Axon is trying to own the whole path from 911 call to courtroom.

The read

1. The product is the connection between products. Speakers describe two loops: after-the-fact "capture to courtroom" and real-time 911, crime center and drone response. Kunins, talent webinar The 2026 pitch joins them as "call to closure". Smith, Axon Week 2026
2. The 911 purchases are the biggest strategic move in the set. Axon spent tens of millions on its own dispatch software, got two customers live and shut it down. It then bought Prepared (about $630M, per Isner) and Carbyne. A host puts the pair at roughly $1.5B; Isner did not confirm that in the transcript.
3. AI is priced, not given away. The AI Era Plan booked over $750M in 2025, its first full year. Isner, May 2026 Draft One is the proven use case. The other AI skills have usage figures but little outcome data.
4. The status quo is still the main competitor in the core. Named rivals matter at the edges: Motorola in cameras and 911, Flock in license plate readers, Anduril in military counter-drone.
5. The growth case moves to buyers who behave differently. Isner says enterprise should eventually be the largest business and that Axon "will have failed" if international does not pass US public safety. Both lack the chief-to-chief network effect that made US sales fast.
6. Trust is the binding constraint, and Axon says so. Face matching, plate readers, Ring and Citizen footage and armed drones all need political consent. if we lose the trust, you lose the tech Smith, Axon Week 2026
7. The core product still has a known weakness. Smith puts TASER at "only about 80% reliable", mainly from heavy clothing and missed darts. The company's goal of replacing the bullet depends on fixing that with the Apollo cartridge.

How to read this

Product map: call to closure

Axon's own framing, laid out in workflow order. Select a product to jump to its entry.

1. The call: 911

The workflow Axon wants to compress runs caller, call taker, CAD, dispatcher, radio, officer. Smith compares the technology to a rotary phone, and a host relays his "game of telephone" description. US figures quoted across the set: about 240 to 250 million calls a year, about 6,000 call centers, about 72,000 responder agencies, 80 to 90% of calls from mobiles, and call takers facing about eight screens.

Prepared

The AI layer that sits on top of whatever call system a center already runs.

WORKS

Live transcription and summary, translation in 30+ languages, CAD entry drafting, radio transcription with keyword alerts, QA on every call against center protocols, and an autonomous voice agent for non-emergency lines.

GROWTH

Started as a school safety app at Yale. Pivoted to a free browser link for caller video and text, gave it to roughly the first 1,000 centers, then sold translation, then AI modules. Chime explicitly treats procurement as the thing to route around.

NUMBERS

1,000+ agencies, 49 states. QA coverage goes from 2 to 3% of calls manually to 100%. Non-emergency calls are 40 to 60% of volume in some big cities; one California agency averaged 40-minute holds. Fort Worth's agent handles thousands of calls a month; agencies can offload "more than a third" of volume. Call counts vary widely, see disagreements.

LIMITS

Chime says the agent is not meant for real emergencies, where AI stays a co-pilot. His framing is This is not person versus agent, it is "agent versus nothing".

DEAL

About $630M, Isner correcting a host's $800M figure. Closed around autumn 2025.

Carbyne

Cloud-native call handling, the infrastructure replacement to Prepared's overlay.

WORKS

Apex replaces on-premise servers. Universe added location, video and chat over existing systems, and Carbyne has stopped investing in it in the US because Prepared's overlay is stronger. Also Call Triage (AI prompts callers near a known incident), two-way translation, and Admin Assist (agentic non-emergency line, launched at APCO 2025).

NUMBERS

A 2026 host credits it with about 400M people and 150M calls a year. Earlier counts are inconsistent. Hundreds of agencies including Denver, Cook County, Atlanta and Miami-Dade. Sales cycle about six months via AT&T, Tyler and Cox as procurement vehicles.

RIVALS

Named: Motorola, Vesta, Solacom, RapidDeploy, Mark43. Elichai says the runner-up is "not even close in references".

COST

Legacy vendors take a one-time payment plus about 17% annual maintenance, per Elichai. He prices language lines at about $4 a minute.

Axon 911

Announced at Axon Week 2026. Rick Smith's plan: go wide and fast with Prepared, then upgrade customers to Carbyne. Chime's metaphor is that AI is a Ferrari and the cloud is the highway it needs. The stated end state is call handling, logging, eventually CAD and radio consoles, with eight screens becoming "a couple".

NUMBERS

1,000+ agencies and 80M+ calls a year. In Winston-Salem, Prepared plus Skydio cut time to eyes on scene from "two minutes at best" to 5 to 10 seconds. One major city saw operator-handled calls fall 33% on a holiday weekend.

WATCH

Radio and CAD are still ahead. Chime's own failed implementations came when centers were mid-change on CAD or radio, and he warns AI could produce 15 to 20 screens if it isn't integrated.

2. Seeing: real-time video and vehicles

Fusus

The map and video layer behind real-time crime centers. Vendor-neutral by design; speakers call it the "Switzerland" of cameras.

WORKS

Fuses city, private and police cameras, plate readers, gunshot detection, CAD, body cams and drones in one view. Private owners keep control through a converter box and share on their own terms, often only during an incident.

NUMBERS

About 200 agencies around 2022. By 2026: 6,000 real-time centers on Fusus and 1M+ cameras integrated, 380,000 of them privately owned. Third-party cameras were 245,000 in March 2025. Definitions likely differ.

WHY BOUGHT

Fusus rolled out to several hundred agencies in the time Axon's CAD got to two. Bagley says Fusus was doing it better and faster than Axon was. It is now moving into enterprise security centers.

LARGEST USE

An undisclosed global logistics customer integrating 300,000 video streams, tied to the largest deal in company history at the time.

Axon Vision

Early access at Axon Week 2026. Turns existing cameras into detectors through "lenses" for fights, fire, fence climbing or inmate counts. Uncertain hits go to a human analyst whose feedback retrains the model. Runs at the edge. Already live in corrections and some enterprise and police sites.

Outpost, Lightpost, Fleet 3 and Works with Axon

Axon entered fixed plate readers only in 2025, after years of building ALPR into patrol cars.

WORKS

Fleet 3 reads plates from the dash camera with passive detection, shipped 2022. Outpost is a pole camera with plate and wide video and on-device vehicle signatures. Lightpost, with Ubicquia, plugs into a streetlight socket.

PRICE

Outpost $2,750 and Lightpost $3,250 a year, against about $3,500 for incumbents, first year free for early adopters (Axon Week 2025).

NUMBERS

Fleet: 8.1B reads, 20,000 felony arrests, 50,000 vehicle recoveries in a year. Concept to general availability in under a year, the fastest major hardware product Axon has built, per Kunins.

RIVALS

Smith calls the market fragmented "walled gardens". Flock was once a partner; see competition. Denver and Syracuse are cited as moving to Axon after other programs were shut.

PITCH

Partner agency searches do not copy your data. Only a hit that becomes evidence is copied. A board member recalls refusing to back buying an ALPR leader unless its data-resale business was shut.

Community Request, Ring, Citizen

Geofenced requests for neighbors' footage, with Ring as first partner, landing in Evidence with an audit log and a public transparency portal. Citizen integration (limited beta in 2025) streams users' live video into Fusus. Kunins says the earlier Ring program drew "bad press" built on unfair assumptions, and that it lacked chain of custody.

3. The air: drones and counter-drones

Drone as First Responder, with Skydio

Axon chose not to build outdoor drones. It resells Skydio as a commissioned channel partner and supplies the software, airspace awareness and evidence chain.

NUMBERS

600+ agencies. Claims: drones beat patrol cars 75% of the time and clear 41% of calls with no officer sent. Skydio's Bry cites about 2M flights a year (excluding military), doubling yearly, and about 1% penetration of addressable use.

SKYDIO

X10 flagship, R10 for indoor tactical work, F10 fixed-wing launched from a dock: two hours, 100 mph, for rural sheriffs covering thousands of square miles. Bry says Skydio's dock bet began in 2021 with "no market" and no budgets.

LIMITS

FAA waiver required. Isner says other drones work with most features, but there is "no close second".

NEXT

Any patrol car as a launch point. SkySwap gives trade-in credit for Chinese-made drones.

Dedrone

Detects drones, identifies make and locates the pilot. Isner said in May 2026 it was "rivaling" the AI Era Plan as the fastest-growing product.

WORKS

RF is the foundation, with radar and cameras added because fiber-optic and non-standard drones emit no usable signal. Mitigation means jammers and nets. Axon partners rather than builds on interceptor drones, which Isner says start to infringe on the mission.

NUMBERS

Revenue up over 300% year on year (Q1 2026, host figure). Almost every NFL stadium, all US World Cup venues, sensors across Ukraine. Up to 125,000 drone alerts a month in Australia alone. Claimed 3,000+ illegal US flights daily.

LIMITS

The FAA ultimately gets to decide when drones come down. Isner, June 2026 Louisiana State Police wanted it for the New Orleans Super Bowl and Axon could not legally sell it to them.

MARKET

Isner concedes Anduril will own the military platform. Axon targets federal civilian, state and local, enterprise and interior ministries.

SkyHero and "Thor"

Tactical indoor drones and robots. The next-generation mesh radio lets one operator run up to four vehicles. "Thor" is a codename for an integrated magazine firing TASER darts, flashbangs or window breakers. Axon's stated rule: no autonomous use-of-force decisions. Isner calls putting less-lethal on drones the thing he is most excited about, for school-shooting response with a department always making the decision.

4. On scene: force, cameras, language

TASER

It works by overriding motor nerves with 19 to 22 pulses a second, and needs about 12 inches of probe spread to lock muscles. Below that it only causes pain. Bagley said in August 2025 that it was under 40% of revenue, with over a million sold.

THE GAP

Smith: reliability "only about 80%", goal high 90s. Agencies using three of four best practices average 93% effectiveness versus the high 60s to 70% elsewhere. His claim: we are not effective enough of the time.

EVIDENCE

Cited by Smith and Kroll: studies showing suspect injuries down two-thirds to 80%; 31 causal deaths from about 4.5M uses. Effect on shootings: one agency down 40%+. Norway introduced it in 2024 and saw 40% fewer shootings on a small base.

MOONSHOT

Halve US police-public gun deaths by 2032 or 2033. In March 2025, deaths were up 9% and Smith said "we're not on pace". By Axon Week 2026 they were down about 10%, and Smith called it too early to claim credit.

CONSUMER

Smith said "2026" for a relaunch aimed at gun owners, in stores that sell guns, then "the next couple of years". Lobbying is underway because TASER 10 counts as a firearm under old rules, which bars corrections and healthcare uses.

RIVALS

Barely present. Arsenal, a Brazilian firm and Chinese makers are named as still winding wire in the cartridge. Isner calls Byrna-type launchers "progress" rather than competition.

Body cameras and Signal

Berkley's point from about 2015 explains the business: you can strap on a GoPro, and the question is what you do with the data after.

SPECS

Body 4: up to 1440p, 30 fps, about 13 hours, rated to 50C, two-way audio, LTE streaming, optional point-of-view camera. Deliberately no night vision, to avoid implying officers see better than they do.

AUTOMATION

Signal starts cameras from a holster sensor, light bar, crash, or TASER arming. Safariland is adding about 30 holster fits.

INTEGRITY

Encrypted storage, no remote deletion, SHA-256 and Merkle-tree hashing, device audit logs of every button press. The on-screen watermark doubles as authenticity mark and advertising.

KEY BET

Adding an LTE chip at about $70 cost without a second SKU. Customers said they didn't want streaming. Connectivity later enabled AI features.

SCALE

350,000 officers (2019) to about 1M (2025), about two hours of video per officer per day. An investor estimates 85%+ of US departments use Axon hardware. Hardware life is long: customers still order Body 2.

Axon Assistant, translation, Policy Chat, Guardian

TRANSLATE

Hands-free on Body 4. 500+ agencies and over a million conversation turns in under a year (April 2026). Language counts vary from 50+ to about 150.

ASSISTANT

Voice plate lookup, policy questions with citations, BOLO drafting, form completion. Rationale: consumer chatbots expose officers to public records and privacy risk. Smartwatch, CAD lookups and live 911 listening are next.

GUARDIAN

New in 2026. Real-time audio analysis that detects escalation and calls for backup. Replayed on one past incident, it would have alerted "more than three minutes" before the radio call. One retrospective test only.

POLICY

Rick thought Policy Chat was "not sexy". It proved popular because agencies have lawyers answering policy questions around the clock.

VR training

A slow start: Smith says early customers called it "a piece of s***" and he apologized to one. About 10,000 users in year one (Axon Week 2025) grew to over 200,000 officers (late 2025) and almost 300,000 active users (2026). Claims: VR-trained officers are 20% more proficient and 81% more prepared. Training Pod is a prefab bay meant to work "like an ATM". AI characters are expected to replace instructors running scenarios. Metacoach, built by Immersion Labs, is resold on revenue share.

5. Paperwork

Draft One

The product that turned AI from a story into bookings. Isner calls it the fastest-selling product in Axon's history.

WORKS

Uses body-camera audio only. Smith says adding vision made reports longer and worse. A draft arrives about three minutes after the incident, with placeholders that block submission until filled, and agency-set acknowledgements. Runs on third-party foundation models, mostly OpenAI via Azure.

NUMBERS

600,000+ reports and 300,000+ hours returned in the past year, about 30 minutes per report (2026). Fort Collins measured a 70% reduction.

SAFEGUARD

Released only after scoring less biased than human reports. An early version wrote that white subjects "ran away" and Black subjects "fled". Berkley on the study: we did pay for the study, but we did not pay for the result.

FRICTION

Kunins says about 95% of skeptical prosecutors relent after seeing it. Noise degrades it. He also calls an imperfect draft "almost a feature": a 99%-perfect one would stop officers reviewing.

SIDE EFFECT

Officers narrate more on camera, which improves the record.

Form 1

Announced 2026. A browser extension, compared to Grammarly, that fills any web form in any records system with no integration, from body-cam transcripts, voice, scanned licenses or screenshots. Strategically notable: it delivers AI value inside rival systems. Smith says Kunins pushed the design over his doubts.

Axon Records

Smith says Axon has more major cities on its records system than any incumbent, and calls Draft One its beachhead. Built on a 7 to 8 year old bet that AI would write reports from video. Kunins' view of why it took that long: LLMs only became good enough around 2023.

6. Evidence and court

Evidence.com

The most durable asset in the set. It began with a data loss: an in-house storage array corrupted its file pointers and customer data was recovered only through an out-of-policy backup. Axon wrote off about $6.5M of hardware and moved to AWS, later Microsoft.

SCALE

Over 90% of law enforcement (Holland). 22 PB in 2018; about 200 PB (investor, 2026); "north of an exabyte" raw (Kunins). Said to ingest more video per year than YouTube.

PACKAGING

Unlimited first-party storage for years; unlimited third-party storage at a flat per-user monthly rate. Evidence-sharing network pulls in DAs, some of which now refuse submissions outside it.

GAPS

From customers on an Axon-hosted panel: Cellebrite workflow is weak, 4 GB browser upload limit, legacy migrations need services. A mistagged category can delete a video for good.

INTL

Scotland runs it as a national backbone: 350,000 evidence items, 15,500 backlog cases cleared, witness callbacks down 78%. AI is what pulls reluctant foreign agencies to cloud.

Justice tools

Software for prosecutors and defenders, started around 2022 as Axon's first launch without a device. Coming: case and key-moment summaries, cross-case search for officer patterns. A simple assault file is now 6 to 12 hours of content, per Berkley, against about one page 30 years ago. Axon Gravity (2026) aims to be an agency-wide AI analyst. A corrections product team is forming.

Forensics

Axon employees analyze devices and testify in court, including for a prosecution in a murder trial of an officer. Bryan Chiles reports 500+ TASER investigations and about 70 testimonies. Pulse logs show connection quality for each pulse. It builds the product's credibility in court, and it also means the vendor is often the expert on its own device.

7. Enterprise

Retail, logistics, healthcare and critical infrastructure. Isner: we will have done something Really wrong if enterprise doesn't eventually become the biggest part of our business

PRODUCTS

Axon Body Workforce and the lighter Body Mini (July 2026), Fusus for security centers, Retail Crime Hub with Auror, Dedrone for sites.

RESULTS

REI pilot at two stores: 58% less use of force. "Dozens of thousands" of retail cameras deployed. Anchorage PD on the Auror hub: 194 cases in about 11 weeks, 181 arrests or warrants (customer self-report).

DEALS

One near $50M and one "well over 100 million" in two years.

LIMITS

Isner says it is "crazy to expect the same ASP" as public safety. Bagley called enterprise still small in revenue (2025). Privacy and wiretap law are the main hurdles at retailers. No network effects between competing CEOs.

8. Bundle and price

Kunins calls it the "Amazon Prime-ification" of how public safety buys: hardware, software and consumables on one price over mostly five-year contracts, with a growing ten-year cohort. Predictable spend helps at city council. Customers adopt modules when ready.

9. What failed, and what that taught

Industry: who buys, and how

• Fragmented customers, small market. About 18,000 US agencies, only about 1,200 with over 100 sworn officers. Many have 20 to 50 staff and no security team. Isner: US public safety is under a million users, so you must win most of them.
• Budgets are nearly all people. One speaker puts labor at about 95% of police budgets. Bagley says budgets rise about 5% a year. An investor says Axon is about 1% of police budgets.
• Peers sell the product. Chiefs call each other; 911 directors keep group chats. A serving officer: "Cops hate to be pitched to."
• Adoption is slow until it's sudden. Smith says tech shifts here take about seven years. Cloud evidence was first called illegal. COVID flipped 911 cloud attitudes in months.
• Procurement can be bypassed or bundled. Prepared went free and bottom-up. Carbyne used partner contract vehicles. Axon's AI plan cleared some councils in about two months.
• Risk aversion is rational. Chime on 911 directors: they get fired for failures, not for lack of innovation.

Staffing is the demand driver

• Pulse of 911 survey (NENA and Carbyne, 2023, about 850 to 1,000 respondents): 82% of centers understaffed, some short by half. Training takes 8 to 12 months.
• Police: vacancies "in the dozens or hundreds" at big-city forces. Officers said to spend 30 to 50% of time on paperwork, depending on the speaker.
• The product story follows directly. Smith calls Draft One a "50% boost to your police force". Isner says departments "don't necessarily need to hire as many people".
• The downside surfaced by a defender-side talk: funders may cut headcount when tools create efficiency.

Competition

Axon's M&A logic is itself a competitive claim. Rick says it buys up-and-comers on modern stacks, while competitors "buy market leaders" carrying tech debt. Openness is the other: Kunins says interoperability is not where people should compete, and Works with Axon includes competitors.

How the AI shift changes the market

• Axon is a model customer, not a model builder. It uses OpenAI, Mistral and others through government clouds, and first-party vision models only where foundation models lose on price-performance, such as plate reads at 140 mph closing speed.
• Data and distribution are the claimed moat. An investor argues the "SaaS apocalypse" misreads Axon because proprietary data allows narrow, cheap models. That is an opinion from a shareholder.
• Features commoditize fast. Elichai says incumbents added transcription and translation once ChatGPT arrived, leaving trust and customer success as the edge.
• Cloud becomes a prerequisite. Isner: you can't run the large models locally. That drives upgrades at home and cloud adoption abroad.
• Hybrid is the consensus. Elichai expects human plus AI until about 2030. Holmes went from fearing replacement to expecting supplementation. Jason Droege warns workflows needing 99.9% reliability break under agents at 90% per step.
• Internal effects. Kunins reports about 25% engineering compression from agentic coding, with a stock pool paying for squads that exceed it.

Regulation and geopolitics

• Airspace: only federal authority allows drone takedowns; cities need waivers. DFR needs FAA waivers. A law is pending for handheld Dedrone Defender use.
• China: legislation phasing out Chinese drones helps Skydio and SkySwap. China sanctioned Skydio and its battery supply was cut, forcing a rebuild. Axon's own about 1,000 suppliers lean heavily on China and Taiwan; it wargamed a Taiwan takeover and holds deliberate inventory.
• Firearms law: TASER 10's propellant makes it a firearm, limiting who can carry it. A camera that picks when to fire would make it a "machine gun" legally.
• Tariffs: "part of the plan", though Smith says they lowered margins. A 2025 tax change produced a GAAP net loss alongside stock compensation.
• Federal: Isner saw DOGE-style cuts as a tailwind once funds shift to law enforcement.
• Funding: Holmes describes a roughly $15B House bill to modernize 911 centers (around 2023), and a push to reclassify 911 staff as first responders.
• Sovereignty: GDPR and data location slow European cloud. Canada's defense procurement is shifting toward domestic supply, per a 2025 panel.

Critics and public trust

• TASER safety remains contested. John Oliver argued these weapons should be used only in extreme situations and criticized "excited delirium" as discredited. Smith defends the term and blames the plaintiffs' bar. A 2012 San Francisco commissioner pressed on evidence that shootings did not fall. Only the rebuttals are in this corpus in full.
• Warnings moved under legal pressure. After Canada's Braidwood Inquiry, Axon lowered recommended point of aim. Smith says lawyers overrode the science board.
• Axon litigates hard. It sued an Ohio medical examiner and won removal of a finding. Smith describes "go nuclear" defense of product suits, about 400 of them.
• Face recognition is moving. The ethics board said no in 2018. Kunins dates a moratorium to 2019. By late 2025 Axon was preparing field trials; by 2026, a narrow evaluation with Canadian agencies. Isner names the condition as demonstrating algorithms are free of racial bias, and expects it to be "very political".
• The ethics council's weight is debated. A host raised the "window dressing" critique. Berkley says the council sometimes rejects ideas. Kunins says it is non-binding. The original board fell apart over the TASER drone.
• Surveillance worry is acknowledged. Axon's answers are owner-controlled sharing, audit logs and transparency portals. Skydio's Bry says public safety is his "most controversial" market and that he doesn't want a "blanket surveillance world".

New markets

Capital markets, as discussed

Other investor issues raised: stock compensation at about 20% of revenue (host), ten-year contracts inflating backlog, inventory build, and a Scottsdale campus fight that Isner says could push Axon out of Arizona. On campus cost, he says a $1.3B figure is false and is closer to 10% of that.

Timeline

Where the sources disagree

These are worth checking before relying on any single figure.

All 130 sources

Every file was read in full. Items tagged "no product content" were read and contributed nothing here. Names are cleaned from file titles.