Why $NET is the most interesting bet on the machine run internet

$NET, the network security and edge compute platform behind a large share of the modern web, is not just another cloud vendor riding the AI wave. It's positioning itself to charge tolls, write policy, and rent compute on whichever side of the agent economy the traffic actually flows through. That's the thesis. Everything below tries to prove or break it.

Start with the traffic. On June 3, 2026, $NET's own CEO Matthew Prince posted on X that "bots have now passed human traffic online for the first time in the Internet's history," something he'd originally expected for late 2027. It arrived 18 months early. Per Radar, the traffic monitoring product $NET publishes to the public, automated requests hit 57.5% of HTML web traffic versus 42.5% from humans. Six weeks later, the Q2 2026 earnings release confirmed the same crossover had happened on $NET's own network, and management went further: non-human traffic could be 1,000 times greater than human traffic within five years if the current trend holds.

That's not a stat you file away. That's a company telling you, in its own earnings report, that the customer it built the entire business for is quietly changing species.

Here's the thing though. Most of that new traffic isn't paying for anything. HUMAN Security's 2026 State of AI Traffic report found agentic AI traffic grew roughly 7,851% year over year. One analysis of $NET's own data found ClaudeBot crawls about 23,951 pages for every referral it sends back to the site it scraped. Training crawlers take content and return essentially zero. That's a real problem for every publisher on the internet, and it's the exact problem $NET decided to monetize instead of just police.

Enter x402. It's an open protocol, built around the old, mostly unused HTTP 402 "Payment Required" status code, that lets any site charge for pages, datasets, APIs, or MCP tool calls per request, settled in stablecoins, no signup required from the agent doing the paying. $NET is building its Monetization Gateway on top of it in partnership with $COIN, the crypto exchange handling settlement, opened to a waitlist on July 1, 2026. If it works, it turns the exact traffic that's currently free riding into a metered utility bill, and $NET sits in the middle, taking a cut of every meter.

One. Traffic. Two. Payments. Three. Compute. That's the actual bet, in three layers, and $NET is trying to own all three at once.

The compute layer is the part that's genuinely interesting rather than just defensively smart. This isn't only about agents visiting the web, it's about agents building it. $NET open sourced VibeSDK, a full vibe coding platform that runs entirely on Workers for Platforms, deploying every generated app into its own isolated Worker with no shared server to fight over. Kenton Varda, the engineer who created Workers in the first place, has been building a related side project called Gadgets, a personal AI code generation platform where every app a user's agent builds is a shareable artifact, sandboxed enough that a whole category of security bugs stops mattering. His argument, roughly: 25 years of server centric cloud architecture and app store gatekeeping leave no real infrastructure for a future where every person's own agent writes custom software just for them. $NET's CTO reportedly blocked Varda from quietly releasing it on GitHub days before a conference talk, not because the idea was wrong, but because they wanted it shipped as a real product instead. That reads like a company arguing with itself about how big this is, which is usually a better sign than a company that already has all the answers.

There's a design detail underneath this worth sitting with. Agent native browsing infrastructure doesn't run Chromium. It doesn't need tabs, extensions, or 60 frames a second of pixel perfect rendering, because agents don't look at pixels, they read machine readable content. Running a full browser per agent session is prohibitively expensive at scale. Running a V8 isolate is not. It's the same kind of bet Amazon made building AWS for developers renting servers, except this time the infrastructure is being built for machines that mostly talk to other machines.

Now the numbers, because a thesis without financials is just a vibe.

Q2 2026 revenue came in at $696.1 million, up 36% year over year, ahead of the low end of guidance going into the print. Paying customer growth accelerated to 74% year over year, with more than 80,000 added in the quarter alone. Customers spending over $100,000 a year hit 4,698, up 27% year over year, the highest annual net addition for that cohort the company has reported. Gross margin held at 73.1%, free cash flow rose 69% to $56.4 million, and remaining performance obligations, essentially contracted future revenue, grew 38% year over year to $2.7 billion. On a trailing basis the picture is similar: revenue growth around 33.5% year over year, gross margin near 72.6%, free cash flow margin of 12.5% for full year 2025.

What the chart says right now

Fundamentals tell you why to care. The chart is telling you the market already does. As of August 12, 2026, 7 of 9 tracked technical signals on $NET read bullish, and the one bearish reading is just short-term froth, not a trend break.

The stock is trading 44% above its 200-day moving average, and the 50-day crossed above the 200-day in a golden cross roughly 65 days ago, the kind of signal that tends to mark the start of a sustained uptrend rather than the end of one. ADX, which measures trend strength rather than direction, sits at 29, meaning buyers are firmly in control, not just drifting higher on thin conviction. MACD flipped bullish about a week ago. In the latest session, $NET broke out to a fresh 20-day high, and it's currently sitting just 4.1% below its all-time high of $324.73, set August 7, after nearly doubling (+96.1%) off its 52-week low. It's also trading 28% above its Ichimoku cloud, another trend-following signal pointing the same direction as the rest.

The one caution flag: RSI at 68, a shade below the conventional overbought line at 70, and stochastics near 82, putting it at the top of its recent 14-day range. That's not a reversal signal by itself, it's the market pricing in a lot of good news quickly and giving it a rest at some point.

Zoom out further and the support structure tells its own story. Going back to August 2024, $NET has turned higher at 15 separate support levels and never once turned lower at a resistance level above the current price, because there isn't one; it's trading in price discovery territory above all prior highs. The nearest support sits at $259.44, about 16.7% below the last close, touched twice between November 2025 and May 2026. Below that, $228.29 (5 touches) and $217.04 (4 touches) mark the floors the stock has repeatedly bounced from over the past year. There's also an unfilled gap at $284.43 left by the August 7 session, when the stock opened 11.8% above the prior close, the kind of gap that sometimes gets revisited even in strong uptrends.

One more wrinkle worth naming honestly: a wave count exists that reads $NET's price action as being in the third leg of a five-part Elliott wave sequence, a pattern some technicians associate with the strongest, most extended part of a move. It's flagged as unconfirmed pending the next data pass, and it fails as a read if price closes below $250.69, so treat it as a live possibility, not a settled call. None of this is a signal to buy or sell. It's a description of what price has already done, and price can undo it fast.

The problems, named

$NET is not profitable. Net margin sits around negative 8%, and the GAAP loss actually widened in Q2 because of a restructuring charge tied to the AI first workforce transition, up to $165 million expected for the full year. The stock trades at roughly 44 times sales and 47 times book, rich by almost any measure. The average Wall Street price target, last updated in early August, actually sits below where the stock currently trades. That's not a detail to skip past because it's inconvenient, it's the exact tension a real thesis has to survive, and it's the same tension the RSI/stochastic readings above are hinting at from a completely different angle. To see how stretched that kind of valuation looks against the rest of the market: the companies trading above 40 times sales →

And insiders have been selling. Over the past 90 days, six insiders made 92 open market sales worth roughly $70.9 million, with zero purchases. That doesn't automatically mean anything, a lot of that kind of activity is routine or pre scheduled under a 10b5-1 plan, but it belongs in the open rather than in a footnote. You can screen for that exact pattern across other names yourself: companies where insiders sold in the last 90 days with no purchases →

The growth without profit setup isn't unique to $NET either, it's basically the defining shape of this entire AI infrastructure cycle. To see who else fits that profile: companies with revenue growth above 30% and negative net margin →

So why does the bull case still hold. Because the thing that has to be true for the bear case to win is that the machine majority internet stops needing a company that already sits in front of most of it, already has the payment rail live with a real partner, already has developers building on it by default instead of by being convinced to, and already has the chart's own momentum agreeing with the story. That's a lot to bet against. The thing that has to be true for the bull case to win is that x402 actually gets adopted at scale, that the restructuring turns into real operating leverage instead of just a bigger loss with a nicer name attached, that 36% growth doesn't decelerate into something the current price can't support, and that a stock sitting near record highs at 44 times sales doesn't need a sharper pullback than the $259 to $284 support band before it earns the next leg up.

Which of those happens first is unknowable. The infrastructure argument looks right more often than it looks wrong, but that doesn't resolve the valuation question or the short-term overbought reading. If growth slips under 30%, x402 stalls out at waitlist stage a year from now, or price closes below that $250.69 wave-count level, the honest read will be that the story got ahead of itself. Right now, both the fundamentals and the chart are still telling the same story.

Not financial advice. Do your own research.