The great unbundling: one AI app is eating your ten SaaS subscriptions

Nobody decides to spend $1,800 a year on AI. It happens $20 at a time. A chat subscription in 2023, because everyone needed one. An image tool in 2024, because the chat app’s pictures weren’t good enough. Then a search tool, a voice tool, a video tool, a music tool, a meeting-notes tool, and a second chat subscription because the first one is worse at code. Each purchase was defensible. The pile is not.

This post makes one argument: the subscription-per-task era of AI is ending, the way it ended for every software category before it. The capabilities you’re renting ten interfaces for have collapsed into a handful of models that are all reachable by API, priced by the drop. The shape that replaces the pile is one multi-modal surface plus your own keys, with a short list of specialist tools that genuinely earn their line on the statement. The math, the history, and the honest exceptions follow.

The ten-tab stack is real, and it bills $151 a month

Price a working 2026 stack for one person who makes things: a marketer, a founder, a creator. Chat and writing on ChatGPT Plus at $20, code and long documents on Claude Pro at $20, research on Perplexity Pro at $20, images on Midjourney Standard at $30, voice on ElevenLabs Creator at $22, meeting notes on Otter Pro at $16.99, video on Runway Standard at $12, music on Suno Pro at $10. That’s $151 a month, $1,812 a year, and it’s a conservative build: Runway is on its cheaper annual rate, and there’s no coding assistant or scheduling tool yet. Add the two of those most professionals carry and the invoice reaches its full ten lines.

The dollars are the smaller half of the bill. Eight products means eight logins, eight histories, and eight places your context goes to die. The research you did in the search tool can’t see the draft sitting in the chat tool. The image tool doesn’t know what the script says. You are the integration layer, ferrying context between tabs by clipboard, and that tax is charged in attention rather than dollars. I’ve written before about what the auto-renew pile does to a month; AI is simply the newest and fastest-growing wing of it.

A subscription is prepaid capacity, and most of it expires

Look at what those flat fees actually buy. Midjourney’s $30 Standard plan is, mechanically, 15 hours of fast GPU time a month. Suno’s Pro plan is 2,500 credits that, per its own pricing page, refresh monthly and don’t roll over. ElevenLabs’ Creator tier is a monthly pool of 121,000 credits, spent per character of speech. A subscription is not access to a tool. It is a prepaid block of capacity with a 30-day expiry date.

Prepaid capacity is a fine deal for steady usage and a quietly bad one for how creative work actually arrives. Real months are spiky: a launch week that burns your image quota in three days, then six weeks where the tab never gets opened. The flat fee is sized so the vendor wins in the quiet weeks, and vendors know how reliably those quiet weeks come. Meanwhile the meter on the same capability reads in cents. You’d never notice the gap from inside one $22 subscription. You notice it when there are eight of them, and every one is holding expired credits you paid for.

Run the arithmetic on a quiet month and the gap stops being abstract. Generate twelve images on a $30 Midjourney month and each one cost you $2.50. The same image, metered through an API, runs about $0.039: a 64x difference, invisible because the subscription never shows you a per-unit price. Flat fees don’t just smooth your spending. They hide the number that would tell you to leave.

Every tool in the stack became an API call

The reason consolidation is possible now, and wasn’t in 2023, is structural. The capabilities stopped being products and became model endpoints. Image generation is a line item on Google’s API price list: $0.039 per image on Gemini 2.5 Flash Image. Frontier-quality chat is $2.50 per million input tokens on GPT-5.6 Terra, with the flagship Sol at $5. Transcription, translation, document understanding, and speech all sit on the same price lists, metered in fractions of a cent.

That changes what a single-purpose AI subscription is. In 2023, a specialized tool often wrapped a genuinely different model. In 2026, much of the stack is a veneer: the same handful of frontier models underneath, with a UI, a quota, and a markup on top. Sometimes the veneer is thick and valuable, a real workflow built by people who understand the job. Sometimes it is a prompt template and a Stripe integration. Either way, the capability itself no longer lives in the tool you’re subscribing to. It lives one API call away, priced by the drop, and any surface that can hold your keys can reach it.

The single-purpose tool is being squeezed from above, too. The big chat apps have quietly become bundles themselves: a ChatGPT Plus subscription now includes image generation, voice conversation, and file analysis in the base $20, and Gemini and Claude are converging on the same all-in-one shape. When the generalist app absorbs a capability, the standalone subscription for that capability has to justify itself on workflow alone. Some can. Most of the stack can’t, and the squeeze only tightens as the frontier models go natively multimodal: one model that sees, hears, and speaks replaces three products that each did one of those.

Software has seen this movie: bundle, unbundle, repeat

In August 1995, at the last stop of Netscape’s IPO roadshow, a banker asked Jim Barksdale how he’d survive Microsoft bundling a free browser into Windows. His answer became the most quoted line in software strategy: “there’s only two ways I know of to make money: bundling and unbundling.” The cycle he named has run through every layer of media and software since: the album unbundled into the track, the track re-bundled into the playlist; the newspaper unbundled into feeds, the feeds re-bundled into aggregators.

Streaming is the rerun most of us lived through. Cable was the hated bundle; à la carte apps were the liberation. Then the apps multiplied, and Deloitte’s 2026 Digital Media Trends survey finds the average subscribing household paying $69 a month for streaming video, churning through services at roughly 40% a year, with 61% saying a $5 price bump would make them cancel a favorite. The liberation quietly rebuilt the cable bill, minus the single login.

The AI stack is the same tape at triple speed. It took television about a decade to go from one bundle to an exhausting pile. AI went from “one ChatGPT subscription” to the eight-line invoice above in three years. Which means the next turn of the cycle, the consolidation, is not a prediction so much as a schedule. Barksdale would shrug: the tasks are unbundling from ten subscriptions, and re-bundling into one surface. Both moves at once, and the money moves with them.

One surface plus your own keys re-prices the month

Here is the consolidated version of the stack, priced honestly. You keep one multi-modal app as the surface: chat, image generation, document work, and transcription in a single place with a single history. Underneath, it spends your own API keys, so every job runs on the meter instead of a flat fee.

Sixteen dollars and twenty cents, against $151, for the work that actually fills an AI-heavy month: conversation, research, drafting, image runs, and long-document reading. It isn’t a rigged comparison so much as an unfamiliar one; when I itemized a maxed-out mixed hour of AI use, the entire hour, video clip included, cost less than one month of one mid-tier subscription. Metered pricing feels risky because a meter can theoretically run forever. In practice, text and image APIs are so cheap that the failure mode of the subscription (paying for capacity you didn’t use) costs more than the failure mode of the meter (a heavy month).

The keys matter as much as the price. Bring-your-own-key means the model relationship is yours: switch providers the day a better or cheaper model ships, keep your history when a tool dies, and route each job to the model that suits it instead of the one your vendor resells. The full BYOK-versus-SaaS accounting is its own post, but the short version stands: the subscription stack rents you ten shallow relationships with the same few models; your own keys give you one deep one. And for the private end of the spectrum, open-weight models running on your own machine take the marginal cost of the everyday errands to zero.

Best-of-breed still wins the 20% that is craft

Now the honest part. Consolidation wins the errands. It does not win the crafts, and pretending otherwise is how essays like this one become sales pages.

If you generate images for hours a day, Midjourney’s control surface, consistency tools, and community are the product, not the pixels, and a heavy user beats the meter on a flat plan anyway. If voice is your product, ElevenLabs’ direction and cloning workflow is worth its subscription. Serious video work lives inside Runway’s editing timeline, not in a prompt box; music makers get commercial licensing and a real studio from Suno’s paid tiers. Specialist tools survive where the workflow, not the model, is the moat. That’s a real moat, and roughly 20% of the stack sits safely behind it.

The test for each line on your statement is whether the tool is craft or errand for you specifically. The same product can be either: Midjourney is craft for the brand designer and an errand for the founder who needs twelve blog images a month. Subscriptions are for crafts. Meters are for errands. The stack got expensive because the default put everything, including the errands, on the craft pricing.

Unbundle the errands, keep the craft

The practical version takes one evening. Pull the statement and list every AI line item with its annual cost, not its monthly one; $22 a month reads as harmless, $264 a year reads as a decision. Mark each line craft or errand. Then move the errands to one surface with your own keys behind it, and let each craft tool re-earn its seat with a month of real use. Most people end this exercise with one surface, one or two specialist subscriptions they genuinely work in, and a bill somewhere near a third of where it started.

The deeper shift is in what you stop owing. Not just the hundred-odd dollars a month, but the eight logins, the eight histories, the quiet dread of the renewal email, the expired credits. The streaming pile taught everyone how this feels; there’s no reason to sit through the rerun with tools you use for work. Barksdale’s line was about vendors, but it reads just as well from the customer’s side of the invoice: there are only two ways to save money, too. Bundle the errands. Unbundle everything else.