Related News
Home » Opinion » China Knowledge
In the startup frenzy to develop AI, the elephant in the room is money
Print Edition
NOT long ago, I found myself in a conference room with a group of wide-eyed entrepreneurs, fresh from raising pre-A round of financing for their AI startup. One of them, a friend, asked me along to advise on their communications strategy.
Their demonstration of what they’ve achieved in application of artificial intelligence was dazzling: drafting legal memos, summarizing earnings reports in five languages, even spinning out passable poetry. Brimming with confidence, they seemed unstoppable.
Then we came to the elephant in the room: money.
Like many swept up in the foundation-model gold rush, their company shines on demo day but can’t foot the bill for graphics processing units (GPUs) that power AI.
Welcome to the world of AI in 2025, a paradox wrapped in hype. On paper, the sector looks promising. Venture capital keeps flowing. PhDs are trading tenure for stock options. Algorithms dominate everything from Go boards to molecular biology.
In reality, most AI startups are torching cash faster than they can raise it. Even the giants aren’t immune: US-based OpenAI reportedly burns through hundreds of thousands of dollars a day just to keep ChatGPT running. AI isn’t a fad but building a profitable business around it is far harder and costlier than most care to admit.
The math is brutal. Training a state-of-the-art model can cost over US$10 million. Running it at scale? Tens of millions more annually. In China, top-tier Nvidia’s GPUs can be rented for up to US$20 an hour, leaving startups with monthly computing bills in the millions of yuan. Elite engineers command salaries on par with star athletes.
Meanwhile, revenues lag far behind.
This isn’t just a Silicon Valley problem. In China, stars like Zhipu AI, MiniMax, Baichuan Intelligence and 01.AI collectively raised more than 20 billion yuan (US$2.8 billion) last year, backed by tech giants and state-linked funds. Yet profitability remains elusive. Zhipu, maker of so-called generalized linear models (GLM), has corporate clients but wrestles with crushing computing costs. MiniMax’s chatbot Glow boasts tens of millions of users, but only a sliver of revenue. Even Kai- Fu Lee’s 01.AI, valued north of US$1 billion, lacks a clear monetization path despite open-sourcing its models to cut expenses.
The root problem is simple: Flashy demos rarely translate into dollars.
Consumers love free AI tools, just as they once downloaded every new photo filter or meditation app. Converting curiosity into steady revenue is a harder sell. Subscriptions are a tough ask when tools feel interchangeable. Everyone wants AI’s magic; few want to pay for it.
Business clients aren’t an easy answer. They demand more than novelty: bulletproof reliability, airtight security and seamless integration with legacy systems that often resemble spaghetti than software stacks. Closing deals takes months of customization, audits and compliance hurdles.
For AI startups, that means acting less like software shops and more like boutique consultancies — expensive, slow and hard to scale. Baichuan Intelligence, for instance, is chasing finance and government clients but faces long sales cycles and regulatory scrutiny, not unlike Palantir embedding AI into industrial workflows.
If the products were indispensable, investors might wait. But much of the industry still chases novelty over necessity. A chatbot writing Japanese-style haiku may amuse millions, but it won’t convince a chief financial officer to approve a seven-figure cloud bill. Businesses want AI that clears hospital billing backlogs, flags fraud in the making or keeps factory lines humming. Those tools exist but require tedious fine-tuning, regulatory approvals and deep domain partnerships. Too many startups, dazzled by elegance, skip the grind.
Even in China, where momentum is strong, survival will favor doers over dreamers. At the recent World Artificial Intelligence Conference in Shanghai, two new alliances were announced. One links China’s leading large language model developers with AI chipmakers, engaging Huawei, Biren, Enflame, Moore Threads and StepFun; the aims to drive AI into industrial transformation, uniting players such as SenseTime, MiniMax, Metax and Uluvatar CoreX.
These coalitions can gather China’s top talents to build infrastructure and frontier applications capable of rivaling US efforts. But not everyone will make it. Companies peddling slick slide decks and lofty promises, banking only on another funding round, will be culled. The industry needs less theater and more substance.
The way forward is clear: Deliver measurable value. Viral demos and freemium gimmicks might grab attention, but they won’t pay seven-figure cloud bills. The only sustainable route is proving impact — cutting costs, boosting revenue or unlocking new capabilities. When an AI tool demonstrably saves a hospital millions or stops fraud, firms will be willing to spend money.
Over the next two decades, AI will shift from spectacle to backbone. Its first act dazzled with TED talks and sci-fi dreams. The next will be quieter but more transformative. AI won’t dominate the spotlight; it will hum beneath it, powering industries from medicine to manufacturing. Less “god in the machine,” more “AI+” — tools that augment, not overshadow, the systems we already trust.
For investors, entrepreneurs and users, the message is simple: the glitzy party’s over. Now comes the grind. The winners won’t be those with the flashiest demos or catchiest slogans, but those who make AI not just smart, but solvent, and in doing so, make our industries and lives meaningfully better.
In business as in life, brilliance is nice, but a bottom line in the black is essential.
(The author is an adjunct research fellow at the Research Center for Global Public Opinion of China, Shanghai International Studies University, and founding partner of 3am Consulting, a consultancy that specializes in global communications.)
- About Us
- |
- Terms of Use
- |
-
RSS - |
- Privacy Policy
- |
- Contact Us
- |
- Shanghai Call Center: 962288
- |
- Tip-off hotline: 52920043
- 沪ICP证：沪ICP备05050403号-1
- |
- 互联网新闻信息服务许可证：31120180004
- |
- 网络视听许可证：0909346
- |
- 广播电视节目制作许可证：沪字第354号
- |
- 增值电信业务经营许可证：沪B2-20120012
Copyright © 1999- Shanghai Daily. All rights reserved.Preferably viewed with Internet Explorer 8 or newer browsers.
