Net Neutrality and Platform Power: Revisiting the Debate in the AI Era

In May 2026, Bharti Airtel launched "Priority Postpaid" — India's first mass-market consumer product built on 5G network slicing, a technology that splits one physical network into virtual lanes tuned to different performance levels. Existing postpaid customers got automatically upgraded, with the promise of steadier speeds whenever a cell tower gets congested. Airtel added a record one million postpaid customers in the following quarter. Within days, TRAI was asking for clarifications, and a parliamentary committee flagged a specific concern: those priority lanes could come at the expense of prepaid users, who make up roughly 90% of India's mobile base. It's 2016's Free Basics fight again, just wearing 5G terminology instead of zero-rating language.

That's a useful place to start, because the AI era hasn't replaced the old net neutrality debate — it's stacked a new, structurally similar debate on top of it, one layer up the technology stack, and most people arguing about "AI and platform power" right now are really having the same argument India settled in principle back in 2018, without noticing it's the same argument.

The old debate, briefly

India's net neutrality fight was decided reasonably cleanly. When Facebook's Internet.org (later rebranded Free Basics) offered free data access to a curated set of partner websites — with Facebook effectively paying the toll on users' behalf — critics argued this let a private company decide which parts of the internet Indians could access without a data cost, giving Facebook's chosen partners a permanent, structural advantage over anyone not on the list. TRAI's response, following over a million public comments, was the 2016 Prohibition of Discriminatory Tariffs regulation: internet service providers could not charge different prices for different content, full stop. That principle — that whoever controls the pipe shouldn't get to pick winners among what flows through it — was upheld through legal challenges and eventually written into India's licensing framework in 2018.

The 2026 Airtel slicing episode reopens a version of that same question at the infrastructure layer, dressed up in newer technical language: does a telecom operator get to sell some users a faster lane at the direct expense of others sharing the same physical network, particularly when 90% of that network's users have no way to buy their way into the faster lane at all? That's still an unresolved live question in India right now, with TRAI's draft rules from August 2026 formally bringing slicing into the quality-of-service framework — a genuinely open regulatory question worth watching, separate from anything AI-related.

Where AI moves the same argument up a layer

Here's the more interesting shift, though: the "neutrality" question that used to live almost entirely at the network layer — does the ISP treat all traffic equally — is increasingly showing up at the application and model layer instead, where it's barely regulated at all.

Consider how "free" AI access actually works in 2026. Meta's Muse Spark model is built natively into WhatsApp, Instagram, and Messenger, with no usage limits and no paywall, monetized instead through advertising. For the roughly 500 million WhatsApp users in India, that means the AI assistant that shows up by default, at zero friction and zero marginal cost, is Meta's — not because it's necessarily the best available model for a given task, but because Meta owns the distribution surface those users already spend hours in every day. A competing AI company, however capable its model, has to fight for a download, a sign-up, and a habit change to reach the same user Meta reaches automatically, inside an app already installed on the phone.

This is structurally the same mechanism Free Basics used — a dominant platform makes its own service frictionless and free while leaving competitors to bear the full cost of access — just relocated from "which websites cost you data" to "which AI assistant shows up in your messaging app without you doing anything." Nobody's proposing to regulate this the way TRAI regulated zero-rating, largely because it doesn't look like the old villain (an ISP throttling a rival's traffic). But the economic effect on competition is comparable: a company that already owns the distribution layer can make its own AI product structurally cheaper and more convenient to use than any competitor's, regardless of underlying model quality — and it can do so entirely within its own app, with no telecom operator or regulator anywhere in the transaction to object to.

The agentic web adds a second, newer front

There's a second and even less settled version of this fight emerging as AI agents start browsing and transacting on the open web on people's behalf. Websites have historically had to treat human visitors roughly equally — a search engine crawler, a browser, a screen reader, all generally got the same content. AI agents and AI-crawlers complicate that arrangement in a genuinely new way: publishers and platforms are increasingly blocking or rate-limiting AI crawlers specifically, some demanding payment for the right to be crawled and used as training or retrieval data, others blocking AI traffic outright to protect ad revenue that depends on human eyeballs rather than agent summarization. Cloudflare and other infrastructure providers have built entire new product categories around detecting and managing this traffic differently from ordinary human browsing.

This is, in a real sense, the inverse of the classic net neutrality problem. The original fight was about whether network operators could discriminate against content based on what it was. The emerging fight is about whether content owners can discriminate against a specific category of visitor — AI agents — based on what's doing the visiting, charging some agents for access, blocking others outright, and creating exactly the kind of uneven, permission-based internet that net neutrality was originally designed to prevent at the network layer. Nobody built regulatory guardrails for this because it wasn't the shape of problem anyone anticipated when the principle was written.

Why this matters more, not less, in an AI economy

The reason this deserves more attention than it's currently getting is that AI concentrates platform power faster than previous technology waves did. This series has already covered how foundation model economics are creating two-tier competitive moats — a commoditized application layer most companies can compete in, and a capital-intensive infrastructure layer only a handful of firms can play in. Add distribution power on top of that, and the companies that already own the phone's default assistant slot, the messaging app billions of people open dozens of times a day, or the cloud infrastructure a startup's AI product has no realistic alternative to renting, hold a form of gatekeeper power that's arguably more consequential than anything a 2016-era telecom operator ever had, because it operates across the entire stack simultaneously rather than just at the wire.

What the AI-era version of this debate should actually ask

The old net neutrality question was simple and largely answered: should the company that owns the pipe get to pick winners among what flows through it? The AI-era question is structurally identical but harder to see, because the "pipe" is no longer just a telecom wire — it's a default assistant slot inside a billion-user app, a cloud platform a startup has no alternative to, or a crawler-access decision a publisher makes unilaterally about which AI companies get to see its content at all. India settled the network-layer version of this question with real public engagement and a clear regulatory answer in 2018. The application and model-layer version of the same question is being decided right now, by the platforms themselves, with essentially no equivalent public debate — which is exactly the condition that made Free Basics controversial enough to generate a million public comments a decade ago, except this time almost nobody's writing in.