MiniMax
Dual engine: (1) AI-native consumer products (Talkie AI-companion, Hailuo video) via subscription + in-app spend, US$53.1M FY2025 rev (+143.4%); (2) Open Platform / enterprise-developer API (usage-metered inference across LLM/video/speech/music), US$26.0M FY2025 rev (+197.8%). >70% of revenue from international markets.
Earnings, margins, COGS & capex
Hyper-growth, pre-profit AI foundation-model company. FY2025 revenue US$79.0M (+159%); gross margin improved to 25.4%; IFRS net loss US$1,871.6M (inflated by non-cash preferred-share fair-value changes), adjusted net loss US$250.9M. >70% of revenue is international — unusual for a Chinese model maker — led by the Talkie companion app and Hailuo video plus a fast-growing enterprise API. Trailing P/S ~170x prices in years of compounding; the print is a growth-and-burn story, not a cash generator. Figures per the FY2025 results release and HKEX prospectus.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~75¢ is cost of goods and ~0¢ operating expense, leaving ~25¢ of operating profit (~1¢ net).
Revenue trend
Margins
up from 12.2% (2024) and -24.7% (2023) — improving but structurally low vs pure software
+437.2% YoY from US$3.7M
+302% YoY, but dominated by non-cash preferred-share fair-value changes
roughly flat vs US$244.2M in 2024 despite ~2.6x revenue — the truer burn read
COGS structure
Dominated by AI inference/compute (GPU + cloud), model-serving bandwidth, and third-party API/content costs. The ~25% gross margin reflects that every consumer/API interaction carries real compute cost; margin expansion depends on inference-efficiency gains (e.g. Lightning Attention / M-series efficiency) and cheaper or more-available accelerators.
Capex
No large owned-datacenter capex disclosed; compute is procured (cloud/GPU rental), so training + inference spend flows through opex/COGS. IPO proceeds earmarked for R&D, model-training compute, and global expansion.
Latest earnings
No consensus/beat-miss framework yet — first fiscal year as a public company; results roughly tracked prospectus disclosures.
No formal issuer guidance. Third-party (unverified): Sacra flags ~US$300M annualized run-rate by mid-2026; analyst 2026 revenue estimates (attributed to S&P Global Market Intelligence) sit in the low-hundreds-of-US$M range, with some multi-year models projecting multi-billion revenue by 2030. Treat all as external estimates, not company guidance, and not independently verified here.
- FY2025 revenue
- US$79.0M (+158.9%)
- AI-native products rev
- US$53.1M (+143.4%)
- Open Platform / enterprise rev
- US$26.0M (+197.8%)
- Gross margin
- 25.4%
- IFRS net loss
- US$1,871.6M (non-cash-heavy)
- Adjusted net loss
- US$250.9M
- Overseas revenue mix
- >70%
- IPO raise
- ~US$620M (HK$4.82B), HK$165/sh, incl. US$350M cornerstones
Growth drivers
- AI-native consumer apps — Talkie (a top-ranked US AI app by 2024 downloads) and Hailuo video — US$53.1M FY2025 (+143.4%)
- Open Platform / enterprise-developer API on the M-series LLMs (M2/M2.5/M3) — US$26.0M FY2025 (+197.8%), the fastest-growing segment
- Video generation (Hailuo 2.x, 1080p) competing in the fast-growing text/image-to-video market
- Multimodal breadth (text+video+speech+music) enabling cross-sell within one platform
- Inference-efficiency architecture (Lightning Attention) aimed at lowering unit compute cost over time
- Global revenue mix (>70% overseas) insulating some revenue from domestic price wars
Bull & bear
A globally-reaching, full-stack multimodal AI franchise growing triple-digits, now public and capitalized, with rare Western consumer traction — if inference costs fall and the enterprise API compounds, today's tiny revenue base could invert the P/S over time.
- One of the few Chinese model makers with a proven overseas consumer hit (Talkie among top US AI apps by 2024 downloads) plus a credible video product (Hailuo) — distribution is the scarce asset
- Revenue +159% FY2025 with the enterprise/Open-Platform segment up +197.8%, showing the API flywheel engaging (Sacra pegs a ~US$300M run-rate by mid-2026 — unverified estimate)
- Gross margin doubled (12.2% -> 25.4%) in a year; efficiency architecture (Lightning Attention) offers a credible path to higher margins
- Fully funded post-IPO (~US$620M raised, US$1.05B cash at year-end) with marquee strategic + sovereign backers (Alibaba, Tencent, ADIA) to sustain the compute race
- Multimodal breadth = multiple shots on goal (video, companion, speech, music) rather than one modality
A ~US$79M-revenue company valued near US$13.5B (~170x sales) with an adjusted net loss of ~US$251M (~3.2x revenue), structurally low gross margins, compute access hostage to export controls, and a share price that has already round-tripped ~6x — a speculative growth-and-burn story priced for perfection.
- ~170x trailing P/S with negative earnings and negative book equity leaves zero margin for a growth stumble
- 25% gross margin caps profitability upside unless compute costs collapse; on an adjusted basis the loss is not narrowing
- Export controls on high-end Nvidia GPUs directly throttle the compute the model roadmap depends on
- Competes against OpenAI/Google/Meta and domestic giants (Alibaba Qwen, ByteDance Doubao, DeepSeek) that can subsidize inference toward zero
- Overseas revenue (its differentiator) is also its biggest single point of failure — Talkie's Dec-2024 US App Store removal already showed the tail risk
- HK$220 -> HK$1,330 -> HK$327 in months implies thin float and momentum-driven ownership, not fundamental support
What it is worth
Revenue-multiple / growth-narrative (earnings and book value are negative, so P/E and DCF are not meaningful today). Anchor: ~US$13.5B market cap on FY2025 revenue US$79.0M => ~170x trailing P/S; on Sacra's ~US$300M run-rate estimate => ~45x forward P/S (estimate, unverified); on low-hundreds-of-US$M analyst 2026 estimates => ~50-60x.
Growth decelerates, compute access tightens, or overseas app-store/regulatory risk bites; with no earnings support and ~170x sales, a large multiple compression is the natural outcome — the HK$1,330 -> HK$327 de-rating shows how far it can travel.
Growth continues (revenue toward low-hundreds of US$M in 2026) but margins stay thin and losses persist; the stock stays a high-beta, sentiment-driven vehicle re-rating on each print rather than compounding steadily. Fair value highly uncertain.
If the enterprise API compounds and the (unverified) ~US$300M run-rate carries into a multi-hundred-million 2026 with gross margin climbing well above 25% and the adjusted loss narrowing, ~US$13.5B could prove cheap on out-year sales — the multiple compresses as revenue catches up.
Valuation is a bet on the growth path, not current fundamentals — ~170x trailing P/S with a persistent adjusted loss (~3.2x revenue) and 25% gross margin. The debut close (HK$345, ~US$11.5B market value per TechNode) already embedded aggressive expectations; the post-listing spike to HK$1,330 and round-trip to ~HK$327 reflect speculative, thin-float dynamics more than re-rated fundamentals. US-first framing: this is context/analysis only, not a buy/own recommendation on a mainland-China-founded, HK-listed name.
SWOT
Strengths
- Rare global consumer distribution for a Chinese model maker — Talkie ranked among the most-downloaded US AI apps in 2024 and hit ~29M global MAU by end-2024; >70% of revenue is overseas
- Full multimodal stack (LLM + video + speech + music) rather than a single-modality point product
- Triple-digit revenue growth (+159% FY2025) with gross margin doubling (12.2% -> 25.4%)
- Deep-pocketed strategic backers (Alibaba, Tencent, HongShan, IDG, miHoYo) and a cash-raising public listing with ADIA as an IPO cornerstone
Weaknesses
- Large losses — adjusted net loss US$250.9M (~3.2x revenue); IFRS net loss US$1.87B; negative book equity at end-2025
- Low ~25% gross margin — compute-cost-heavy unit economics unlike typical software
- Tiny absolute revenue base (US$79M) against a ~US$13.5B market cap => ~170x trailing P/S
- Extreme share-price volatility (HK$220 to HK$1,330 in ~2 months) signals speculative, thin-float ownership
Opportunities
- Text/image-to-video is early and expanding fast; Hailuo has genuine product traction
- Enterprise/Open-Platform API is the fastest-growing segment (+197.8% FY2025)
- Inference-efficiency gains could lift gross margin materially from a low base
- Continued Western / SE-Asia consumer expansion where Talkie and Hailuo already show product-market fit
Threats
- US export controls on advanced GPUs (Nvidia) constrain training/inference compute access
- Brutal competition + price wars from far larger, better-capitalized players (OpenAI, Google, Meta; Alibaba, ByteDance, DeepSeek domestically)
- US-China app-store / data-scrutiny risk to overseas consumer revenue — already materialized: Talkie was pulled from the US App Store in Dec 2024
- Cash burn vs runway — adjusted loss ~3.2x revenue; the valuation demands near-flawless execution
Moats, dependencies & bottlenecks
Moats
Real download-chart presence abroad, but AI-companion/video switching costs are low, rivals are catching up, and Talkie's Dec-2024 US App Store removal shows the distribution is not fully in its control.
Harder to replicate than a single modality, but each modality faces a specialist (Runway/Sora video, ElevenLabs speech, Suno music).
Efficiency edges erode as the whole field advances; not a durable structural moat.
Alibaba/Tencent/ADIA funding buys runway, but capital is not a defensible product moat.
Dependencies
Advanced-GPU access is constrained by US export controls; domestic-chip substitution is unproven at frontier scale.
Compute is rented, so cost and availability directly set gross margin and training cadence.
>70% overseas revenue rides consumer apps exposed to app-store policy and US-China scrutiny; Talkie was removed from the US App Store in Dec 2024.
Capital & ecosystem Backers fund the burn and provide cloud/ecosystem ties; concentration and strategic-interest risk.
Frontier progress depends on data + shared research advances; IP/licensing exposure on training content.
Advantages
- Proven Western consumer traction rare among Chinese AI labs
- Triple-digit revenue growth off a fast-improving margin base
- Multimodal one-platform breadth enabling cross-sell
- Public listing + sovereign/strategic capital for the compute arms race
- Global revenue mix diversifying away from the domestic China price war
Weaknesses
- Nosebleed valuation (~170x sales) on a sub-US$100M revenue base with persistent losses
- Low ~25% gross margin and negative book equity
- Compute access dependent on a geopolitically contested supply chain
- Extreme, speculative share-price volatility and likely thin float
- No structural moat that survives well-funded incumbents subsidizing inference
Bottlenecks
- Access to frontier compute under export controls — the binding constraint on model quality and cost
- Gross margin ceiling from inference costs until efficiency or cheaper silicon arrives
- Cash burn vs runway despite the IPO raise (adjusted loss ~3.2x revenue)
- Monetizing consumer engagement (Talkie MAU) into durable, high-margin revenue amid app-store risk
- Geopolitical ceiling on overseas expansion — its main growth vector is also its main political risk
Top signals & trends
Top signals
Demand and monetization inflecting from a small base.
Unit economics improving, though still low.
Headline is alarming but largely non-cash (preferred-share fair-value); adjusted loss US$250.9M is roughly flat.
Momentum/float-driven, not fundamentally anchored.
Rare overseas traction, but removed from the US App Store Dec 2024 and monetization/retention unproven.
Priced for multi-year flawless compounding.
Trends
Directly benefits Hailuo; also the most competitive frontier (Sora/Veo/Kling).
Talkie is a leader, but monetization, safety, and app-store/regulatory scrutiny are rising.
Helps margins but also lets rivals subsidize toward zero, compressing pricing.
Constrains compute and threatens overseas consumer distribution.
Opens capital-markets access but invites intense comparison and volatility.
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.
Primary source of training/inference GPUs; access gated by US export controls.
Cloud/compute capacity and strategic-investor tie.
Upstream fab for the AI accelerators MiniMax's compute depends on.
>70% of revenue via Talkie companion app and Hailuo video subscriptions/in-app spend (AI-native products, US$53.1M FY2025).
Usage-metered access to M-series LLM, video, speech, music models (Open Platform, US$26.0M FY2025) — the fastest-growing segment; 214,000 enterprise customers/developers.
Frontier LLMs (GPT) + Sora video; the global scale/quality benchmark. Private.
Gemini LLMs + Veo video + full compute/distribution stack — most complete multimodal rival.
Llama open-weight models + massive consumer distribution and AI companions.
Qwen open models + cloud; both a MiniMax backer and a domestic competitor.
Ernie foundation models; established Chinese AI incumbent.
Hunyuan models + distribution; backer and rival.
Kling — a leading text-to-video generator competing with Hailuo. (HK-listed; context only.)
Chinese multimodal AI incumbent; MiniMax founder Yan Junjie is an ex-SenseTime executive. (HK-listed; context only.)
Fellow 'Six Tiger', listed on HKEX 8-Jan-2026 (first Chinese AI firm public); enterprise/foundation-model focus. Context only, not a buy call.
'Six Tiger' rival; long-context reasoning + agents. Private, China. Context only.
Ultra-low-cost open models pressuring pricing across China. Private. Context only.
Massive distribution + Doubao models + CapCut/Jimeng video. Private. Context only.
Video-generation specialist; direct Hailuo competitor. Private, US.
AI-companion app; Talkie ranked ahead of it in the H1-2024 US AI-app download chart, though Character.ai led Western MAU. Private, US.
Speech and music-generation specialists competing with MiniMax's Speech/Music models. Private.