section 02 of 11

Google Is Kodak

That was not a history lesson. It is happening right now, at a scale that dwarfs photography, to a company that makes Kodak’s foresight look negligent by comparison.

In 2017, a team at Google Research published a paper called “Attention Is All You Need.” It introduced the transformer architecture — the foundational technology behind GPT, Claude, Gemini, and every large language model that now dominates the headlines. Google didn’t just see the future of AI coming. Google built the lens.

They have the talent. They have more data than any company on Earth. They have the computational infrastructure to train models at a scale few competitors can match. They launched Gemini, invested billions, acquired DeepMind years before the current wave began. By every metric that mattered in the Kodak story — early awareness, technical capability, willingness to invest — Google is better positioned than Kodak ever was.

And they are in the exact same trap.

Google’s search advertising business is the most elegant money machine since Kodak’s film. A user types a query. Google displays results. Advertisers pay to appear among those results. The marginal cost of serving one more search is effectively zero. The revenue is recurring, almost passive, and enormous — more than two hundred billion dollars a year. Every search is a tiny transaction where the user gets information and Google gets paid. The machine is so efficient that it feels like a natural law.

Now consider what an AI assistant does. A user asks a question. The AI gives the answer. There is no results page. There is no natural place for an ad, no moment where the user’s attention can be auctioned to a third party. Every query that an AI answers directly is a query that never generates a click, never loads an ad, never completes the transaction that funds Google’s existence. The better the AI works, the faster it dismantles the business model that pays for it.

Google sees this clearly. They are not blind. But seeing the trap and escaping it are different things, and Kodak already proved that. The question Google cannot answer — the question no one in the industry has answered — is how you monetize a conversation the way you monetize a search results page. A subscription model generates a fraction of the revenue. A usage-based model collides with the fact that every AI interaction costs real money in compute, unlike search, where the marginal cost was nearly nothing. The economics are not just different. They are inverted.

None of this shows in Google’s numbers yet. Search revenue is still growing, and Google says the ads around its AI answers earn as much per query as the ones on the old results page. That is exactly what the trap looks like from inside. Kodak’s revenue peaked in 1996, two decades after its own engineer built the digital camera, and film kept selling for years while its replacement was already on the shelf. An AI answer laid on top of a results page is still a results page. The trap closes when the assistant stops sending anyone to the page at all — when the agent books the flight, buys the product, files the form — and that part has barely begun.

This is not only Google’s problem. The entire AI industry is running on the assumption that scale will fix the economics — that costs will fall, efficiency will rise, and eventually the numbers will work. OpenAI is burning cash at a staggering rate. So is Anthropic. So is every company racing to build foundation models. The assumption may prove correct. It is also, for now, an assumption and not a fact, and the history of technology is littered with industries that scaled beautifully and never turned a profit.

But the deeper Kodak parallel is not about any single company’s balance sheet. It is about the convergence that nobody is correctly pricing in.

Right now, every major software company is adding an AI copilot to its existing product. Microsoft adds Copilot to Office. Adobe adds AI to Photoshop. Salesforce adds AI to its CRM. The strategy is obvious: enhance the existing product, preserve the existing subscription, defend the existing margin structure. This is the equivalent of Kodak launching a digital camera — a rational response that completely misreads the nature of the disruption.

The disruption is not “your product, but with AI assistance.” The disruption is an AI that does the task without needing your product at all. A copilot helps you write a better email in Outlook. An agent writes the email without Outlook. A copilot helps you build a better spreadsheet. An agent does the analysis without the spreadsheet. A copilot helps a consultant produce a better slide deck. An agent produces the strategic recommendation without the consultant.

Every incumbent adding a copilot to their product is defending a margin structure that may be about to cease to exist — not because the product becomes obsolete in some slow, manageable way, but because the economic structure beneath it evaporates, just as the printed photograph evaporated beneath Kodak. And just like Kodak, the companies that were born digital, the ones with no legacy to protect, will be just as vulnerable. Being native to the current paradigm is no protection when the paradigm itself is what’s disappearing.

And this is what Google was chosen to make plain. Not that Google will die — it has the models, the capital, and the reach to survive almost anything, most likely by becoming something we don’t yet have a name for. The point is that if the trap can close on the company that invented the technology, the one that saw it first and clearest and richest, then no company’s scale, foresight, or head start is a defense. Kodak was a story about one firm in one industry. Google is the proof that the story is now general. Every company should look at the printed photograph and ask whether it is holding one — whether its product, stripped to the thing the customer actually wanted, is something an agent now does without it. Most will be certain they are the exception. So was everyone before them.

This comparison still flatters us. Kodak lost a business. Google might lose a business. These are corporate tragedies, not civilizational ones. The reassurance forming in your mind right now — that we have been through disruptions before, that the industrial revolution also destroyed old industries and created new ones, that humanity adapts — is the most dangerous comfort of all.

going deeper · optional

Why this holds the mechanism, and where it's soft

The section’s claim: Google is structurally trapped the way Kodak was, and it’s happening now. Here’s the load-bearing part, and the part that’s a bet.

Google did build the foundation. In 2017 a team at Google published “Attention Is All You Need,” introducing the transformer — the architecture under GPT, Claude, Gemini, and effectively every current large language model. This is not a company that missed the technology. It originated it. That’s what makes the Kodak parallel bite: like Kodak, Google saw it first and clearest.

The trap is a margin structure, not a technology gap. Search advertising is close to a perfect business: the marginal cost of one more query is almost nothing, and the results page auctions attention to advertisers on every search. That is why it throws off enormous, recurring, near-passive revenue — well into the hundreds of billions annually across Google’s ad business. An AI assistant that answers directly has no results page, no ad slot, no attention to auction, and a real compute cost per response. So the better the assistant, the more it erodes the surface that funds it. Nobody has yet shown how to monetize a conversation at search-page economics: subscriptions bring in a fraction, usage-based pricing collides with per-query compute cost. The economics aren’t just lower — they’re inverted, from near-zero marginal cost to real marginal cost.

The revenue hasn’t turned, and that is not a refutation. Google Search revenue grew through 2025 to roughly $225 billion, and growth accelerated into 2026 (about 19% year on year in Q1). Google reports that ads shown with AI Overviews monetize at about the same rate per query as ads without them. Taken at face value, that shows AI answers can be grafted onto a results page without loss. It does not show that a conversation or an agent can be monetized at search-page economics, because an AI Overview is still a results page with ads around it. Kodak’s own revenue peaked in 1996, near $16 billion, two decades after Sasson’s camera: the trap was visible years before it showed in the accounts. What would refute the section is agentic or conversational use monetizing at margins comparable to search, not search growing while the agent layer is still small.

The cash-burn point is real but contingent. The industry is running on the premise that scale drives costs down until the numbers work. Frontier labs are, by broad and repeated reporting, spending far more than they earn. That premise may hold — costs per token have fallen sharply — but “scale fixes it” remains an assumption, and the history of technology includes plenty of industries that scaled beautifully and never turned a durable profit.

Copilot versus agent — the mechanism. Incumbents are bolting AI assistants onto existing products (Copilot in Office, AI in Photoshop, AI in the CRM), which preserves the subscription and the margin. That’s the rational, Kodak-shaped move. The distinction the section draws is between an assistant that helps you use the tool and an agent that does the task without the tool. Where the second holds, the defended margin structure is the thing at risk — not the product’s quality, its reason to be paid for.

It’s not hypothetical — it’s already visible below Google’s scale. The clearest early casualty is in education technology: firms built on homework help and study aids saw usage and market value fall hard once students could ask a chatbot directly, and at least one prominent homework-help company’s leadership attributed the collapse in its growth explicitly to students switching to ChatGPT. That’s the Kodak shape at company scale on a compressed timeline — not a worse product, a vanished reason to pay for it.

What the section does and does not predict about Google itself is in the thinking-aloud note below.

— thinking aloud —

Let me be straight about what this section is doing, because the title invites a misreading.

“Google Is Kodak” is not a prediction that Google dies. If you’re reading it as a confident forecast of Google’s funeral, that’s an overreach, and I’ll own that the literal claim doesn’t survive scrutiny — Google has its own frontier models, more cash than most nations, and distribution to billions. It will very likely survive, the way Fujifilm survived, by becoming something whose economics we don’t recognize yet. Betting on Google’s collapse would be foolish, and I’m not.

The reason I still put Google at the top of the section is that it’s the hardest possible case for the reassurance everyone reaches for: “we’re too big, too smart, too far ahead to be Kodak.” Google saw the technology first — it built the transformer. It has every advantage Kodak lacked. And it is still holding a razor-and-blades machine it has to cannibalize to compete. If the trap can close on that company, the comforting exemptions — scale, foresight, a head start — are worth nothing. That’s the whole move. Not “Google is doomed,” but “if even Google, then who exactly is safe?”

Someone built the punchline better than I can argue it: a “SaaSpocalypse survival scanner” called Death by Clawd that asks whether your entire product could be replaced by a single Claude Skill — whether you are, functionally, just a .md file. It’s a joke, and it’s the whole section in one question, pointed at every company rather than one. That’s the register to read this in. Not a death notice for Google. A test I’m handing to everyone, Google included, and the uncomfortable part is how many companies fail it.

Objections 6 the strongest counters, answered
Google will just put ads in the AI answers. Problem solved.
Maybe — but it's unproven that it works at the same economics. Search ads monetize intent at the exact moment of a query, on a page built to auction attention, at near-zero marginal cost. An AI answer has no auction surface and a real compute cost per response. Google can absolutely put ads in Gemini; whether that reconstructs a hundreds-of-billions, near-passive margin is the open question. Nobody has shown it yet, and the section's claim is about that gap, not about whether ads are technically possible.
Google has its own top models, infinite cash, and distribution to billions. Comparing it to a bankrupt film company is absurd.
Fair — and Google being strong is the point, not a hole in it. Google sits at the top of the section precisely because it's the hardest case for 'we're too big and too smart to be Kodak': it invented the transformer and still can't step out of the trap. The claim was never that Google dies — it probably survives by becoming something else, the way Fujifilm did. The claim is that if the trap closes on even Google, then scale and foresight buy no exemption, and every company should run the test on itself. 'Google is fine' concedes nothing, because the target was never Google.
Compute costs are collapsing. At scale the economics obviously work.
They might — cost per token has dropped fast, and that's the strongest optimistic case. But it's an assumption doing load-bearing work, not a settled fact. Plenty of technologies scaled brilliantly and never found a durable profit. The section doesn't claim the economics can't work; it claims 'scale fixes it' is a bet the whole industry is making with real money, and betting isn't the same as knowing.
People still Google everything. Search isn't going anywhere.
Search as a behavior isn't the claim; the revenue model is. Every query answered directly by an assistant is one that doesn't load a results page or an ad. Search can remain enormous and still have its monetizable share erode at the margin — and margin erosion is exactly what breaks a business built on near-passive, near-total capture of query intent.
Google isn't Kodak. It has YouTube, Android, Cloud, Workspace, Waymo and DeepMind. Search can shrink and the company survives.
The company very likely survives; the section says so. But look at what the portfolio is made of. YouTube is advertising on attention, the same model as search. Android earns through the app store and default search deals, which erode under the same forces. Workspace is the clearest case of a copilot defending a product that an agent makes unnecessary. Gemini and DeepMind face the same question of how to monetize a conversation. Cloud is the one large business outside the trap, and it is not on the scale of search. The portfolio is less a diversification away from the trap than several exposures to it.
Google's search revenue is growing faster than ever, and Google says AI Overviews earn as much per query as classic search. The trap isn't closing — AI is making search more valuable.
For now, true, and it's the strongest objection to this section. But an AI Overview is an answer laid on top of a results page; the ads still have a page to sit on. The section's trap is the step after: an assistant or agent that completes the task and sends no one to any page. That layer is still small, so today's numbers can't test it. Kodak is the warning on reading the accounts: its revenue peaked in 1996, two decades after its own engineer built the digital camera. Healthy revenue is what a business looks like before the structure under it goes. The section is wrong if agentic and conversational use comes to monetize at search-like margins.
Argue it yourself ↓ a scoped prompt for any AI — bring your own disagreement

Broken yet?

The irony is noted. You may continue either way. Disagreeing opens a short form that asks why; skip it if you like (privacy).