Runway
Tiered SaaS subscriptions ($12-$95+/user/mo) with metered GPU credits, plus usage-based API for developers and enterprise/studio licensing deals (e.g. Lionsgate, AMC Networks)
All rounds are press-reported private primaries (~$860M cumulative since 2018 across seed through Series E); Runway does not disclose audited figures. Mid-2026 secondary marks ($10.69 Notice to $14.58 Nasdaq Private Market per share) are too thin and non-comparable to plot as valuation points.
Earnings, margins, COGS & capex
Private company; no audited public financials. Press-reported trajectory: ~$44M recognized revenue in 2024 (+236% YoY per Sacra), ~$70M ARR at end-2024, ~$90M ARR by mid-2025, with a company forecast of $265-300M ARR by end-2025 driven by Gen-4/4.5 adoption, the API, and enterprise/studio deals. Losses are large (reported ~$155M EBITDA loss in 2024) because model training and GPU inference outpace sales. Feb 2026 Series E ($315M at $5.3B post, led by General Atlantic with NVIDIA, Adobe Ventures, AMD Ventures, Fidelity, AllianceBernstein, Mirae Asset, Felicis, and Premji Invest participating) funds pretraining of next-generation world models.
Revenue trend
Margins
GPU inference cost per generated video is the swing factor; improving with model efficiency but structurally lower than classic SaaS
large absolute losses driven by training spend; revenue scaling faster
dependent on venture funding
COGS structure
Dominated by cloud GPU compute for inference (video generation is orders of magnitude more compute-intensive per request than text), plus model hosting and data pipeline costs; AWS was named Runway's primary cloud in 2024 and NVIDIA silicon (incl. announced Rubin/Vera Rubin platform adoption) is the underlying supply.
Capex
Asset-light on paper - compute is rented, so the capex-equivalent shows up as opex/committed cloud spend; the Series E is effectively a compute war chest for pretraining next-gen world models.
Latest earnings
n/a
No formal guidance; company forecast (per press) was $265-300M ARR by end-2025; one Jul 2026 item claims ~$40M ARR added in Q2 2026, but it appears only in a low-quality syndicated press release and is unverified
- Last priced round
- Series E, $315M at $5.3B post (closed Feb 10, 2026), led by General Atlantic; NVIDIA, Adobe Ventures, AMD Ventures, Fidelity, AllianceBernstein, Mirae Asset, Felicis, Premji Invest participating
- Prior round
- Series D, $308M at ~$3.3B (Apr 2025), also led by General Atlantic - Series E was a ~60% step-up
- Cumulative funding
- ~$860M since 2018 founding (press-reported)
- Secondary marks
- Nasdaq Private Market est. ~$14.58/share (Jun 17, 2026); Notice quotes ~$10.69 - wide spread typical of thin private marks
Growth drivers
- Gen-4.5 (ranked #1 on the Artificial Analysis text-to-video leaderboard at its late-2025 launch; native audio and multi-shot editing added Dec 2025) driving prosumer and enterprise subscription upgrades
- GWM-1 world model family (Dec 11, 2025) — GWM Worlds (explorable environments), GWM Avatars (conversational characters), GWM Robotics (synthetic training data and policy testing) - opens simulation, gaming, and robotics TAM beyond creative video
- API and enterprise channel — Lionsgate custom-model deal (Sep 2024) expanded Jun 2026 with a Lionsgate equity stake and joint IP-development program; AMC Networks deal (Jun 2025) expanded Mar 2026 with Gen-4.5 and GWM Avatars endpoints
- NVIDIA partnership (Series E investor — Gen-4.5 was ported to the Vera Rubin NVL72 platform ahead of its release)
- Hollywood/marketing production cost deflation - previsualization, VFX, and ad creative workflows
Bull & bear
Runway is the credible independent lab in generative video with real enterprise distribution, and its world-model pivot converts a commoditizing video niche into a platform bet on simulation, avatars, and robotics - with NVIDIA's capital and silicon behind it.
- ARR trajectory (press-reported ~$44M 2024 revenue to a $265-300M end-2025 ARR forecast) is one of the fastest scaling curves in applied AI outside the frontier labs
- Gen-4.5 beating Google and OpenAI video models on the Artificial Analysis leaderboard at launch proves a startup-scale lab - a fraction of frontier-lab headcount - can stay frontier-relevant in video
- GWM-1's three-variant strategy (Worlds, Avatars, Robotics) opens TAM far larger than creative tooling - robotics simulation and interactive environments are where world models monetize next
- Studio deals are moats-in-progress: Lionsgate deepened to an equity stake plus joint IP development (Jun 2026) and AMC Networks expanded (Mar 2026); workflow integration and custom models create switching costs that consumer video apps lack
- Strategic cap table (NVIDIA, General Atlantic, Adobe Ventures, AMD Ventures, Fidelity) provides compute access, credibility, and a path to IPO-scale later rounds; the Nasdaq Private Market mark (~$14.58, Jun 2026) sits above implied round pricing at some other venues
Runway is a structurally unprofitable, sub-scale model lab whose core product is being commoditized from above (Google, OpenAI, Meta) and below (Chinese video models), and whose $5.3B valuation prices in flawless execution of a still-unproven world-model pivot.
- Unit economics are unproven: ~$155M EBITDA loss on ~$44M of 2024 revenue (press est.); every model generation requires another nine-figure raise, and gross margins on GPU-heavy video inference may never reach software norms
- Model leadership decayed within months - Gen-4.5 fell out of top leaderboard slots in 2026 as Seedance 2.0 (Feb 2026), Kling 3.0, and newer entrants advanced - showing no durable technical moat in raw video quality
- Hyperscalers bundle: Veo ships inside Google's ecosystem effectively free to consumers; price competition structurally favors players who monetize elsewhere
- Copyright litigation and training-data provenance (defendant in Andersen v. Stability AI with trial set for Sep 2026; reported YouTube scraping) is an unresolved tail risk to both cost structure and enterprise trust
- The world-model/robotics pivot is pre-revenue and pits Runway against NVIDIA Cosmos, Google DeepMind Genie, and well-funded specialists (World Labs); all revenue figures are unaudited press estimates - end-2025 forecast attainment is not independently confirmed
- At ~$5.3B, the round implies ~18-20x forward ARR on forecast numbers - little margin for a growth stumble; a miss risks a down round and talent flight
What it is worth
Last priced round + forward ARR multiple cross-check (private; no public comps trade on video-gen revenue)
$2-3B down-round scenario
if video-gen commoditization stalls ARR growth, GWM monetization lags, and crossover investors mark down - roughly the Series D level
$5-6B
grows into the Series E mark as ARR scales toward $400-500M while the multiple compresses toward ~10-12x
$8-10B+ at a next round or IPO
if ARR sustains toward $600M-$1B with world-model/robotics revenue materializing and margins improving (premium multiple retained)
Series E (Feb 10, 2026) set $5.3B post-money, a ~60% step-up from the Apr 2025 Series D ($308M at ~$3.3B). Against the company's end-2025 ARR forecast of $265-300M (press-reported, unaudited), that implies roughly 18-20x forward ARR - rich but within range for hypergrowth AI labs in 2025-26 private rounds (frontier labs commanded far higher). The multiple assumes continued ~2-3x ARR growth and eventual gross-margin normalization as inference costs fall. Secondary-market marks in mid-2026 (~$10.69 Notice to ~$14.58 Nasdaq Private Market) bracket a wide range without clear direction. All revenue inputs are estimates; treat the valuation bands as scenario framing, not price targets. Not financial advice.
SWOT
Strengths
- Frontier-quality video models — Gen-4.5 topped the Artificial Analysis text-to-video leaderboard at its late-2025 launch, ahead of Google and OpenAI offerings - research credibility despite startup scale
- First-mover brand with filmmakers and studios (organizer of the Runway AI Festival; Lionsgate and AMC Networks deals, with Lionsgate taking an equity stake in Jun 2026) - distribution into Hollywood that pure model labs lack
- Full-stack product — models plus an integrated editor/workflow (references, character consistency, Act-Two performance capture), not just an API
- Backing from General Atlantic, NVIDIA, Adobe Ventures, AMD Ventures, Fidelity - capital access and strategic silicon relationships
- Early, credible pivot into world models (GWM-1) diversifies beyond the commoditizing text-to-video niche into avatars, simulation, and robotics
Weaknesses
- Deeply unprofitable (~$155M EBITDA loss in 2024 per press) with GPU-heavy COGS; burn is structural until inference costs fall
- Sub-$1B revenue scale competing against Google, OpenAI, Meta, and ByteDance, which can subsidize video generation indefinitely
- Model leadership is perishable — Gen-4.5 led leaderboards at launch but was displaced in 2026 by Kuaishou's Kling 3.0, ByteDance's Seedance 2.0, and other entrants
- Training-data provenance exposure — added as a defendant in the artists' copyright suit Andersen v. Stability AI (trial set for Sep 2026) and reported (404 Media, 2024) to have trained on scraped YouTube content
- No owned compute; dependent on rented cloud GPUs and continued venture funding for each model generation
Opportunities
- World models for robotics (GWM Robotics) — riding the physical-AI wave with a video-pretrained backbone; NVIDIA alignment helps
- Enterprise/studio licensing expansion beyond Lionsgate/AMC into advertising, gaming, and previsualization pipelines
- API monetization as video generation gets embedded in third-party creative tools
- Hollywood production-cost deflation adoption cycle is still early - the Jun 2026 Lionsgate program moves AI from marketing/previz into short-form episodic content
- Potential IPO or strategic acquisition as consolidation hits mid-tier AI labs
Threats
- Google (Veo 3.x, bundled with Gemini/YouTube), OpenAI (Sora), and Meta shipping video generation at consumer scale and near-zero marginal price
- Rapidly advancing video models from mainland-China players (Kuaishou Kling, ByteDance Seedance, MiniMax Hailuo) compressing quality differentiation and pricing globally (competitive context only)
- Adverse copyright rulings on training data could force licensing costs or model retraining (Andersen trial scheduled Sep 2026)
- Compute cost inflation or GPU allocation scarcity in a capacity-constrained market
- Down-round/valuation compression risk if ARR growth misses the multiple implied by $5.3B (~18-20x forward ARR on forecast numbers)
Moats, dependencies & bottlenecks
Moats
low-to-moderate Real but perishable; leaderboard leadership lasted months. Durability depends on sustained compute funding.
Integrated editor, character consistency, Act-Two, Runway AI Festival community - harder to copy than raw model quality.
Enterprise/studio integration (Lionsgate equity-linked partnership, AMC Networks custom deployments) Custom models and pipeline integration create switching costs; still only a handful of anchor logos.
Proprietary usage data / RLHF loop from millions of creator generations weak-to-moderate Helps model tuning but rivals have larger consumer data flywheels (YouTube, TikTok/Douyin).
Sub-scale versus hyperscalers on compute; this is an anti-moat today.
Dependencies
compute silicon + strategic investor Training/inference on NVIDIA GPUs; Gen-4.5 ported to the Vera Rubin NVL72 platform ahead of its release. Investor alignment mitigates allocation risk but concentrates it.
primary cloud infrastructure Named AWS its primary cloud provider in 2024; no owned datacenters. Pricing and capacity terms are a margin lever it does not control.
Loss-making; each model generation needs fresh capital. Feb 2026 Series E funds the next pretraining cycle.
Defendant in Andersen v. Stability AI (trial set Sep 2026); scraped-video reporting could impose licensing costs or retraining.
enterprise revenue + credibility (Lionsgate is also now an equity holder) Concentrated anchor deals; loss would dent the enterprise narrative more than revenue.
Advantages
- Only independent video-AI lab with both benchmark-credible models and real Hollywood distribution (incl. a studio equity partner in Lionsgate)
- Product-led growth engine (creator community, Runway AI Festival) lowers CAC versus enterprise-only rivals
- World-model pivot started early (GWM-1 shipped Dec 2025) with a differentiated video-pretrained backbone for robotics
- NVIDIA on the cap table - silicon access and validation; AMD Ventures also invested, hedging silicon supply
- New York base gives proximity to media/advertising buyers rather than only tech buyers
Weaknesses
- No path to profitability demonstrated; ~$155M EBITDA loss (2024, press est.)
- Video-generation quality lead proved transient against ByteDance, Kuaishou, and Google releases
- Revenue scale (sub-$300M ARR) is a rounding error versus competitors' AI budgets
- All financials are unaudited press estimates - low information quality for investors
- No owned compute or distribution platform; rents both
Bottlenecks
- GPU compute cost and availability - video/world-model inference is the most compute-intensive genAI modality
- Capital: sustaining frontier-scale pretraining on a startup balance sheet
- Talent retention against frontier-lab compensation (OpenAI, Google DeepMind, Meta)
- Enterprise sales motion still young relative to the product-led prosumer base
- Unresolved training-data IP framework limits some risk-averse enterprise adoption
Top signals & trends
Top signals
Crossover and strategic investors at a ~60% step-up from the Apr 2025 Series D ($3.3B).
TAM expansion beyond creative video; robotics angle aligns with the physical-AI investment theme.
First major studio equity alignment; deepens the Sep 2024 custom-model deal. Some reporting flags IP-rights complications to watch.
Second major media anchor validates the enterprise motion.
Quality moat in raw generation is eroding fast; differentiation must come from workflow and world models.
Thin private-market price discovery; treat marks with caution.
Appears only in a low-quality syndicated press release; unverified - do not weight until corroborated by a credible outlet.
Trends
Google Veo, Sora-class models, and Chinese entrants compress pricing and quality differentiation.
GWM-1 positions Runway in the simulation-and-robotics narrative alongside NVIDIA Cosmos and DeepMind Genie.
Lionsgate (now an equity partner) and AMC deals are early instances; guild agreements now permit bounded AI use.
Andersen v. Stability AI heads to trial (Sep 2026); could impose licensing costs industry-wide, or advantage players who license early (e.g. Runway's prior Getty collaboration).
Inference cost declines (incl. next-gen platforms like NVIDIA Rubin) are the main path to viable gross margins for video generation.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GPUs (incl. Vera Rubin platform port of Gen-4.5); also a Series E investor
Named primary cloud provider (2024) for training and inference
Prior licensed-data collaboration for commercially-safe model variants
Custom model (Sep 2024); expanded Jun 2026 with an equity stake and joint AI short-form IP program
First cable-TV partner (Jun 2025); expanded Mar 2026 across marketing incl. Gen-4.5 and GWM Avatars endpoints
Advertising/creative agencies and prosumer creators Core subscription base ($12-$95+/user/mo tiers plus metered credits)
Veo 3.x bundles into Gemini/YouTube with native audio; Genie competes in world models.
Private. Consumer-scale video generation; product strategy has shifted repeatedly but research capability is frontier-tier.
Kling 3.0 topped video arena leaderboards in 2026; global API distribution. Mainland-China player - competitive context only, not a buy/own call.
Private. Seedance 2.0 (Feb 2026) took top leaderboard slots; TikTok/CapCut distribution. Competitive context only.
Commercially-safe training data and Creative Cloud distribution target the same pro-creative buyer - though Adobe Ventures also invested in Runway's Series E (dual relationship).
Research-strong, distribution-massive; productization slower so far.
Private frontier video startup; raised large rounds incl. Saudi-backed capital.
Private; consumer-leaning video generation, smaller scale.
Private; spatial-intelligence world models - competes with GWM Worlds ambitions.
Partner, investor, AND world-model competitor for robotics simulation - dual-natured relationship.
Private avatar-video leaders; overlap with GWM Avatars in enterprise talking-head use cases.