Li-Ning Brand and Curry Brand Reach Long-Term Cooperation 李宁品牌与库里品牌达成长期合作

Li-Ning signs Curry, CITIC Securities discusses lithium batteries... but the real undercurrents are surging in the gaps of the trending lists. While everyone is buzzing about brand collaborations and industry cycles, ByteDance’s AI assistant “Doubao” is set to launch its paid services in late June. This is the heavyweight move in the summer 2026 AI chess game.

Free has always been the most expensive trap. Large language models have burned through money for three years—from the showy demos in labs to the red ocean of applications. Now, it’s finally time to confront the financial statements. Doubao is not the first to go paid, and it certainly won’t be the last. Its choice is extremely pragmatic: rather than directly forcing subscription fees on users, it aims to “accelerate integration with Douyin e-commerce.” This means the AI assistant is being forced to transform from a cost center into a precise traffic and conversion engine. It’s no longer just an “encyclopedia” for answering questions; it must become a “shopping guide” to drive sales. This pivot may lack grace, but its business logic is starkly clear—the endgame of generative AI might just be becoming the next era’s super advertising space and the binding agent for closed-loop transactions.

This warning bell is also ringing on the other side of the ocean. Silicon Valley giants, who once propelled their stock prices to dizzying heights with AI hype, have quietly begun restricting internal employees’ AI token usage. How ironic! On one hand, they paint a limitless future of AI-driven productivity revolution in public markets; on the other, they tighten their belts internally due to spiraling token costs. This reveals an awkward reality: as AI transitions from “demos” to daily use, its operational costs expand exponentially. So-called “intelligence” is, at its core, a war of attrition burning through computing power and capital. When even the giants themselves are feeling the pinch, can we expect them to offer free lunches indefinitely? The market is already voting with its feet.

This pressure is now rippling up the industry chain. Why are capital-favored unicorns like Zhipu AI and MiniMax suddenly rushing to list on the A-share market in 2026? The answer is starkly written on their balance sheets. Financing in the primary market is tightening, and valuation myths require liquidity to sustain them. More crucially, once the narrative is told, an IPO is needed to lock in returns for early investors and stockpile resources for the “war of attrition” ahead. The AI startup race has entered its second half—shifting from competing on technical specs to competing on survival capabilities.

Looking back domestically, Doubao’s shift toward monetization and commercialization may herald the exploration of a more arduous survival model. As the capabilities of general large models become homogenized, simple API calls and subscription fees can no longer support massive training and inference costs. ByteDance’s choice is to deeply integrate AI capabilities into its vast content and e-commerce ecosystem, transitioning from “selling tools” to “selling results.” This might be a bloody path, or it might be a narrow one, because it demands that AI deliver quantifiable commercial returns, rather than just being “somewhat useful.” Will this make AI assistants increasingly utilitarian, more like a pushy salesperson than the pure, intelligent companion we once imagined?

The entire industry is undergoing a collective disenchantment. From the frenzy of the “hundred-model battle” to each player now seeking its own path to monetization, capital has become stingy and cautious. CITIC Securities’ optimistic forecast for the lithium battery industry is underpinned by the logic of capacity clearance and demand recovery in the real economy. In contrast, the AI industry’s logic is the reverse: it is painstakingly seeking sustainable business models amid disorderly overcapacity (a flood of models and applications). Doubao’s paid model, Silicon Valley’s restrictions, and the IPO rush of unicorns all speak to the same theme: after the technological hype fades, the gravity of commerce begins to manifest.

We have finally reached a more realistic, albeit harsher, juncture. AI is no longer an infinite sci-fi story to dream about; it must face the cold arithmetic of cost, revenue, and profit. The end of the free era marks a necessary, if unromantic, beginning of the industry’s maturation. The future winners may not be the models with the highest benchmark scores, but the practitioners who first strike the balance among “technology, cost, and value” and inspire ordinary users to willingly pay for it. In this long-distance AI race, the supply line is the true lifeline.

李宁签下库里，中信券商谈锂电……但真正的暗流，涌动在热榜的缝隙里。当所有人都在谈论品牌联名和行业周期时，字节跳动旗下“豆包”AI助手即将在6月下旬启动付费，这才是2026年夏天AI棋局里落下的一枚重子。

免费，从来都是最昂贵的陷阱。大模型烧钱烧了三年，从实验室的炫技烧到应用层的红海，现在终于到了必须直视财务报表的时刻。豆包不是第一个付费的，也绝不会是最后一个。它的选择极其务实：不直接向用户强索订阅费，而是“加速打通抖音电商”。这意味着，AI助手正从一个成本中心，被迫转型为精准的流量和转化引擎。它不再只是回答问题的“百科全书”，而要成为带货的“导购员”。这一转身，优雅尽失，但商业逻辑清晰得可怕——生成式AI的终局，或许就是成为下一个时代的超级广告位和交易闭环的粘合剂。

这记警钟，也敲在大洋彼岸。硅谷那几位靠AI概念把股价推上云霄的巨头，最近开始悄悄限制内部员工的AI Token使用量。多讽刺！一边在公开市场描绘着AI生产力革命的无限前景，一边却在内部因为Token成本超支而勒紧裤腰带。这揭示了一个尴尬的现实：当AI从“Demo”走向日常，其运营成本是指数级膨胀的。所谓“智能”，本质上是一场持续燃烧算力和资金的消耗战。巨头们自己都心疼了，还能指望他们无限期地为公众提供免费午餐吗？市场正在用脚投票。

这股压力，正沿着产业链向上游传导。智谱AI、MiniMax这些被资本追捧的独角兽，为何在2026年突然着急回A股？答案赤裸裸写在资产负债表上。一级市场融资趋紧，估值神话需要流动性来支撑，更关键的是，故事讲完了，得用上市来锁定早期投资人的回报，并为下一阶段“持久战”储备粮草。AI创业进入了下半场，从拼技术参数，到拼生存能力。

回看国内，豆包的付费化与商业化，可能预示着一种更艰难的生存模式探索。当通用大模型的能力趋于同质化，单纯的API调用和订阅费已无法支撑庞大的训练和推理成本。字节的选择，是将AI能力深度植入其庞大的内容与电商生态，从“卖工具”转向“卖效果”。这或许是一条血路，也可能是一条窄路，因为它要求AI必须产生可量化的商业回报，而不是停留在“有点用”的层面。这会不会让AI助手变得越来越功利，越来越像一个喋喋不休的销售，而非那个我们想象中的、纯粹的智能伙伴？

整个行业正在经历一场集体去魅。从百模大战的狂热，到如今各自寻找变现的路径，资本开始变得吝啬而谨慎。中信证券对锂电行业的乐观预测，背后是实体经济产能出清与需求复苏的逻辑；而AI行业的逻辑恰恰相反，它是从无序的产能过剩（模型和应用泛滥）中，艰难地寻找可持续的商业模式。豆包的付费，硅谷的限制，独角兽的上市，都在诉说同一个主题：技术热潮褪去后，商业的重力场开始显现。

我们终于走到了一个更现实、也更残酷的节点。AI不再是那个可以无限畅想的科幻故事，它必须面对成本、收入和利润的冷酷算术题。免费时代的落幕，恰恰是行业走向成熟的、虽不浪漫却必要的开端。未来的赢家，可能不是参数跑分最高的模型，而是那个最先找到“技术-成本-价值”平衡点，并让普通用户心甘情愿为其买单的实践者。这场AI长跑，补给线才是真正的生命线。