The bundling apocalypse — why most standalone AI tools will be dead by 2027
Midjourney fell from the top 10 to #46. OpenAI shut down Sora after burning $15M a day for $2.1M in lifetime revenue. The map of which standalone AI tools survive the bundling pressure, and which categories get absorbed.
In September 2023, when a16z published the first edition of its Top 100 Gen AI Consumer Apps report, Midjourney sat in the top ten. It was the iconic AI product. Discord-only. Bootstrapped. The first thing most people pointed to when they wanted to show a non-technical friend what generative AI could actually do. By January 2026, Midjourney was at #46.
Midjourney didn't get worse over those three years. By most accounts the product is better now than it was at peak, and the team has continued shipping confident updates throughout. What happened to it is what tends to happen to almost every category of standalone AI eventually. ChatGPT and Gemini both built decent image generation directly into their default chat interfaces, and a lot of the people who had been paying for Midjourney to make Pixar-style portraits of their cat decided they didn't really need a separate subscription for that anymore.
This is the bundling pattern in action, and image generation was just the first category to face it. Three more categories of standalone AI tool are due to face the same compression over the next eighteen months or so. For builders working in those categories the runway is shorter than most people realise. For anyone trying to track or bet on which AI tools will still matter in 2027, the survival map looks fairly different from the leaderboards everyone is currently optimising against.
The cleanest picture of the bundling effect is in a16z's sixth-edition report, published in March 2026 using SimilarWeb data from January and Sensor Tower data for mobile. The headline number got most of the attention. ChatGPT now has 900 million weekly active users, up by 500 million in a single year. More than 10% of the global population uses it weekly.
The follow-on numbers are more interesting. ChatGPT is 2.7× larger than the second-place product on web traffic, and 2.5× larger on mobile MAU. There hasn't been a category in modern consumer software where one product has dominated this hard since the early days of mobile, and unlike search, which had network effects on its side, this is happening because every query teaches ChatGPT something about the user. Memory and context compound. The longer someone uses one assistant, the higher their cost of switching. Sam Altman has hinted at a "Sign in with ChatGPT" identity layer; if it ships, the moat goes from deep to structural.
Underneath the headlines, the more telling signal is in the creative-tools category. Three years ago, seven of the nine creative slots in the top 50 were image generators. Now there are three. The slots that opened up didn't disappear; they got refilled by video, music, and voice products, all in categories where the bundled solutions inside ChatGPT and Gemini still aren't quite good enough to absorb the use case.
The threshold seems to sit somewhere around the point where a bundled version is good enough for everyday non-specialist use. Past that point, only the specialists with strong workflow opinions or community moats keep their seat at the table. Notion's paid AI attach rate climbing from 20% to over 50% in twelve months, with AI features now accounting for roughly half of total revenue, is the same dynamic told from the other side of the bundle. So is CapCut having 736 million monthly mobile users on an editor whose top features are AI background removal, auto-captions, and text-to-video. Canva built its entire growth engine around its Magic Suite. The old distinction between "AI tools" and "tools with AI features" has effectively collapsed, with the second group capturing most of the active usage.
If you wanted a clean test of whether a category was about to be bundled away, OpenAI's own video product was the experiment.
Sora launched and immediately rocketed. Number one on the US App Store for twenty consecutive days. Faster to a million users than ChatGPT itself. A planned Disney partnership covering more than 200 characters from the Marvel, Pixar, and Star Wars catalogues was announced in December 2025. Everything looked like it was working.
On March 24, 2026, OpenAI announced the shutdown. The web and app versions went dark on April 26. The API will follow on September 24. The economics, once they leaked, explained everything.
| Job | Default pick | Why |
|---|---|---|
| Daily cost (peak) | ≈ $15 million | Each ten-second clip cost roughly $1.30 in compute. Forbes / Cantor Fitzgerald analyst estimate, March 2026. |
| Lifetime revenue | ≈ $2.1 million | Total in-app purchase revenue from launch through shutdown announcement, per Appfigures. |
| Burn-down ratio | Lifetime revenue every 3.4 hours | Even using WSJ's more conservative $1M/day estimate, Sora burned its lifetime revenue every two days. |
OpenAI itself couldn't make a standalone consumer video tool work. The company with the best distribution, the most capital, and the closest relationship with the underlying compute layer concluded the GPUs were better spent inside ChatGPT.
If OpenAI couldn't make that work for itself, the implication for everyone else operating with less capital and less distribution should be reasonably clear. Most consumer AI products launching today with a flashy demo and a $20-a-month price tag are solving the same equation Sora was solving, often without showing both sides of the math at the same time when they pitch.
Walking through the consumer AI map with the bundling lens, four categories of standalone tool look most at risk through 2027.
Four categories under bundling pressure
Where the absorption lands hardest in the next eighteen months — and what the bundles already ate.
Any product whose pitch is 'ChatGPT but for [vertical]' without owned distribution or proprietary data is in the most direct line of fire. The window for general-purpose-chatbot-but-specialised closed in late 2025.
Sora is one data point. Kling, Veo, and Hailuo are consolidating the high end. Anything that isn't workflow-deep like Runway or quality-supreme like Veo will hit the same compute economics that killed Sora.
ElevenLabs survives because of dubbing and pro-audio workflows. The tier below it (point-and-click 'type text get speech' tools) is being absorbed by ChatGPT voice mode and Gemini speech in real time.
Pure autocomplete is losing to integrated coding agents. Cursor, Claude Code, and the Devin family now do code review, agent execution, and full-loop development. Standalone copilots have nowhere to retreat to.
The standalone AI tools that aren't just surviving but accelerating in 2026 tend to share three structural features. One of these is usually enough to keep a product alive through the bundling pressure. Two of them together starts to look like a real moat.
Suno held its rank because it doesn't just generate music — it owns the loop from idea to lyric to mixed track to share, tuned to a specific creative culture. Examples: Suno, Runway, Cursor.
Civitai survives because the model marketplace is the product, not the foundation model under it. Hugging Face works at developer scale. Character.AI works because user-created characters are the moat. Examples: Civitai, Hugging Face, Character.AI.
Manus, Lovable, OpenClaw, and the wave of horizontal agents from late 2025 onward sit in a category the foundation-model companies are structurally slower to build. Independents move faster on agency.
Anyone running a leaderboard, an aggregator, or a tool directory has a measurement problem getting worse before it gets better. As a16z themselves noted, web traffic increasingly undercounts AI usage. A developer spending eight hours a day in Claude Code is invisible to SimilarWeb. A knowledge worker dictating every email through Wispr is invisible to app store rankings. The most-used AI products in 2027 may not show up on a leaderboard built around web sessions and downloads, because they will live inside other products.
The metrics that matter for the next phase look different. Real-time usage signals from API aggregators. Attach rates inside bundled products. Retention curves rather than launch spikes. Migration patterns between tools.
The graveyard at the bottom of the a16z list, where the companies that were in the top 50 in 2023 sit largely forgotten, is going to grow before it shrinks. That isn't really a story about AI failing, though. It's a story about the parts of AI that worked being absorbed into the few products with the distribution to absorb them, and almost everything else getting compressed against that wall. The bundling itself isn't really the question to be asking anymore. What matters is which part of the survival map you decide to position on while the absorption is still finishing.
Track who's surviving on AgentTape
Stars get gamed, leaderboards get embedded out of, and bundles get thicker every quarter. AgentTape tracks AI tools on usage flow and sentiment — the signals that survive the absorption. Watch which standalone tools are gaining momentum this week, and which are being quietly compressed.
View the live indexes