Wabi valuation
$1.25B
Oct. 29, 2026; $55M round led by IVP
Re-rating since February
2.5x
from about $500M, per TechCrunch's Oct. 29 report
Pivot to price
31 days
Wabi 2.0 announced Sept. 28; round closed Oct. 29, 2026
Wabi revenue to date
Not disclosed
free product, no business model settled as of Oct. 19, 2026
The Deal
Two and a half times in eight months, and a full re-rating in 31 days
Wabi announced on Oct. 29 that it had raised $55 million at a $1.25 billion post-money valuation, led by IVP, with Andreessen, Forerunner and Amazon's Alexa Fund on the round and angels including Shopify CEO Tobi Lütke, Marc Benioff, Marissa Mayer, Dick Costolo and Alex Pall. That is roughly 2.5 times the valuation TechCrunch reported for the company in February, and it came 31 days after founder Eugenia Kuyda abandoned the product that had defined the company. Nothing about Wabi's revenue changed in those 31 days: it is still free, still has no business model, and has disclosed no user count.
| Company | Date | Raised | Valuation | Lead investor |
|---|---|---|---|---|
| Wabi | Nov. 5, 2025 | $20M | Not disclosed | Andreessen |
| Wabi | Feb. 2026 | Not disclosed | ~$500M | Not disclosed |
| Instinct | Aug. 26, 2026 | Not disclosed | $2.5B | Not disclosed |
| Instinct | Sept. 28, 2026 | $1B | $10B | Sequoia, Benchmark, Coatue |
| Wabi | Oct. 29, 2026 | $55M | $1.25B | IVP |
Read the sequence, not the headline. Instinct quadrupled from a $2.5 billion valuation in late August to $10 billion on Sept. 28, a month after its invite-only launch. Wabi moved from about $500 million to $1.25 billion in eight months, and the entire re-rating landed after it announced on Sept. 28 that it was becoming an AI messenger. OpenAI shipped its own always-on assistant, Dots, on Sept. 29, one day after Wabi's pivot. Private capital repriced the consumer agent category in under three weeks, and not one of these companies sells a product with disclosed revenue.
The Product
Wabi is buying a surface, not building a business
Wabi 2.0, announced Sept. 28, is a messenger where the agent builds the interface on demand: meal plans that become grocery lists, restaurant shortlists, date-night plans, fitness plans, trivia games for a book club. The wedge is shared use — couples, families and friend groups coordinating inside one thread — rather than a single-user chat box. On Oct. 19 the company opened it to all users, free, and confirmed it has not settled on a business model.
- The wedge is the group, not the individual: one shared thread per household, book club or friend group, with the agent holding context across people rather than across one person's chat history.
- The mini-apps became the output rather than the destination — the interface is generated on demand inside the conversation instead of built, downloaded and maintained.
- Pricing is deferred: the service is free today, and the terms describe recurring subscriptions sold through Apple in-app purchase, with liability for AI features capped at $100.
- No user count, no revenue and no pricing have been disclosed. The company says it does not train models on user conversations, and its privacy notice extends that promise to the model providers processing requests.
The Supply Chain
The last layer rents its brain, its traffic and its billing
Wabi's privacy notice is more revealing than its funding release. It names Google, Anthropic and OpenAI as the model providers that process user content and connected-account data to generate responses and carry out tasks, with provider-side retention typically up to 30 days for abuse monitoring. Hosting sits with Cloudflare, live sessions with LiveKit, and payments and financial-account connections with providers including Apple.
| Layer | Named provider | What it means for the thesis |
|---|---|---|
| Intelligence | Google, Anthropic and OpenAI | Model providers bill every inference; the agent keeps whatever is left over |
| Infrastructure | Cloudflare, LiveKit | Rent is usage-based, so it rises with usage rather than with revenue |
| Acquisition and measurement | Meta advertising pixel, conversions API and custom audiences; TikTok pixel and events API; Snap pixel, conversions API and audience match | The ad market is still paid on the way in, so the agent does not bypass the gatekeeper |
| Billing | Apple in-app purchase | The store collects the subscription, not the agent |
| Browser execution | Third-party browser-automation infrastructure | Every completed task adds metered inference plus a hosted session |
That table contradicts the industry line. The argument for a personal agent is that it becomes the app and therefore escapes app-store economics and the advertising gatekeepers. Wabi's own filing says the opposite: it buys users through the pixels of Meta, TikTok and Snap, measures them through the same pipes, and routes any future subscription through Apple's billing. The agent owns the conversation and the memory, and everything on either side of it is a tollbooth.
The Benchmark
The free agent took 10.2 million US downloads in 40 days
US downloads in the first 40 days after launch (millions)
Muse launched Sept. 8, 2026. Sensor Tower data reported by TechCrunch, Oct. 20, 2026.
Unit: million US downloads
Meta Muse
10.9 million downloads globally in 40 days
10.2
ChatGPT
US downloads in its own first 40 days
3.7
Grok
US downloads in its own first 40 days
0.5
Claude
US downloads in its own first 40 days
0.2
Meta launched Muse on Sept. 8 and, on Sensor Tower numbers reported by TechCrunch, it reached 10.9 million downloads in 40 days, 10.2 million of them in the United States, against 3.7 million US downloads for ChatGPT over its own first 40 days. Retention was better too: 30% of Muse users were still active on day seven and 23% on day 14, versus 19% and 16% averages for ChatGPT, Claude and Grok. On the Q3 earnings call, Meta said more than 10 million people in the US and Canada had set up a Muse agent in about a month and a half, with shopping the primary use case. Meta owns the distribution that Wabi is buying.
What Muse actually does matters more than the download count. Meta's launch describes an agent that keeps working after the app is closed, returns when something needs approval, sends email, books travel, opens a browser, fills in forms and negotiates on a user's behalf, connected to the apps a user already has — free for most of what people need, with subscription plans for more. That is Wabi's product description with the price removed and the distribution attached.
- Price: a free-for-now product has to beat a free incumbent that already owns the social graph, the permission layer and the daily habit.
- Reach: 10.2 million US downloads in 40 days buys a scale that no seed-stage round can buy, so the standalone agent competes for a slice of the relationship rather than all of it.
- Wedge: the shared, multi-party context is the one place a platform agent is weakest, because a consumer assistant holds one person's context rather than a group's. Wabi's bet is that group context is worth paying for.
- Contest: Instinct put its agent into group chats on Oct. 5, including for friends without accounts, weeks after a $10 billion round, so the wedge is already crowded.
The Bill
The agent war is being paid for out of platform cash flow
| Company | Revenue, latest quarter | Capital spending | Free cash flow |
|---|---|---|---|
| Meta Platforms | $64.6B (Q3 2026) | $104.4B | $28.7B |
| Alphabet | $127.6B (Q3 2026) | $159.8B | $19.7B |
| Amazon | $201.9B (Q3 2026) | $215.2B | -$46.9B |
These are the balance sheets underwriting a category whose leading consumer product is free and whose leading startups have no revenue. Meta's Q3 filing guides 2026 capital expenditure to $135–145 billion, including $85.82 billion spent in the first nine months, and carries $454.52 billion of non-cancelable commitments plus roughly $355.73 billion of leases not yet commenced for data centers and colocations. Quarterly free cash flow was negative $2.12 billion and operating margin was 20%, with legal-related costs and infrastructure spending both named as drags on income from operations.
Amazon is the starkest case: $215.2 billion of trailing capital spending against negative $46.9 billion of free cash flow. Alphabet spent $159.8 billion to generate $19.7 billion of free cash flow. Meanwhile NVIDIA's revenue rose 106% year over year to $96.22 billion in the quarter ended July 26, 2026, at a 75.0% gross margin. The agent war is being financed by balance sheets, and it is being paid for, in the end, by the chipmaker.
NVIDIA quarterly revenue ($B)
Revenue by fiscal quarter, from NVIDIA's quarterly filings through the quarter ended July 26, 2026.
Unit: $B
Q1 FY2026
44.1
Q2 FY2026
46.7
Q3 FY2026
57
Q4 FY2026
68.1
Q1 FY2027
81.6
Q2 FY2027
96.2
Horizons
What to watch over the next year, and what to watch over three
- Wabi's first pricing announcement: the first subscription tier converts the round from a narrative into an economy, and until then it is financing a waiting room.
- Muse's day-30 retention: the 30% day-seven number is the one that matters, because an agent that peaks on download and hollows out afterwards is worth nothing to an advertiser or a store.
- Meta's capex against cash generation: $135–145 billion of 2026 spending and negative quarterly free cash flow force a monetization decision inside four quarters.
- Instinct's first disclosed revenue: a $10 billion mark on a free, invite-only agent is a distribution claim until somebody books a number.
- Inference price per completed task: with three model providers metering Wabi's requests, the gross margin of the whole category hinges on their pricing.
- Surface concentration: if the agent becomes the app, the winners own the default surfaces — the phone, the browser, the social graph — and the losers are the app-store commission pool and the advertising inventory behind it.
- Merchant rails: agents that plan and buy need catalogs, carts and checkout, which is why commerce platforms are building agentic rails and why Shopify's CEO is among Wabi's backers.
- Rent inflation: inference prices, ad costs and store fees are the three line items a standalone agent cannot control, and each is set by a platform with more leverage than the startup.
- What would break this thesis: a consumer agent disclosing revenue per user, or Meta charging for Muse and still holding retention. Either would move value from the platforms back to the agent layer.
Conclusion
Own the surface and the silicon, not the agent
Wabi is the cleanest available read on where private capital thinks the consumer agent market goes: it is not paying for a business, it is paying for the right to be the default surface for a group of people. The same logic produced Instinct's $10 billion mark, and a free Meta agent with 10.2 million US downloads in 40 days. So the durable profits sit with the surfaces that own distribution and with the silicon underneath them, while standalone agents are options that get repriced upward when a surface is proven and downward the moment inference, ad or store fees bite. In a category where the leader is free, the agent is the product, and the margin belongs to somebody else.
Where the rent lands
- Meta's advertising pixel, conversions API and custom audiences appear in Wabi's privacy notice, so every consumer agent app buys its users from Meta.
- Muse took 10.2 million US downloads in 40 days with 30% day-seven retention versus 19% for ChatGPT, Claude and Grok.
- Q3 2026 revenue of $64.6 billion and a 20% operating margin show what a free consumer agent costs.
- Next 12 months: capex of $135–145 billion against negative quarterly free cash flow forces a monetization decision.
- Google is named in Wabi's privacy notice as a model provider, so agent inference is billed to Alphabet's books.
- Google Play carries Wabi's Android distribution and Firebase its analytics, so the default surface is already in the model provider's hands.
- Q3 2026 revenue of $127.6 billion against $159.8 billion of trailing capital spending leaves only $19.7 billion of free cash flow.
- Over 1-3 years the risk is disintermediation: agents that book and buy hit APIs directly, not paid search results.
- Wabi's terms route iOS subscriptions through Apple in-app purchase, so the store collects the recurring revenue, not the agent.
- Revised EU terms cut the in-app purchase commission to 26% from Oct. 1, 2026, with take-rate pressure arriving alongside agents.
- Q3 FY2026 revenue of $109.4 billion including a record $30.7 billion of Services funds Apple's own assistant build-out.
- Over 1-3 years the exposure is discovery: agent-built interfaces bypass App Store search, where Apple's discovery economics live.
- Quarterly revenue of $96.22 billion, up 106% year over year at a 75.0% gross margin: agent competition is a commodity sold to all participants.
- Every model call from Wabi's three named providers, plus its hosted browser sessions, lands on the same silicon.
- Trailing revenue of $303.0 billion and trailing net income of $192.9 billion make inference the one line that grows with each agent.
- Next 12 months: the watch item is whether hyperscaler capex keeps converting into accelerator orders.
- Shopify's CEO personally backed Wabi, and its Sidekick agent already runs merchant-side workflows, so the commerce endpoint is where agent demand lands.
- Q2 2026 revenue of $3.58 billion, up 34%, on GMV of $115.6 billion, with $654 million of free cash flow.
- Agents that plan and buy need catalogs, carts and checkout, which turns merchant APIs into the tollbooth for agentic commerce.
- Over 1-3 years: if agents transact, Shopify monetizes the transaction while Meta and Alphabet own the conversation.
- Amazon's Alexa Fund backed Wabi, an implicit admission that Amazon does not own a default consumer surface of its own.
- Trailing capital spending of $215.2 billion against negative $46.9 billion of free cash flow makes the AI build self-funded.
- Q3 2026 revenue of $201.9 billion shows scale, but scale is the wrong metric where the category leader is free.
- Over 1-3 years the upside is agent-led commerce through Amazon's retail APIs; the risk is Alexa's irrelevance.
- Snap's pixel, conversions API and audience match appear in Wabi's privacy notice, so Snap is paid for agent land-grabs it does not own.
- Q2 2026 revenue of $1.60 billion came with a $164 million net loss and a $171 million operating loss.
- Meta's free Muse took 10.2 million US downloads in 40 days in the messaging surface Snap wants to own.
- Next 12 months: if agents absorb messaging, Snap keeps the measurement revenue but loses the relationship.
