Vidaio sells AI video compression and upscaling by the minute through a website and an API, and runs a Bittensor subnet whose miners compete every two weeks to supply the models. Its best-known claim is that its files are "up to 80% smaller", restated this week by an investor as "half the size of AWS compression"; the team's own commissioned benchmark (17 Jun 2026) found 7.8–9.7% fewer bits than open-source FFmpeg at the same picture quality, with the cost advantage coming from hitting a quality target in one encode. Revenue is claimed and not evidenced: a credit-billed API, a joint venture with a UK post-production house and one enterprise test are described, but no customer count, usage or dollar figure is published. The token is a claim on $4.43M a year of issuance, 86% of its $5.17M market cap; the owner's account has been a net seller of alpha for the last 30 days.
| Claim | Who said it | What the check shows |
|---|---|---|
| Files "up to 80% smaller" (elsewhere 95–96%) | Revenue Search, Oct 2025; repeated by Stillcore (Jan 2026) and TAO Daily | No baseline or method published. The team's own matched-quality benchmark (next row) shows 7.8–9.7% fewer bits than FFmpeg, so the 80% figure is against something other than a tuned modern encoder |
| Ranked first for compression efficiency against AWS MediaConvert, Bitmovin and Brightcove; ~$0.067 a minute against up to $2.42; $4k against $145k per 1,000 hours | @vidaio_, 17 Jun 2026; TAO Daily, 18 Jun | Commissioned by Vidaio, firm unnamed, report not linked. 30 clips, AV1 and H.265, three VMAF targets: 7.8% (AV1) and 9.7% (H.265) fewer bits than self-hosted FFmpeg at matched quality; on target in 68% of jobs, within one VMAF point in 95%. The cost gap comes from rivals needing three to five encodes to hit a target, so it compares workflows, not rate cards |
| "Video files are HALF the size of AWS compression"; could halve Netflix's streaming bill | Mark Jeffrey (Stillcore), 9 Oct 2026, relaying the CEO's Exploit Summit talk | Talk not reviewed. The nearest primary is the same June benchmark, where TAO Daily reports AWS's file 90% larger than the FFmpeg baseline at matched VMAF after a single encode. Unresolved |
| Joint venture with PIP Studios: "a gateway to 50+ major partnerships", Netflix, Amazon, Sony, Universal, Paramount | @vidaio_, 28 Apr 2026 | PIP Studios is a UK audio-post and localisation house (Winnersh, Molinare group); the studios are PIP's clients. Terms undisclosed; revenue split "in proportion to actual usage". PIP's own site does not mention Vidaio or CABIN |
| CABIN, PIP's AI video platform, is "powered by Vidaio" and showed at IBC Amsterdam | @vidaio_, 12 Aug; IBC debrief, 15 Sep | The debrief reports seven demos to prospects (a public broadcaster, a film studio, an entertainment franchise, a post house) and six follow-ups; no contract. No PIP-side or IBC-side source found |
| An enterprise client switched from ElevenLabs after Vidaio scored 82/100 against 52/100 on an English-to-Japanese dub | @vidaio_, 4 Aug 2026; TAO Daily adds "hundreds of millions in annual revenue" | Vidaio's own QA score on one test; client unnamed; no payment stated |
| The winning miner "gets a revenue share from Vidaio for as long as that miner's model is used with clients" | @vidaio_, 8 Oct 2026 | The only mechanism in code is the licensing treasury, funded from emission (MINING.md); the whitepaper puts revenue mechanics "out of scope". No share of customer money is coded |
| "Accelerated revenue and market traction lately" | Mark Jeffrey, 8 Oct 2026 | A relay with no figure; Stillcore's own publications list Vidaio without a position size |
| Vidaio moved its storage from AWS S3 to Hippius and saved about 50% | Hippius blog, 25 Aug 2026 | Team estimate, no amounts, AWS and Backblaze kept as backups; both companies share an adviser |
| Item | Detail |
|---|---|
| Products | Video upscaling (2×/4×) and AI-driven compression, by upload on the site or by URL through the API; live streaming with AI dubbing and subtitles is "coming soon" (the beta streamed the Exploit Summit on 28–29 Sep); dubbing and localisation are sold through PIP Studios as CABIN |
| How it is billed | Credits: the API refuses a job without credits (HTTP 402) and consumes them when the result is fetched. No public price list; the June benchmark priced Vidaio at about $0.067 a delivered minute. The October 2025 plan was about $0.05 a minute at a 75% margin, with enterprise fiat split 75% to miners and 25% to Vidaio; none of that is in the September 2026 whitepaper |
| Channels | "One stack. Five verticals" (18 Aug): PIP (localisation dashboard), Sentinel (own compression tool), Manako (compression embedded on a client's site), Vidaio Enterprise, Vidaio Retail; "PiP and Manako are rolling out" |
| Customers | None named. The 4 Aug enterprise test and the IBC prospects are unnamed; PIP's studio clients are PIP's. No usage or revenue figure on the site, API, X, TAO Daily's twenty-plus articles or the podcasts |
| Financing | DSV Fund OTC purchase, 1 Dec 2025, amount undisclosed (TAO Daily); MEXC listing 6 May 2026; Stillcore covers Vidaio and its partner mined it, no round disclosed. Incubated by the parent company of Taostats (Graichen, Apr 2025) |
| Item | Detail |
|---|---|
| Company | No legal entity named on the site or in UK Companies House ("vidaio": no result). Incubated by Taostats' parent; Marcus Graichen (Taostats, Hippius) is listed as "Angel Advisor" on the site and as co-founder in his own bio |
| Named people (eight on the site) | Gareth Howells, Director and CEO ("25+ years in video operations"); Dr Ahmad Ayad, ML engineer, called CTO by Hippius; Gopi Jayaraman, video technology; "Bruhvatar", subnet developer (ex nineteen.ai); William McCrea, business development and ISO 27001; Chinaza, design; Akin, front end; Graichen, adviser |
| Team size, checked | The team has never stated a headcount. Seven people have ever committed to the original repository (one of them 599 of about 760 commits); every 2026 commit to it is by one person. The live code is a mirror of a private repository pushed by two identities since September, so two or three committers are visible in the last 90 days against eight names |
| Encoder or method | Compared with | Bits saved at equal quality | Source and caveat |
|---|---|---|---|
| Vidaio (AV1) | Self-hosted FFmpeg, same codec | 7.8% | Vidaio-commissioned; hit the target in one encode in 68% of jobs |
| Vidaio (H.265) | Self-hosted FFmpeg | 9.7% | Same test |
| AWS MediaConvert, single encode | Self-hosted FFmpeg | −90% (file larger) | TAO Daily, 24 Jun on the same test; AWS was not allowed its usual multi-pass search |
| Beamr CABR (content-adaptive bitrate) | The customer's existing encode | 30–50% claimed | Beamr on AWS Marketplace; vendor claim |
| Deep Render neural codec | SVT-AV1 | 45–46% | Jan Ozer, May 2025, subjective test; low-latency configuration only |
| AV1 (any encoder) | x265 (H.265) | about 35% | Streaming Media; the codec step buyers take first |
Upscaling has no published head-to-head: the MSU upscaler benchmark lists Topaz and Real-ESRGAN but Vidaio has not entered it, and Vidaio's competitions score entries on its own hidden clips, so quality against Topaz (now Adobe's) is unrated rather than rated and losing.
| Vendor | What the buyer gets | Price per output minute |
|---|---|---|
| Vidaio | Compression to a VMAF target in one encode; upscaling 2×/4×; credit-billed API | about $0.067 (benchmark estimate); upscaling not published |
| AWS Elemental MediaConvert | HD AVC single pass $0.024, multi-pass $0.042; HEVC $0.048–0.084; 4K higher | $0.024–0.084 |
| Bitmovin | Per-title encoding, pay as you go after 2,000 free minutes a month | $0.02 |
| Google Transcoder API | HD output, any codec | $0.030 |
| Beamr Cloud | Content-adaptive bitrate reduction on AWS | $0.04 plus compute |
| Mux, Coconut, api.video | Encoding bundled with hosting and delivery | free–$0.031; $0.015 |
| Topaz Video (Adobe) via fal.ai | AI upscaling to 720p, 1080p, 4K | $0.60–3.60 |
On rate card Vidaio is two to three times the cloud encoders. Its pitch is that it reaches a quality target in one pass where the others need three to five, which only matters to a buyer who encodes to a VMAF target rather than a preset; the benchmark's $145,000 alternative is that search loop priced at list.
| Measure | Figure |
|---|---|
| The layer | Per-minute cloud video encoding and upscaling sold as an API (the MediaConvert, Transcoder, Bitmovin, Beamr slot), not a hosting platform |
| Floor: revenue the named encoding vendors disclose | About $0.32B: Harmonic's video business $190.5M (2024), Ateme €95.6M (2025), Bitmovin "in the twenties" of millions (2024), Beamr $3.1M (2025). AWS and Google do not break out their encoders, so this is a hard floor |
| Ceiling: video transcoding, software and hardware | $2.56B in 2025 (IMARC), 11% a year to 2034; the largest named vendor fits inside it. Platforms that bundle hosting and delivery are a wider $7.5B (MarketsandMarkets), which Vidaio does not sell into |
| What winning companies are worth | Topaz Labs, AI upscaling: bought by Adobe for about $340M (Jun 2026, Adobe 10-Q). Harmonic's video business: $145M cash from MediaKind, 0.76× revenue (Jun 2026). Beamr: $27.5M market cap on $3.1M revenue. Deep Render: $30M (2023). Mux: over $1B (2021). Vimeo $1.38B and Brightcove $233M to Bending Spoons at 3.3× and 1.2× revenue. Apple/WaveOne and Sony/iSIZE undisclosed |
| Vidaio's issuance against the market | $4.43M a year of alpha, about 1.4% of the disclosed floor. Size is not the constraint; a public price, a named paying customer and a usage figure are |
| Who | What they do | How they are paid | Per day | Customer money? |
|---|---|---|---|---|
| Miners (five paid slots) | Enter a sealed code competition every two weeks (compression, upscaling, object removal, now frame interpolation); the latest result's five placed entries share all miner emission for two weeks, 89.55% to the winner. The 73 miners serving everyday compression and upscaling requests earn nothing since 27 Sep | New alpha | $5.0k | No |
| Validators and stakers (one paid hotkey) | Copy the weight vector that Vidaio's own scoring server publishes each epoch, after checking its hash on chain. The owner's hotkey receives all validator dividends; 25 outside hotkeys delegate through it and earn 30–37% a year | New alpha | $5.0k | No |
| Vidaio, the subnet owner (coldkey 5GTPB…) | Runs the scoring server, the competitions, the website and the API; holds 2.2% of alpha; sold a net 134 τ of alpha in the 30 days to 11 Oct | The 18% owner cut in alpha | $2.2k | No |
| Licensing treasury (UID 111) | A hotkey the team registered on 6 Oct to pay "licensing fees" to winners whose code runs on the website; it takes 0.5% of miner emission while a result pays and 50% of any gap | New alpha | $0.0k | No |
| Alpha holders | No buyback, burn-from-revenue or fee sink. 30 days to 11 Oct: owner buys −$38,864 (net sales), burn $14,795 (emission sent to the sink slot, not a purchase), outflow $337,120, team-verified | — | — | No |
| Step | What happens |
|---|---|
| Scoring | A server run by Vidaio (the "scoring authority") scores miners on CPU with libvmaf and PieAPP and writes one log every 360 blocks (about 72 minutes) to a public bucket on Hippius; it anchors the log's hash on chain |
| Validators | Download the log, check the hash, and submit the authority's weight vector unchanged. Two "auditors" recompute and report, without affecting weights. In the old design (to August) validators scored miners themselves on synthetic clips |
| Who earns | Since 27 Sep a competition result pays 100% of miner emission for two weeks: a result over the "crown" bar pays 90/4/3/2/1 to the top five, one under it pays 50/24/13/8/5. Serving everyday requests pays nothing. A gap with no result pays 50% to the burn and 50% to the licensing treasury |
| Customer jobs | Run through a separate gateway "outside the consensus loop" with "no effect on emissions": the token pays for the contest, the website pays (if anything) for the serving |
| Question | Answer |
|---|---|
| Cadence | A competition every two weeks, each result paying for two weeks. Live: frame interpolation 003b (double the frame rate; speed counts 30% of the score), enrolment closes 16 Oct 14:00 UTC, results not before 20 Oct |
| Entry cost | Registration 0.0005 τ; a stake floor of 500 α (about $840) on the entering hotkey; the miner's own development and GPU time. Entries are private repositories at a pinned commit, run sealed without network on an L40S against 16 hidden clips; commercial licences only |
| Split, on chain | Crown (score at or over the bar, 0.85 here): 90/4/3/2/1 to five places; podium: 50/24/13/8/5. Today UID 202 holds 89.55% for winning object removal 002 (score 0.789 against 28 ranked entries); the licensing treasury takes 0.5% |
| Miners paid against entries | Five paid of 28 ranked in the last result; three entries so far in the live one. The 73 miners enrolled on the inference tracks (51 compression, 22 upscaling) are paid nothing |
| Is the winning code published? | A crown winner is paid only after its source archive is released, and the results page links one; podium code stays private. The team's premium encoder is "not part of the public build" |
Miner slots (UIDs) validators paid each day, read from chain (latest block 9,248,983); the owner's own key is left out because the chain burns its share. Slots, not companies. About 100 slots were paid until mid-June 2026; the June rule change paid the top five per pool (ten), and since 27 September only the latest competition's five placed entries and the licensing-treasury slot are paid, so the fall tracks the payout rules, not miners leaving: 73 miners are enrolled on the inference tracks and earn nothing.
| Component | Public? | Detail |
|---|---|---|
| Original validator and miner code | Yes (README says MIT; no licence file) | vidaio-subnet/vidaio-subnet, last commit 18 Aug 2026 |
| Live validator, miner and scoring rules | Mirror only (MIT) | miner-vidaio and validator-vidaio are "release mirror snapshots" of a private repository; the scoring server itself runs only at Vidaio |
| Crown-winning submissions | Yes, after payout | Released archive required before a crown pays; commercial licence so Vidaio can run it on the website |
| Podium submissions, premium encoder, website, API, dubbing pipeline | No | What a customer pays for is the private serving stack plus the licensed winner |
Open crown code lets any competitor run the same model; what Vidaio keeps is the serving stack, the licence rights and the PIP channel.
| Alpha price and market cap | 0.00582 τ = $1.68; 3,071,932 α outstanding = $5.17M; 13,643 τ ($3.95M) in the pool |
| Alpha issued | 7,200 α/day = $12.1k/day, $4.43M/yr at spot, 86% of market cap; $12.01M since April 2025 at the prices of the time (chart below) |
| TAO the network injects (its cash cost) | 6.4 τ/day = $1.9k/day; 0.56% of all subnet TAO inflow, 36th of 93 subnets |
| The owner coldkey's holding | 66,813 α = 2.2% of outstanding ($112.5k), nearly all on its own validator hotkey. The account has signed 179 transactions; taorevenue shows it sold a net 134 τ ($38,864) of alpha in the last 30 days |
| Owner lock | 113,874 α (3.7% of outstanding) locked permanently to the owner hotkey by an outside account, 5HMtFEQ9…, which holds similar locks on at least eight other subnets; the owner itself placed a decaying lock of 28,480 α (0.9%, about a third of what it holds). Automatic locking of the 18% cut is off |
| Owner's 18% cut | $2.2k/day, $0.80M/yr; against no measured customer revenue. Net insider flow: the cut plus net sales, no verified buyback |
| Party | Today | What would change it |
|---|---|---|
| Vidaio | Keeps the credit payment less its serving cost; the 2025 plan to pass 75% of enterprise fiat to miners is not in the 2026 design | A published price and a usage figure |
| Miners | Emission only; the "revenue share" announced on 8 Oct is the emission-funded licensing treasury | A fiat licence fee paid to winners, visible in a ledger |
| Alpha holders | $0.00: no buyback, burn-from-revenue or fee sink; the owner account is a net seller | The promised profit-funded buyback (Jun 2026) executed from an identified wallet, at a stated share of revenue |
| Plan | Source | Status |
|---|---|---|
| Frame-interpolation competition, then two-week competitions on a rolling calendar | competitions page | Live; enrolment closes 16 Oct |
| Licensing treasury paying winners whose code serves the website | MINING.md, 6 Oct | Registered (UID 111); fee rules unpublished |
| Live streaming with AI dubbing as a product | vidaio.io ("coming soon"); @vidaio_ | Beta used for one conference |
| Enterprise, Retail, Sentinel and Manako channels | @vidaio_, 18 Aug | "Rolling out" for PIP and Manako; no customers named since |
| Profits fund alpha buybacks | TAO Daily, 2 Jun | Announced; no wallet, share or purchase |
| Enterprise fiat split 75/25 with miners; "AlphaBond" locking | Revenue Search, Oct 2025 | Not in the 2026 whitepaper or code |
The plan that matters most to the case is a public price with a usage figure behind it; the buyback is second, and both are promises today.
| Bear | PIP's CABIN stays a demo channel, the API finds no named buyer at $0.067 a minute against $0.02–0.04 list prices, competitions keep paying one winner 90% of a shrinking emission, and the owner keeps selling its cut |
| Base | A few PIP-sourced localisation jobs and library-compression deals produce real but undisclosed revenue; the licensing treasury pays winners in alpha; the token trades on emission and ecosystem flows with no customer money reaching it |
| Bull | A named enterprise or broadcaster contract, a public price card, an independent benchmark that reproduces the one-encode advantage, and the promised profit-funded buyback executed at a stated share of revenue |