Goldman Sachs Keeps Buy on MiniMax as ARR Doubles in Two Months, M3 and Hailuo 3 Set to Test Monetization Ceiling
Goldman Sachs reaffirms a HK$1,000 price target on MiniMax Group (0100.HK) — implying 25% upside from current levels — after the Hong Kong-listed AI model company disclosed that April annualized recurring revenue surged 100% versus February, with API revenue now commanding more than half of total sales, a structural shift that analysts say validates the company's bet on open-platform commercialization.
The revenue mix inflection is the sharpest signal yet that MiniMax is successfully converting model capability into repeatable enterprise cash flows. Open-platform and API revenue climbed from roughly 30% of the total at end-2025 to above 50% by May 2026 — a near-doubling in share within five months — compressing the relative weight of consumer-facing products including Hailuo AI, MiniMax Agent, and social AI app Xingye. Goldman Sachs analysts Ronald Keung, Steve Qiu, and Damian Xie, who hosted MiniMax management at the GS Asia Communacopia + Technology Conference on May 19, described the ARR trajectory as consistent with the company's own year-end guidance, without quantifying a specific target.
Markets have already begun pricing in the acceleration: at HK$800.00 per share as of May 18, MiniMax carries a market capitalization of HK$250.9 billion (approximately US$32.0 billion), with three-month average daily turnover of HK$1.7 billion (US$213.2 million) — liquidity that reflects institutional conviction in a name Goldman upgraded to Buy as recently as May 4, 2026.
Upcoming M3 and Hailuo 3 Launches Redefine the Pricing Playbook
The near-term catalyst calendar is dense. MiniMax expects to release a smaller-parameter M3 model within weeks, followed by a larger-parameter variant within months, with both versions targeting step-change improvements in coding and agentic task execution. For video generation, the company has penciled in a June launch for Hailuo 3 — a roughly eight-month cadence following the Hailuo 2.3 release in October 2025 — built on a native understanding-generation architecture that management explicitly benchmarks against Seedance 2.0's design philosophy.
The strategic significance extends beyond product novelty. Management indicated that the 3-series model generation will serve as the primary lever for price increases across API tiers. This matters for margin arithmetic: Goldman Sachs currently estimates MiniMax's text API gross margin at approximately 40%, already double the 20–25% range it attributes to industry peers. Multi-modal API margins are estimated even higher, at 60–70%. If the M3 models deliver the intelligence uplift management anticipates, the firm sees a credible path to further gross margin expansion — a rare combination of volume growth and margin accretion in a sector where most players are still burning compute to buy market share.
Cost Architecture Gives MiniMax a Structural Edge Over Cloud-Dependent Rivals
Underpinning the margin advantage is an infrastructure strategy that diverges meaningfully from the asset-light defaults of most Chinese AI startups. MiniMax operates self-owned computing clusters and expects the majority of its computing infrastructure to be self-operated by year-end 2026 — under an arrangement where third parties own the physical hardware but MiniMax retains operational control, preserving flexibility while reducing unit inference costs.
The company positions itself as one of the largest pure-play AI firms in China by secured computing power scale, a claim that carries weight in a market where GPU access remains a binding constraint amid ongoing US export controls. Critically, MiniMax has completed software adaptations for major domestic chips and targets full domestic inference capability by the third quarter of 2026. That milestone would substantially insulate the company from further supply-chain disruptions and reduce its exposure to the US-China technology decoupling risk that Goldman explicitly flags as a key downside scenario.
Revenue Ramp Narrows Losses on Paper, But Profitability Remains a 2028-Plus Story
Goldman's financial model reflects a company in deliberate investment mode. Revenue is forecast to rise from US$79.0 million in fiscal 2025 to US$300.0 million in 2026E, before accelerating to US$880.1 million in 2027E and US$2.47 billion in 2028E — a compound growth rate that would rank among the fastest in global software. Yet EBITDA losses are projected to widen in absolute terms through 2027, from negative US294.7 million in 2025 to US454.5 million in 2027E, before narrowing to negative US$227.1 million in 2028E as scale economics begin to assert themselves.
The loss trajectory reflects a conscious choice to front-load compute and talent investment. Free cash flow yield turns modestly negative in 2026E at -0.8% before deepening to -1.4% in 2027E, suggesting the company retains sufficient runway but will require disciplined capital allocation to avoid dilutive fundraising. Goldman's EPS estimates of -US$1.31 for both 2026E and 2027E imply the market is effectively valuing MiniMax on a revenue-multiple basis, with profitability visibility a 2028 event at the earliest.
Global TAM Ambition Anchors the Bull Case Despite Geopolitical Overhang
Goldman's Buy thesis rests on four pillars: comprehensive multi-modal product coverage spanning text, code, video, and agentic workflows; demonstrated commercialization capability evidenced by the API revenue inflection; per-token cost advantages supported by the self-operated infrastructure model; and organizational efficiency that the analysts argue is atypically high for a Chinese AI startup of this scale.
The bear case is equally well-defined. Model performance could disappoint relative to OpenAI, Google DeepMind, or domestic rivals including Zhipu AI and Moonshot AI. The path to profit visibility remains long. And geopolitical risk — particularly any escalation in US technology export restrictions targeting Chinese AI infrastructure — could disrupt both the domestic chip rollout timeline and international expansion plans for Hailuo AI and Talkie, which have accumulated meaningful user bases outside China.
For now, the ARR doubling in two months and the imminent launch of two flagship model families give bulls a concrete near-term scorecard. The June Hailuo 3 release, in particular, will serve as the first real-world test of whether MiniMax's native understanding-generation architecture can close the quality gap with leading Western video models — and whether enterprise customers will pay the higher token prices that management is counting on to fund the next leg of growth.
Related Coverage:
MiniMax Unveils M2.5 AI Model, Enabling Hour-Long Agent Tasks at $1 Cost