant at thinking and completely useless at everything else.

When an AI agent isn’t actively thinking, CPUs handle the behind-the-scenes work like memory, routing, scheduling, and tool calls.

Right now, NVIDIA GPUs do the intelligence while Intel, AMD, and Arm CPUs handle the paperwork.

NVIDIA announced Vera, their own CPU, purpose-built specifically for agentic AI orchestration.

Source: NVIDIA News

And Huang said it opens a $200B total addressable market that NVIDIA has never competed in before.

This is a land grab into territory that two legacy chipmakers have spent decades defending, and NVIDIA just showed up with $20B in visible FY2027 CPU revenue already in the order book before the product ships at volume.

The strategic logic here is almost uncomfortably clean.

An AI agent running on an NVIDIA GPU already depends on NVIDIA's CUDA software.

If NVIDIA can also get the CPU that orchestrates that same agent, they own the entire compute stack, GPU for thinking, CPU for managing, and NVIDIA networking fabric connecting everything. One vendor. Completely locked in. For the entire AI factory.

It's a second tollbooth on a highway that is still being built, and they already have the contracts to prove the traffic is coming.

The parts keep getting wrong

There is a legitimate bear case on NVIDIA, and it deserves a fair hearing.

YoY revenue growth is decelerating.

NVIDIA’s revenue growth was 250% in 2024, 85% today, and will likely continue slowing from here.

This is math, and markets are forward-looking machines that will price the deceleration before it shows up clearly in the revenue line.

Custom ASICs from Google and Amazon are now in the tens of billions, and Broadcom expects its AI ASIC revenue to grow 60% in 2026.

Gross margins came in at 74.9%, below the 78% levels NVIDIA was posting in FY2024-2025, and the transition from Blackwell to Rubin creates margin pressure that repeats every product cycle.

And 60% of NVIDIA's trillion-dollar revenue forecast comes from exactly four customers. If any one of them blinks, the impact is immediate and ugly.

These are real risks and anyone who tells you otherwise is selling something.

But here is what the bears keep getting structurally wrong about the tollbooth analogy: they keep modeling NVIDIA as if traffic on the highway might slow down.

Google, Microsoft, Amazon, and Meta are collectively spending $725B on AI infrastructure in 2026.

That is up 77% from a year that was already a record.

These are the four most profitable tech companies in history, all independently concluding they need more compute and all arriving at the same destination: Jensen Huang’s tollbooth.

The traffic isn’t slowing down, the highway is just getting bigger.

And then there is the China situation, which is the most underappreciated number in the entire earnings report.

NVIDIA guided Q2 at $91B with zero China revenue assumed.

But ByteDance, Alibaba, and Tencent have already received approval to purchase over 400,000 H200 units.

If even a fraction of those orders convert to revenue in the back half of 2026, the guidance is conservative by billions, and every dollar that comes in from China is pure upside against a number Wall Street already thinks is too low.

The tollbooth is already collecting $81.6B a quarter, with the China lane completely closed.

Imagine what happens when they open it back up.

Alright, that's it for this edition of Milk Road AI. We want to hear from you.

What's NVIDIA's biggest risk from here?

WHEN MACHINES PAY EACH OTHER 🤖

In today's episode, we sat down with Guillaume Ponce, Head of Engineering at Alchemy, to talk about how blockchain infrastructure is being rebuilt for an agent-first future where AI transacts without humans in the loop.

Here's what you'll hear:

Why payments are the real bottleneck for agent economies and how crypto rails can finally unlock machine-speed settlement.
How Alchemy is shifting from dashboards to APIs as the primary interface, with new primitives like Agent Cards and agent-pay.
Real use cases where agents commission ads, hire micro-contractors, or buy assets for anywhere from $0.50 to $100 per task.
Where the biggest opportunities sit for builders: agent marketplaces, payment orchestration, identity, and attestation layers.

Don't sleep on this one 👇️

YouTube | Spotify | Apple Podcasts

Real Finance Blockchain is an EVM-compatible L1 that is built specifically for RWA tokenization. Read more about Real Finance Blockchain here.

Arc is an open Layer-1 blockchain built for real-world financial applications. Public testnet is live, App Kits are here, and mainnet is getting closer. Start building on Arc.