MiniMax Faces Lock-Up Cliff as Pricing Misstep Exposes Structural Fault Lines in China AI Race
MiniMax, China's AI model startup that went from zero to IPO in four years, is now fighting a three-front war — a user trust crisis triggered by an unannounced price hike, a structural profitability gap between its consumer and enterprise arms, and a July lock-up expiry that could flood the market with shares equivalent to nearly 10 times its current free float.
The convergence of these pressures has been brutal for shareholders. As of June 24, 2026, MiniMax's Hong Kong-listed shares had surrendered HK$267.6 billion in market capitalization since their March 18 peak, leaving the company valued at HK$149.8 billion — roughly one-sixth of domestic rival Zhipu AI, whose market cap stood at HK$969.3 billion on the same date. For context, Zhipu's stock surged as much as 47.6% intraday on June 15 alone, following the release of its GLM-5.2 model. The valuation chasm between the two companies is no longer just a number; it is a verdict on strategic timing.
Stealth Price Hike Destroys Developer Trust, Sends Stock to Record Low
The immediate trigger for MiniMax's stock rout was self-inflicted. On June 1, 2026, the company launched its flagship M3 model with technically impressive credentials: per-token compute cost reduced to one-twentieth of its predecessor, prefill speed up 9.7 times, and native multimodal capability paired with a 1-million-token context window. Founder Yan Junjie publicly claimed M3 "surpasses GPT-5.5."
Yet simultaneously — and without prior notice — MiniMax raised its API pricing across the board and restructured its Token Plan subscription tiers. The entry-level monthly subscription jumped from RMB29 (approximately $4.03) to RMB49 (approximately $6.81), a 69% increase. More disruptively, the billing model shifted from per-call pricing to per-token consumption, and a new weekly usage cap was imposed on top of existing session limits.
The developer community reacted immediately. Users who had specifically chosen MiniMax for its reputation as a high-volume, low-friction platform — colloquially known in developer circles as "all-you-can-eat" — found their effective usage capacity slashed to less than one-third of prior levels within the same five-hour window. One developer noted that a single hour of coding work now consumed what previously took five hours of quota. Annual Plus subscribers filed refund requests, citing breach of service terms.
Markets moved in lockstep with user sentiment. MiniMax shares fell 15.71% on June 1, the largest single-day decline since its IPO. The company issued an apology that evening, grandfathering existing users out of the weekly cap and offering new users a 50% bonus quota. On June 8, it introduced a permanent 50% discount on API pricing. Neither measure arrested the slide. The stock continued falling, breaking through the HK$400 level on June 12, before beginning a tentative recovery on June 15.
The episode crystallized a concern that had been building among institutional investors: MiniMax lacks the pricing power to extract premium margins from its technology, at least at this stage of the competitive cycle.
Financials Reveal a Business Running on Consumer Volume, Not Enterprise Margins
MiniMax's 2025 annual results, filed ahead of its IPO, illustrate the structural imbalance at the core of its business model. Total revenue reached US$79.04 million (approximately RMB534 million), up 159% year-on-year. Adjusted net loss (excluding approximately US$1.6 billion in financial liability fair-value charges related to the IPO process) came in at US$250 million, up only 2.7% year-on-year against the 159% revenue expansion.
On a reported basis, however, the net loss attributable to shareholders was US$1.872 billion (approximately RMB 12.651 billion), a threefold year-on-year deterioration — a headline figure that continues to dominate investor perception.
More telling is the margin split by segment. Consumer-facing AI products — Talkie, Hailuo AI, and related applications — generated over 67% of total 2025 revenue but carried a gross margin of just 4.7% through the first three quarters of 2025. The B2B API and open-platform business, which accounted for less than 40% of revenue, posted a gross margin of 69.4%. Revenue from the enterprise segment grew from US$8.72 million in 2024 to US$25.96 million in 2025, a near-tripling, but the absolute base remains modest.
MiniMax COO Yun Yeyi disclosed in late May 2026 that total users have exceeded 300 million globally, with enterprise and developer clients surpassing 1 million, and that enterprise revenue now roughly matches consumer revenue on a run-rate basis. If accurate, that would represent a significant compositional shift — but the gross margin differential means the strategic imperative is unambiguous: enterprise and API monetization must scale faster, or unit economics will remain structurally challenged.
Coding Pivot Arrives Late, But the Race Is Not Over
The M3 launch was not merely a product update; it was a strategic repositioning. MiniMax has placed AI coding capability at the center of its enterprise commercialization push, a pivot that Yan Junjie acknowledged came later than it should have.
Speaking at a MiniMax technical forum on June 14, Yan recalled asking DeepSeek founder Liang Wenfeng two years ago whether he planned to build AI coding tools. The answer was no — the addressable market of Chinese developers was estimated at only 1 to 2 million people, deemed too narrow. Yan admitted he shared that view at the time.
The market has since proved that assessment wrong. Following Anthropic's aggressive reinforcement of coding capabilities in its Claude model family, enterprise buyers globally demonstrated high willingness to pay for verifiable, ROI-positive coding automation. Liu Zhiyong, chairman and chief scientist of Zhenji Intelligence, framed the dynamic precisely: "Coding has become the primary battleground for large model commercialization. It is the only scenario that simultaneously possesses high payment willingness, verifiable correctness, and direct ROI."
Zhipu AI and Moonshot AI pivoted to this narrative earlier. MiniMax began its M2 series in 2025, but the primary focus remained multimodal consumer applications. The result is a perception gap in developer communities, where MiniMax is still widely viewed as a follower rather than a leader in coding benchmarks. That perception gap translates directly into the HK$819.5 billion market-cap spread between MiniMax and Zhipu as of June 24.
The company's "10x Team" recruitment initiative, launched in May 2026, signals an attempt to accelerate enterprise co-development across vertical industries. MiniMax's argument is that its multimodal application-layer experience gives it differentiated user-scenario understanding that can be embedded into foundational model capabilities — a bottom-up advantage that pure infrastructure players may lack. Whether that thesis resonates with enterprise buyers remains to be demonstrated in benchmark data.
Lock-Up Expiry and A-Share Ambitions Define the Next Sixty Days
The near-term test for MiniMax is structural, not just operational. In early July 2026, the company's IPO lock-up period expires. The restricted shares represent approximately 63% of Hong Kong-listed equity, with financial investors — as distinct from strategic or founder shareholders — holding more than one-third of that total. Current free float is estimated at approximately 5% of total shares outstanding. Post-expiry, available supply could increase nearly tenfold.
Whether the stock can absorb that supply depends entirely on whether MiniMax can demonstrate a credible commercialization inflection before the unlock date. Bai Wenxi, vice chairman of the China Enterprise Capital Alliance, told local media that improvements in MiniMax's coding benchmark scores would be interpreted by Hong Kong investors as a signal that a monetization turning point is approaching, providing a basis for valuation recovery.
In parallel, MiniMax signed an IPO counseling agreement with CITIC Securities on May 29, 2026, initiating the process for a potential STAR Market listing — an "A+H" dual-listing structure. A June 17 announcement that STAR Market would expand its eligibility criteria for hard-technology companies added market excitement to the prospect. A successful A-share listing would provide a second capital channel to fund model training costs and extend the cash runway ahead of a potential profitability inflection.
Bai cautioned, however, that STAR Market regulators have tightened scrutiny of unprofitable technology companies, requiring demonstrable commercial pathways rather than technology narratives alone. For MiniMax, that means the M3 model's coding performance metrics in the second half of 2026 are not merely a product story — they are the primary variable in both its stock price defense and its A-share listing eligibility.
Related Coverage:
MiniMax Faces Triple Threat: Pricing Backlash, Benchmark Doubts, and a July Unlock