The “AI Oprah” Era is Over

For years, Z.ai (formerly Zhipu AI) operated like a benevolent AI Oprah, tossing out model releases, 4.5, 4.6, 4.7, with the casual generosity of “you get a model, you get a model, everybody gets a model!” That approach built a massive developer ecosystem: 45 million developers worldwide, 12,000 enterprise customers, and 80 million end-user devices running GLM models across public governance, industrial manufacturing, finance, and education sectors.

But public companies have fiduciary duties, not philanthropic missions. As one developer forum comment summarized: Once shareholders and big money are involved, they are not the same company that produced 4.5, 4.6, and 4.7. The math is stark, Z.ai’s prospectus reveals the company loses money on its GLM Coding Plan even before accounting for marketing or R&D. With 70% of IPO proceeds earmarked for research and development, investors will demand returns, not just releases.

The discounting strategy that fueled growth, annual plans at 60% off plus referral credits totaling 75-80% discounts, has created a user base accustomed to near-free access. But as one analyst noted, the segment of users hosting models locally is too small to matter commercially. The real money lies elsewhere.

Built on Open Source, Now Trapped by It

Z.ai’s technical foundation is impressive. Its GLM-4.7 model ranks first among open-source models on CodeArena’s blind testing with a million users, surpassing GPT-4, and tops the AA AI Index among domestic Chinese models. The architecture supports over 40 domestic chips, a strategic necessity after the US Commerce Department added Z.ai to its Entity List in January 2025, claiming the company worked with China’s military.

This “full domestication” approach reflects China’s technological self-reliance drive, but it also means Z.ai can’t easily leverage the most advanced Western semiconductor technology. Instead, it’s optimized for cost-effectiveness, with compute costs for its coding API at about 1/10 of cloud alternatives while delivering 90% performance.

The open-source strategy built trust and adoption, but as CEO Zhang Peng acknowledged in Bloomberg interviews, the company must now balance openness with profitability. The GLM series, China’s first 10-billion-parameter model, first open-source 100-billion-parameter model, first conversational model, and first multimodal model, created a moat of goodwill that may prove harder to monetize than anticipated.

The Sanctions Paradox

Being placed on the US Entity List should have been a death sentence for a company requiring cutting-edge chips. Instead, Z.ai turned constraint into strategy. The company now maintains an “open attitude regarding the use of chips” focused on cost-effectiveness, betting that domestic alternatives will improve while prices drop.

This pivot reveals a deeper resilience in China’s AI ecosystem. Z.ai’s prospectus shows revenue of 312.4 million yuan in 2024, with the Chinese government as a primary customer for on-premise deployments citing national security concerns. While US sanctions limit access to advanced semiconductors, they also create a captive domestic market where foreign competitors can’t easily play.

The Hong Kong listing serves as both financial gateway and geopolitical statement. As Zhang Peng told SCMP, “As an international city, Hong Kong serves as an excellent gateway and bridgehead, enabling us to do more in the internationalisation space.” The city provides access to global capital while operating under Beijing’s sphere of influence, a delicate balance for a company navigating US-China tech tensions.

The Profitability Mirage

Here’s where the controversy gets spicy: Z.ai is currently unprofitable, and its path to profitability requires fundamentally changing the behavior it encouraged. The company has grown its cloud-based revenue tenfold, but lower pricing means it contributes minimally to total sales. Meanwhile, deep discounts have trained users to expect cheap or free access.

The IPO changes this equation. With rival MiniMax also filing for public offering, the “AI tiger” cohort faces pressure to demonstrate business models that work. Z.ai’s strategy involves shifting from custom B2B deployments to standardized products, targeting 50%+ growth over three years with cloud-based clients driving the majority.

But can you convert a community built on free access into paying customers? The company’s own data suggests caution: GLM Coding Plan has 150,000 global users, but many are likely hobbyists or researchers with limited willingness to pay. The real revenue comes from Chinese government contracts and enterprise clients with data sovereignty concerns, not the open-source community that amplified Z.ai’s reputation.

What This Means for AI’s Future

Z.ai’s IPO represents more than one company’s public debut. It validates an alternative AI development path: state-influenced, domestically-sourced, and commercially pragmatic. While Western labs like OpenAI and Anthropic debate openness versus safety, Chinese labs face a different tension: openness versus solvency.

The market’s 13.2% first-day premium suggests investors believe Z.ai can thread this needle. But developer communities are fickle. If GLM-5 arrives with restrictive licensing or premium pricing, users can fork existing open versions or migrate to alternatives. The moat isn’t the model, it’s the ecosystem and the government relationships that ensure steady enterprise revenue.

For AI practitioners worldwide, Z.ai’s listing serves as a case study in how geopolitics shapes technology. US sanctions intended to slow China’s AI progress instead accelerated domestic chip optimization and created a parallel ecosystem. Hong Kong’s role as financial intermediary shows how capital finds ways to fund innovation despite political barriers.

The question isn’t whether Z.ai will survive, it’s whether the open-source AI movement can survive commercial success. Every AI lab eventually faces the same choice: serve the community or serve the shareholders. Z.ai just made that choice very publicly, and the entire industry is watching to see what breaks first, the business model or the ideals that built it.

The takeaway: AI development is splitting into parallel tracks. Western labs optimize for capability and safety debates, Chinese labs optimize for self-reliance and commercial viability. Z.ai’s IPO proves both paths can attract capital, but only one has to answer to quarterly earnings calls. For developers who built their projects on free GLM models, the party’s over. The question is whether they’ll be invited to the new, paid party, or left outside looking in.