Our Predictions for 2030, and How Our 2014 Calls Held Up

In February 2014 we wrote that .com would remain the standard web extension, that good names would become genuinely scarce, and that prices would climb into the millions. Hundreds of new extensions launched in the years after. .com held. Voice.com sold for $30 million and CarInsurance.com for $49.7 million, both in 2019.

Five months later we wrote that cars would be connected to the internet and that a single hub would run your television, appliances, heating, cooling and security, reachable from your phone anywhere. That is an unremarkable description of a house in 2026. It was not unremarkable in July 2014.

We also got something badly wrong, and it is the most useful thing we have to offer here. We predicted that hub would be a computer in your house, specifically so you would not have to put your data on someone else’s servers. The opposite happened. The hub became Amazon, Google and Apple, home servers stayed a hobby, and the dependency we wanted designed out became the default.

That is the pattern. The capability is the easy part to call. Who ends up owning it is where almost everyone is wrong, and the answer is almost always “whoever can rent it to you.” Every prediction below is made with that in mind.

1. The independent artists come for AI music, and win something

The majors have settled. Warner reached a deal with Suno reported at $500 million in November 2025 and Universal settled with Udio the month before, while Sony remains in litigation. Independent artists — who hold roughly 46.7% of the global recorded music market on an ownership basis — were not at the table, have no opt-in, and received nothing. A class action was filed on their behalf in June 2025.

The prediction: a collective licensing body for independent catalogues emerges and becomes the mechanism by which indie music gets paid for AI training, because individual litigation does not scale and the alternative is a permanent two-tier system.

What would prove us wrong: the class actions settle quietly for small sums with no structural change, and independents simply absorb the loss.

2. Search stops being how people find things to buy

Nearly half of secondhand shoppers already discover their next purchase through creators and social feeds rather than search, per ThredUp’s 2026 resale report. That number only moves in one direction.

The prediction: by 2030, search is a minority discovery channel for consumer purchases. And here is the ownership twist — the winner is not the social platforms. It is AI assistants. The practical consequence is that businesses stop optimising to be ranked and start optimising to be cited, which is a different craft with no established playbook.

We should be straight about the self-interest: this prediction runs against us. We are building a magazine that currently depends on search. We think it is true anyway, which is why we are also building an audience that does not arrive through a search engine.

What would prove us wrong: AI assistants keep sending traffic onward to publishers at scale rather than answering in place.

3. Secondhand becomes a tab, not a destination

The global secondhand market is projected to reach $393 billion by 2030, growing roughly twice as fast as apparel retail overall, with about 71% of that growth coming from Gen Z and Millennials — mostly people who have never shopped secondhand before.

The prediction: major retailers stop running resale as a side programme and fold used inventory into the main shopping experience as a filter beside size and colour. Buying used stops being a separate trip.

What would prove us wrong: margin pressure kills retailer resale programmes and the category stays with the specialists.

4. “Plant-based” gets a legal definition, and it is a story when it does

Right now there is no FDA definition of “plant-based” and none of “vegan” either. The main certification for plant-based alternatives permits up to 10% of a product’s formula to come from ingredients that are neither plant nor animal. The FDA has already issued draft guidance on naming and labelling plant-based dairy alternatives.

The prediction: the term gets regulated this decade, and the moment it does, the gap between what shoppers assumed and what the label permitted becomes a consumer story rather than a footnote.

What would prove us wrong: regulators leave it to industry self-certification indefinitely.

5. The AI hub will be rented, not owned — again

This is us applying our own 2014 mistake rather than repeating it.

The appealing prediction is that capable models run locally on your own device, privately, with no subscription and no company reading your data. It is the same instinct that made us predict the home server, and we think it fails the same way for the same reason: convenience and capability win, the frontier stays expensive, and the thing that is good enough to depend on will be the thing you rent.

The prediction: on-device AI becomes genuinely good and still does not displace cloud AI for mainstream consumers by 2030. Local models fill the private and offline edges, as home NAS did.

What would prove us wrong: a device generation ships local models good enough that most people stop paying monthly — which is precisely what we wanted to happen in 2014 and did not.

6. The computer becomes professional equipment

The prediction: the general-purpose computer narrows to a tool for people who produce things — developers, film and video editors, designers, engineers, anyone whose work needs a large screen and real compute. Everyone else stops owning one. Not because computers get worse, but because the phone absorbed every reason an ordinary household had for buying one.

This is already most of the way true and under-acknowledged. A generation is arriving at university having never owned a desktop.

What would prove us wrong: a consumer use case appears that genuinely needs local compute and pulls people back — running capable AI models on your own hardware being the obvious candidate, which is exactly the tension with our fifth prediction above.

7. The assistant moves to your wrist and your ear

The prediction: the primary way people use AI is through an assistant on a phone or a wearable, not a browser tab on a computer. The interface stops being something you go to and becomes something you talk to while doing something else.

What would prove us wrong: assistants stay bound to apps and screens because ambient voice never gets reliable enough in noisy, public, real-world conditions.

8. Brain interfaces move from control to augmentation

This one is closer than most people realise. Neuralink reported 21 participants enrolled in its trials worldwide as of late January, up from 12 the previous September, with no serious device-related adverse events reported. Participants move cursors, browse, play games and post to social media by thought.

Worth being precise about what that is: a medical trial for people with paralysis, not a consumer product. The implant restores control that illness or injury removed.

The prediction: the step that matters is from control to augmentation — running an AI model through the interface rather than simply operating a cursor with it. Musk has said as much, and the trial numbers are now compounding rather than creeping.

What would prove us wrong: regulators keep the approved use strictly therapeutic, and augmentation in healthy people stays unapprovable for a decade or more. That is the likeliest brake, and it is a regulatory question rather than an engineering one.

9. Billboards start reacting to whoever is standing there

The prediction: outdoor advertising becomes as targeted as online advertising — the board changes based on the audience in front of it, bought programmatically the way display ads are today.

Early versions already run, using anonymous audience measurement to estimate who is nearby rather than identifying individuals.

What would prove us wrong: privacy law stops it. This is the prediction most likely to be decided in a legislature rather than a market, because a camera pointed at a pavement is a far harder sell to the public than a cookie.

10. Diabetes gets managed from the watch, with one important caveat today

The caveat first, because this one can hurt someone. On 21 February 2024 the FDA issued a safety communication telling people not to use smartwatches or smart rings that claim to measure blood glucose without piercing the skin. No such device has been authorised to measure or estimate blood glucose on its own. Inaccurate readings can lead to the wrong insulin dose, and the FDA names confusion, coma and death among the possible consequences.

That is distinct from watches that display readings from an FDA-authorised continuous glucose monitor. Those are not the problem and are already common — the sensor does the measuring, the watch does the screen.

The prediction: the finger-prick test disappears from routine management, replaced by continuous sensing read from the wrist. Whether truly non-invasive optical sensing arrives, or the sensor simply becomes small and cheap enough to stop counting as a procedure, the daily experience converges on the same thing: you glance at your watch.

The prerequisite nobody mentions: battery. The debate fixates on whether optical sensing can read glucose through skin at all, and skips the fact that it would have to run continuously. Mainstream smartwatches are already daily-charge devices, and a sensor sampling around the clock is a constant draw rather than an occasional one. The reason a continuous glucose monitor manages ten to fourteen days on a tiny cell is that the sensor does very little and transmits in short bursts; a watch doing the measuring itself has no such luxury. Battery density improves slowly and unglamorously, which makes it the likeliest thing to set the timeline here — not the physics and not the regulator.

What would prove us wrong: non-invasive glucose sensing turns out to be a harder physics problem than a funding problem, and finger-sticks persist as the accuracy standard, or the power budget simply never closes. If anyone sells you one before it is authorised, that is a reason to be more suspicious rather than less.

11. AI enters policy making, and the disclosure fight follows

The prediction: governments use AI in drafting and modelling policy, openly. Impact assessment, forecasting, consultation analysis and legislative drafting are all document-heavy work that AI does cheaply.

The real fight will not be whether it happens but whether it is declared. Expect the first genuine scandal to be about undisclosed use rather than a bad outcome — a bill, a filing or a consultation response that nobody will admit to having drafted.

What would prove us wrong: procurement rules and political risk keep it confined to back-office analysis, and nothing touching actual legislative text.

12. Microplastics become the defining health issue of the next decade

The prediction: microplastics move from an environmental story to a human health story, and become the health concern that defines the 2030s the way smoking defined the second half of the twentieth century and sugar defined the 2010s.

The evidence has shifted fast. A prospective multicentre study published in the New England Journal of Medicine in 2024 examined carotid plaque removed from 257 patients and found polyethylene in 58.4% of them. Over a mean follow-up of about 34 months, the composite outcome of non-fatal heart attack, non-fatal stroke or death occurred in 20.0% of patients with plastics in their plaque versus 7.5% without — an adjusted hazard ratio of 4.53.

And the exposure is not exotic. A study by Columbia and Rutgers researchers in PNAS found an average of roughly 240,000 plastic particles per litre of bottled water, ranging from 110,000 to 400,000 across three brands.

The honest reading of that evidence. The NEJM authors state plainly that their results do not prove causality. It is an observational study, laboratory contamination could not be firmly ruled out, the population was asymptomatic surgical patients, and dietary, water and air-pollution exposures were not measured. A hazard ratio of 4.5 in a study with those limitations is a reason to pay close attention, not a finding to quote as settled.

That is exactly why we think this becomes the big one. The pattern of a plausible mechanism, ubiquitous exposure, and a strong early association is how a health issue graduates from fringe to mainstream — and the institutional position has already moved, from treating microplastics as a low-concern unknown to funding serious work on them.

What would prove us wrong: larger, better-controlled studies fail to reproduce the association, or the particles turn out to be inert passengers in plaque rather than participants in it. Attention would then fade the way it did for several earlier contamination scares.

One we are noting, not grading

Pulling drinking water out of the air works on the same principle as the dehumidifier in your basement, and it has quietly stopped being a thought experiment. The newer approach uses porous materials called metal-organic frameworks that capture moisture even where the humidity is too low for plain condensation to work. That chemistry won the 2025 Nobel Prize in Chemistry, shared by Omar Yaghi, Susumu Kitagawa and Richard Robson, and Yaghi’s company is now moving to commercial rollout in containerised units producing thousands of litres a day, tested as far down the humidity scale as Death Valley.

We are flagging it rather than making a call on it. It is industrial equipment today and nobody is drinking air-harvested water at home this year, but the gap between a basement appliance and a water source is narrower than it was. More on where this fits in treating water at home.

How to hold us to this

All twelve predictions above name what would falsify them, which is the part usually left out. We will grade these the same way we graded 2014: in public, with the original text unaltered, including the ones we get wrong.

The 2014 posts are still up if you want to check our working — Digital Toy Box and Dot Com Real Estate, both with their scorecards attached.

Predictions 10 and 12 touch on health topics. Nothing here is medical advice, and an association found in a study is not a proven cause. Decisions about glucose monitoring, insulin or any medical device belong with a clinician.