Who Actually Controls Your Information?

Quick Summary

The AI race isn’t about who builds the cleverest chatbot – it’s about who controls the data, the infrastructure and default settings that billions of people never question. Google and OpenAI are locked in a battle that’s being marketed like a tech thriller, complete with heroes and villains. Strip away the noise and you’re left with something far less exciting but infinitely more important: a fight over who gets to decide what information you see, which creators get paid, and whether the open web survives or gets swallowed by AI summaries that keep you trapped inside a single company’s ecosystem. Neither deserves your uncritical support. Both deserve your attention.

Key Points

• Google’s advantage is structural, not innovative – it owns the pipes (search, Chrome, Android) and can push AI into products you already use without asking permission
• OpenAI still sets the innovation pace – it ships first, captures imagination, and holds the frontier model crown, even as Google catches up fast
• Data moats matter more than clever code – Google owns behaviour data from billions of logged-in users; OpenAI scraped the web and hopes fair use saves it
• Cost structures reveal fragility – Google owns the factory (TPU chips, data centres); OpenAI rents compute from Microsoft and runs a high-risk margin game
• AI search might kill the web it feeds on – Google’s AI Overviews are cutting publisher traffic by 40-50%, strangling the ecosystem that made search valuable
• Regulation is the only real brake – the EU and UK can slow Google’s dominance and threaten OpenAI’s training data model, but neither will be knocked out
• It’s not a two-horse race – Anthropic is winning enterprise spend, Meta is flooding the market with free models, and open source is quietly breaking the monopoly
• Control beats cleverness – the real prize is owning the default assistant in your phone, browser, and search box, not having the “smartest” model

Google, OpenAI, and the AI Delusion: Who Actually Holds the Power?

The AI race is being sold like a Marvel film. On one side, the heroic startup that will “end poverty” and “fix climate change”. On the other, the search giant that already owns half the internet. The truth is uglier, more boring, and far more important: this isn’t about who has the cleverest chatbot. It’s about who controls the pipes, the data, and the default choices billions of people never think about.

What follows is not a fanboy love letter to Google or a hit piece on OpenAI. It’s a blunt look at who actually has the structural advantage, where that story falls apart, and why none of us should be relaxed about any of them.

Google’s Quiet Advantage: Pipes, Not Hype

Google doesn’t need to “win AI” to survive. It’s already making tens of billions a year from ads, search, cloud and apps. AI is an upgrade to a working money machine, not a last-throw gamble. OpenAI is the opposite: almost all of its revenue is some flavour of “rent our model”, with multi-billion compute bills and near-total dependency on keeping that lead.

Geoff Hinton, who helped invent half the techniques everyone is now bragging about, said recently that Google is “beginning to overtake” OpenAI and that it’s more surprising it took this long. He points at the obvious things: Google has its own chips, ridiculous amounts of data, huge data centres, and armies of good researchers.

Google’s real trick is distribution, not genius:

It pushes Gemini into Search, Android, Chrome, Maps and Workspace, so users “adopt AI” by doing exactly what they already do.

Its AI Overviews and AI Mode sit where organic results used to be, turning the search page itself into an assistant.

From a power perspective, that’s brutal: OpenAI has to persuade you to open ChatGPT. Google just has to persuade you not to change your default search or browser.

The Uncomfortable Counterpoint: OpenAI Still Sets the Pace

If you only looked at infrastructure and cash, you’d call the race for Google and go home. That would be lazy. On innovation and mindshare, a lot of credible analysis still puts OpenAI first.

Recent scorecards rate OpenAI as:

First on “frontier model” innovation, with GPT-5 leading agent-style use cases and complex reasoning.

First on consumer brand and daily usage inside a single app (ChatGPT), even as growth slows and competition bites.

Google is often rated as “catching up fast” rather than definitively ahead, with Gemini praised but still framed as part of a close race. Anthropic’s Claude is quietly overtaking both in enterprise spending for certain use cases, which makes the narrative even messier.

So any honest piece needs to admit this: OpenAI still sets a lot of the agenda. It ships first more often, it still captures imagination, and its biggest partner (Microsoft) can push GPT-5 into Office and Windows at massive scale.

Google has the deeper foundations. OpenAI still throws the bigger conceptual punches.

Data: Who Actually Owns the Fuel?

It’s easy to say “Google has more data” and leave it there. The detail matters.

Google has:

First-party behavioural data from Search, Chrome, Android, Maps, YouTube, Gmail, Docs and billions of tagged page views.

Logged-in user histories that tie queries, locations, clicks and purchases together in a way most companies can only dream of.

That data isn’t just text. It’s real-world behaviour at ridiculous scale, which is gold for ranking, personalisation and reinforcement learning.

OpenAI’s fuel is very different:

Huge web crawls, public datasets, code repositories and licensed corpora from publishers, media and platforms.

Increasingly controversial training on copyrighted and paywalled material, including evidence of O’Reilly books and other commercial works in its training mix.

That mix makes OpenAI powerful and vulnerable at the same time. It can boost quality and coverage, but it hands every rights holder a potential lawsuit and every regulator a reason to dig.

Google has its own copyright headaches, but crucially it can lean far more on data generated inside its own products, under its own terms, with user consent folded into the fine print years ago. That’s a quieter, deeper moat than “we scraped the web and hope fair use saves us.”

Cost and Compute: Who Owns the Factory?

AI at this scale isn’t just about clever algorithms – it’s about owning or renting the factory that runs them.

Google:

Designs and runs its own TPU chips, controls the data centres, and sells that capacity to others through Google Cloud.

Can spread the cost of training and serving Gemini across Search, Ads, YouTube, Cloud and Workspace, so the economics don’t have to stand up as a separate product from day one.

OpenAI:

Buys the bulk of its compute from Microsoft’s Azure. Even with deep partnership pricing, it’s still a tenant, not the landlord.

Has built a £8-12 billion ARR business on top of that, but public reporting still shows huge spend on compute and R&D, with profitability fragile and tightly coupled to model costs.

On raw cost structure, Google has the advantage. It owns the factory and can charge others to use it. OpenAI is running a high-risk, high-margin software business on somebody else’s hardware.

The counterpoint is simple: being lean and dependent can also force OpenAI to move faster, experiment more aggressively, and chase bolder use cases than a giant with a money-printing search empire. The question is whether that speed translates into durable power, or just better demos.

Google’s AI Search Problem: Winning by Eating Its Own Ecosystem

On the surface, Google’s strategy looks flawless. Answer more queries directly. Keep users on the results page. Use Gemini to summarise the web so people never need to click.

Underneath, it might be sawing off the branch it’s sitting on.

Evidence from publishers, SEO platforms and media groups is grim:

Analyses show AI Overviews cutting click-through rates to publishers by around 40-50% on some result types.

The UK’s PPA and Europe’s Independent Publishers Alliance accuse Google of “double daylight robbery”: using publisher content to train models, then using AI summaries to keep traffic away from those same publishers.

Some publishers warn that without intervention, a third of independent outlets could face closure because AI search breaks the link between visibility and revenue.

Google, predictably, insists that AI features are “increasing overall traffic” and that it sends “more clicks than ever”, but that claim is increasingly at odds with the numbers publishers are sharing.

So yes, Google’s AI integration is a competitive superpower. It’s also a real risk to the information ecosystem that made Google valuable in the first place. If you strangle the web with AI summaries, what’s left to summarise in ten years?

Regulation: The Only Thing Big Enough to Hurt Them

Regulation is messy, slow and often behind the curve. It’s also the only realistic brake on the power grab both companies are attempting.

For Google, the threat is structural:

In the EU and UK, it’s already classified as a gatekeeper, and AI Overviews are triggering antitrust complaints for self-preferencing and misusing publisher content.

The EU AI Act and DMA give regulators tools to demand transparency on training data, limit how AI can be combined with ads, and force interoperability or opt-outs for content owners.

For OpenAI, the threat is existential:

It faces lawsuits and investigations about whether training on copyrighted and personal data is lawful, and whether outputs leak or replicate protected works.

Because OpenAI is effectively “just the model”, a hard regulatory hit on training data or output liability goes straight to its core business, rather than being softened by ten other revenue streams.

There’s a middle ground here: regulators can slow Google and hurt OpenAI at the same time, while still leaving both utterly dominant compared with almost anyone else. Anyone writing honestly about this should admit that “regulation will save us” is at best a partial comfort.

The Rest of the Field: Anthropic, Meta, and the Open-Source Wildcard

Treating this as a two-horse race is comforting for headlines and completely wrong.

Anthropic has:

Overtaken OpenAI on some measures of enterprise LLM spend, with surveys showing Claude now capturing 30-40% of US enterprise LLM budgets, ahead of OpenAI.

Built a reputation for compliance, safety and enterprise-friendly tooling that large organisations are increasingly prioritising over raw capability.

Meta is:

Flooding the world with Llama models under permissive licences, making “good enough” AI essentially free for anyone willing to host it.

Baking AI assistants into WhatsApp, Instagram and Facebook, which gives it stupidly broad reach in chat and social, even if its models lag the absolute frontier.

Open-source in general means:

Smaller companies can deploy strong models on their own infrastructure, keeping data in-house and avoiding dependency on any single US mega-corp.

The idea that Google or OpenAI will tax every AI interaction like they do search is much less certain than it looked in 2023.

If you care about concentration of power, this is the only vaguely hopeful bit: there are real alternatives, and some of them are already winning in specific slices of the market.

The Boring Truth: This Is About Control, Not Cleverness

The loud story is “which model is better?” Gemini vs GPT-5 vs Claude 4.5, benchmarks, “IQ scores”, viral threads about one model “feeling more human”. The more important story is who owns:

The default search box

The default browser

The default assistant baked into your phone, car, TV and operating system

On those fronts, Google is in the strongest overall position: the biggest share of search, the Android footprint, Chrome, Gmail, YouTube, Maps, Docs and a model family it can slide under all of it. OpenAI still leads in frontier innovation and brand, but is constrained by cost, dependency on Microsoft, and growing regulatory and legal heat around training data.

The middle ground is uncomfortable and honest:

Google probably has the best long-term odds of staying in charge of how people access information.

OpenAI still has a real chance to own the “assistant” relationship if it can keep ChatGPT ahead and turn it into something people deliberately choose, not just tolerate.

Anthropic, Meta and open-source tools together make it harder for either of them to own absolutely everything.

If you’re a normal user, publisher, or small business, none of this is neutral. The “winner” is whoever gets to decide what you see, what gets suppressed, what gets summarised, and which creators get paid versus quietly erased by an AI answer box.

What This Actually Means for the Rest of Us

The real fight isn’t about who builds the smartest model – it’s about who controls the infrastructure that decides what information reaches you, how it’s packaged, and whether the people who created it get paid or get buried. Google has the distribution advantage and the data moat. OpenAI has the innovation lead and the brand momentum. Anthropic is winning enterprise trust. Meta is making good-enough AI free. Open source is chipping away at the monopoly narrative.

None of them are your friend.

None of them should be trusted to act in your interest without scrutiny, regulation, and competition keeping them honest. The hype merchants want you to pick a side and cheer. The smarter move is to watch all of them carefully, support the alternatives where they exist, and push back hard when any of them try to lock down the future of information access behind their own walls.

That’s the bit the hype merchants rarely spell out.

Questions & Answers and AI Data