Ambi: The AI That Listens to You All Day

A critical assessment of Ambi, the $26M ambient AI assistant from a former Plaud executive — backed by China-focused Luminous Ventures and smartphone ODM Huaqin, built by a team half-based in Beijing, Shanghai, and Shenzhen, and shipping on a privacy promise that its own website quietly contradicts.

ProofStory Research October 12, 2026

$26M Raised, Lead: Luminous Ventures — October 12, 2026

Ambi (ambi.ai) builds an all-day ambient AI assistant for iPhone, Mac, and Apple Watch that listens, transcribes, and acts live during meetings. Co-founder & CEO Peter Mo was a China-side executive at Plaud, the AI-recorder maker that weathered its own data-security scrutiny in 2025.

$26M
Raised To Date
~50
Staff, 4 Cities
0
Model / Cloud Vendors Disclosed
1
Prior China-Data Controversy (Plaud)

Three Core Questions

01

“Where Is the Data Actually Processed?”

The site says transcripts are “encrypted and stored in the United States.” It never says where they are processed. Roughly half the company sits in Beijing, Shanghai, and Shenzhen — the exact structure that drew scrutiny at the CEO’s prior company, Plaud. Storage location does not resolve processing access.

02

“Do They Really Not Store Recordings?”

The CEO says Ambi “does not store users’ recordings.” True — of the audio. The company’s own site confirms it keeps the transcript: the searchable text of everything it overhears. “We delete the audio” is not “we forget what you said.”

03

“Is $26M a Real Round?”

The headline reads like a milestone raise. The substance is a running total across unnamed rounds, with no lead-round size, no valuation, and no filing or investor release corroborating it. Every number here traces to a single TechCrunch article.

Key Finding: Ambi is entering a validated category — Meta bought Limitless, Amazon bought Bee, Granola raised to a $1.5B valuation — with a credible, hardware-literate founder. But its privacy narrative is built on careful framing, and the structural question it has never addressed publicly is where a team half-based in mainland China processes a recording of your entire day. On the most privacy-sensitive product category that exists, that silence is the story.

The Numbers

Founded
Not disclosed (est. 2024–2025); San Francisco HQ
Founder
Peter Mo (co-founder & CEO), former partner & China CEO at Plaud
Funding
$26M total “across multiple rounds” (company-stated, single-sourced); no round or valuation named
Investors
Luminous Ventures (lead, China-focused); Huaqin (ODM, likely strategic/hardware partner)
Team
“Over 50” across San Francisco, Beijing, Shanghai, Shenzhen
Product
All-day ambient AI assistant for iPhone, Mac, Apple Watch; live in-meeting agent; desktop app + hardware device planned
Hardware
Screen, speaker, camera, microphones; waitlist-gated, “early next year”
Pricing
None published; FAQ cites “device, plan, region, and connected services” with no figures

What Happens to What You Say

Ambi’s privacy pitch rests on one line: “Your audio belongs to you. Once it’s transcribed, we delete it.” Trace the pipeline and the reassurance narrows.

The Ambient Capture Pipeline

01

Capture

Continuous ambient listening across phone, Mac, and watch — off by default, opt-in all-day mode.

→
02

Transcribe

Speech converted to text by “speech-recognition models” the company is still hiring to build.

→
03

Delete Audio

The raw recording is discarded. This is the step the “we don’t store recordings” claim describes.

→
04

Retain Transcript

“What remains is encrypted and stored in the United States” — the searchable text of everything overheard.

→
05

Inference

A live agent reasons over the transcript, calendar, and web — via an unnamed model and cloud, seen in plaintext at inference.

“We delete the audio” is not “we forget what you said.” The sensitive asset in an always-listening product is the transcript, and Ambi retains it. “Encrypted, stored in the US” also says nothing about which third-party model provider sees the plaintext at the moment of inference — a dependency the company has chosen not to name.

Stored in the US. Processed Where?

Ambi states where data is stored. It is silent on where data is processed — and on which foundation model and cloud it depends on. With dedicated engineering offices in Beijing, Shanghai, and Shenzhen, and PRC law (the Data Security Law and National Intelligence Law) that can compel data access from China-based infrastructure, storage location does not settle the question. This is the precise structure that forced Ambi’s predecessor, Plaud, to issue a six-point public defense in September 2025. Ambi has published no data-residency architecture and no government-access policy.

Peter Mo

CEO; was Plaud’s partner and China CEO — carries both the hardware playbook and the China-data history.

Luminous Ventures

Lead investor; China-focused fund spun from Lightspeed China, backer of ByteDance, Meituan, Unitree.

Huaqin

Top-two global smartphone ODM (Shanghai/Hong Kong listed). Its presence implies it builds Ambi’s planned device.

Transcript Retention

Audio deleted; encrypted transcript kept in the US. The sensitive content survives the “no recordings” claim.

Bystander Consent

CEO says the duty to disclose capture “falls on users” — pushing wiretap liability onto the customer.

Name Collision

“Ambi” clashes with Ambi Robotics (which also raised $26M, in 2021) — muddying trademark and search.

A Crowded, Already-Consolidating Field

The always-listening category has two proven exit paths and one venture breakout — and the two hardware plays both got absorbed by big tech, then wound down.

01

Granola — $192M

The category’s valuation leader at a $1.5B valuation (Series C, March 2026), now pushing from meeting notes into enterprise AI. Roughly 7× Ambi’s total raise.

02

Limitless & Bee — Acquired

Meta bought Limitless (ex-Rewind) in December 2025; Amazon bought Bee in July 2025. Both validated big-tech appetite — and both saw their always-listening hardware wound down post-deal.

03

Plaud — The Template

Peter Mo’s alma mater; hardware-first AI notetaker. The direct blueprint for Ambi — including the China-data controversy Ambi now inherits structurally.

04

Otter, Friend, Omi

Otter (~$70–80M) holds the enterprise transcription incumbency; Friend and Omi are sub-$10M pendant plays that became lightning rods for anti-AI backlash.

The read: timing cuts both ways. The category is real and moving — but the best-capitalized player is 7× Ambi’s raise and pivoting to enterprise, while the two prior always-listening hardware startups both exited to big tech and killed their devices. Ambi plans to ship that same hardware — a screen/camera/mic device — on a $26M cumulative base.

Weaknesses & Threat Vectors

Six structural risks the $26M does not resolve.

High

Geopolitical / Data Sovereignty

Half the team in Beijing/Shanghai/Shenzhen, a Chinese ODM and China-focused VC on the cap table, and an undisclosed processing location — on a product that captures a user’s entire day. Directly echoes the Plaud controversy the founder already weathered. Likely un-sellable to US enterprise or government without a hard data-isolation story Ambi has not provided.

High

Wiretap / Two-Party Consent

Always-on capture collides with all-party-consent regimes (California Penal Code §632, and peers in PA/WA). The CEO explicitly offloads disclosure duty to users. Notetaker peers already face litigation; a planned hidden-capable camera/mic device sharpens the exposure.

High

Privacy-Claim Credibility Gap

“We don’t store recordings” obscures retained transcripts; “encrypts data” with an unspecified method; no SOC 2, ISO, or HIPAA claim and no third-party audit. One investigative teardown could puncture the entire trust narrative.

Medium

Undisclosed Model / Cloud Dependency

No LLM vendor or cloud provider named anywhere. Margins, latency, plaintext exposure, and geopolitical risk all ride on third parties the company won’t reveal. No evidence of a proprietary model despite “building” speech recognition.

Medium

Hardware & Category-Absorption

A screen/speaker/camera/mic device “early next year” is a capital-intensive lift on $26M cumulative. Both prior always-listening hardware startups (Limitless, Bee) exited to big tech and then killed their devices.

Medium

Thin Capitalization & Name Collision

$26M total — not a fresh round — against Granola’s $192M and Meta/Amazon-backed rivals, with no disclosed runway or valuation. The “Ambi” name also collides with Ambi Robotics (also a $26M raise), muddying SEO and trademark.

Assessment Matrix

Technical Credibility
Medium
Ambitious live cross-device agent; founder has hardware pedigree — but no proprietary model shipped and core AI deps undisclosed
Privacy / Trust
Low
Misleading “no recordings” framing, unspecified encryption, consent offloaded to users, zero certifications
Moat / Defensibility
Low
Crowded field, better-funded incumbents, no disclosed proprietary tech; “always-listening” is commoditizing
Market Timing
Medium
Real momentum (Granola $1.5B, Meta/Amazon M&A) — but big tech may already own the oxygen
Geopolitical / Regulatory
High Risk
China engineering footprint + Chinese ODM/VC + undisclosed processing + always-on capture + wiretap law — the single most disqualifying dimension for Western adoption
Funding Transparency
Low
$26M is cumulative and single-sourced; no round size, no valuation, no filing or investor release
Investor Signal
Mixed
Luminous and Huaqin are real and serious — but reinforce, rather than offset, the China-exposure thesis

Ambi is a credible founder’s bet on a real, validated category — wrapped in a privacy story its own website undercuts. The product is plausible and the category is moving. But the question that decides whether a Western buyer can ever trust an app that records their entire day — where a team half-based in mainland China processes that transcript — is the one Ambi has never answered in public. Until it does, “encrypted and stored in the United States” is an answer to a different question.

Research Sources

Based entirely on publicly available information, including the TechCrunch announcement of October 12, 2026. Company-stated figures — the $26M total, the ~50 headcount, and the investor list — trace to a single primary source and are not independently verified by filings.

  1. TechCrunch — “Former Plaud exec raises $26M for AI app that listens to you in the background” (Ivan Mehta, October 12, 2026)
  2. Ambi website — ambi.ai, including the privacy section, FAQ, download, and terms pages (product claims, retention and encryption language, pricing absence)
  3. Secondary re-reporting — Mezha, TradersUnion, BigGo (all tracing to the TechCrunch source)
  4. 36Kr — identification of Peter Mo as Plaud’s partner and China CEO
  5. Plaud data-security controversy — DigiTimes (Sept 2025), Sally.io GDPR analysis, Sacra company profile
  6. Luminous Ventures — DealStreetAsia, VCSheet (China-focused fund; Lightspeed China lineage; portfolio)
  7. Huaqin (ODM) — GlobalVenturing, AVCJ, Caplight (top-two smartphone ODM; Intel Capital Series A; dual listing)
  8. Always-listening / wiretap law — MacDailyNews (Sept 2026), Oklahoma Law Review, practitioner commentary on §632 all-party consent
  9. Competitor set — CNBC (Meta/Limitless), TechCrunch (Amazon/Bee; Granola $125M at $1.5B), Omi, SF Standard (Friend), PitchBook (Otter)
  10. Brand-collision reference — Ambi Robotics $26M Series A (2021), an unrelated entity frequently conflated in databases
Work With ProofStory

We Bring This Rigor to Your Business

We build and maintain websites, help businesses of every size put AI to work, and run growth marketing that’s held to the same standard as our research.