On June 22, Zhipu’s market capitalization soared past HKD 1 trillion.

On the same day, Tencent’s market cap stood at approximately HKD 3.94 trillion, Alibaba’s at HKD 1.97 trillion, Xiaomi’s at HKD 612.9 billion, Meituan’s at HKD 444.5 billion, JD.com’s at HKD 302.9 billion, Ctrip’s at HKD 229.3 billion, and Kuaishou’s at HKD 196.9 billion.

A company founded less than seven years ago—still unprofitable, with annual revenue just exceeding RMB 700 million—now commands a market valuation surpassing that of China’s most profitable internet giants.

Capital markets are placing a tangible bet: the “gold mines” marked on old maps are losing value; the outlines of a new continent are emerging, and pricing power is changing hands.

Most market analyses of Zhipu focus on its technological edge. The GLM series has excelled in evaluations, its open-source strategy has won developer loyalty, enterprise clients are expanding, and official recognition in technical competitions on Weekend X has further fueled its surge.

The waves are active, but they don’t dictate the tide’s direction.

Tencent reported net profits exceeding RMB 200 billion last year, with a price-to-earnings ratio below 15. Alibaba’s market cap is even lower than the combined value of its stake in Ant Group and its core e-commerce business. Capital markets are revaluing internet giants ruthlessly: their existing assets are depreciating.

Meanwhile, a large model-native company with RMB 700 million in annual revenue and still in the red has secured a trillion-dollar valuation.

Capital is clearly betting not on intrinsic value but on the disruptive potential of business models: large models’ ability to replace search, matching, content creation, and code writing will systematically erode the core profit pools of internet giants over the next decade.

Of course, many argue Zhipu’s trillion-dollar valuation is merely a “greater fool” gamble—a market impulse driven by tech hype. This skepticism isn’t entirely unfounded—a money-losing company is hard to justify at such a valuation, no matter how you dissect its cash flow.

But the “greater fool” label may obscure a deeper shift in valuation logic: when the moats of internet giants develop structural cracks due to large models, market share transfers become highly probable. The market isn’t pricing Zhipu’s present but the timeline for large models to devour old profit pools.

The giants see this trend but are trapped.

Tencent’s WeChat ecosystem relies on advertising revenue. When users stop scrolling through Moments and official accounts, instead asking AI assistants directly, WeChat’s commercial foundation begins to crumble.

Alibaba’s e-commerce empire is built on the “search-compare-purchase” chain. When AI assistants handle product selection and price comparisons for users, Taobao and Tmall’s traffic gateways become obsolete.

Even ByteDance, the most aggressive in AI investment, faces similar challenges: Douyin retains users through recommendation algorithms, but when AI can directly generate desired content, the algorithm’s moat is bypassed.

The three broadest moats of the internet era—social connections, e-commerce matching, and content recommendation—are quietly being outflanked by large models: you know where the enemy is coming from, but your walls face the wrong direction.

At this moment, internet giants confront a deeper dilemma: if large models render existing business models obsolete, will they remain the same companies even if they win (or fail to win) the large model race?

Tencent’s core asset has never been WeChat ads but “connection”—WeChat connects people, official accounts connect people and content, and Mini Programs connect people and services. Advertising, payments, and game distribution are all byproducts of connection.

In the large model era, interaction logic is shifting from “connection” to “agency.” Users no longer open Mini Programs or browse official accounts; they simply tell AI assistants, “Book me a flight” or “Write my weekly report.” AI assistants directly invoke services, generate content, and complete tasks, pushing connection into the background.

An intermediary layer now stands between users and Tencent. The more powerful this layer becomes, the farther Tencent drifts from users. The more users rely on AI assistants, the lower WeChat’s strategic value as a “connector” becomes. Tencent faces a crisis of purpose: when your mission is connection, but the era declares connection no longer scarce, who are you?

Alibaba faces two racing curves. One is Alibaba Cloud: large model training and inference require massive compute power. The stronger Tongyi Qianwen becomes, the greater API call volumes grow, securing cloud infrastructure revenue. The other is Taobao and Tmall’s ad revenue: when AI assistants handle product selection and price comparisons, information asymmetry shrinks, weakening merchants’ willingness to buy traffic.

This won’t happen overnight, but the direction is clear. Alibaba straddles both curves, which move independently. The true test of its transformation is how fast Alibaba Cloud can grow and whether it can establish a new growth pole before e-commerce ad revenue hits a ceiling.

ByteDance’s dilemma is subtler. Despite its AI-first facade, struggles persist. Douyin’s recommendation algorithm relies on precise capture of user behavior—what they watch, how long they linger, whom they like. Large models are shifting recommendation logic from “behavioral data” to “semantic understanding.” They no longer need users to watch 1,000 videos to learn preferences; a single statement like “I’m in a bad mood today” suffices. Behavioral data barriers may collapse before semantic understanding matures.

Moreover, while Doubao series performs well in evaluations, ByteDance cannot abandon its most profitable recommendation ad business. Large models remain confined to “auxiliary recommendation” roles within ByteDance, unable to touch core commercial logic. The reason is obvious: this isn’t a matter of courage but company DNA. A firm built on “behavioral matching” cannot easily pivot to “semantic understanding.”

The three giants, three types of confusion, share a common underlying logic: the core asset of the large model era is “intelligence,” while the internet era’s core asset was “data.” Data accumulates, but intelligence is universal. No matter how much you accumulate, a stronger foundational model can instantly surpass you.

This paradigm clash repeats throughout business history. Kodak invented digital cameras but locked them away, fearing they would kill film. Nokia developed touchscreens first but clung to physical keyboards.

Every technological paradigm shift forces old-era giants to oscillate painfully between “protecting existing profits” and “embracing the new paradigm.”

The cruelty of the large model era is that it offers no decade-long transition. GPT-3 to trillion-parameter models took under three years; Kodak struggled for two decades between digital camera invention and bankruptcy. The window is visibly narrowing, yet the steering wheel remains locked to old-era profit logic.

Zhipu secured its trillion-dollar valuation precisely because it lacks these burdens. It need not protect ad revenue, worry about e-commerce platform interests, or defend recommendation algorithm moats. It can relentlessly allocate all resources to model advancement and developer ecosystem building.

Internet-era winners—Tencent, Alibaba, ByteDance—first amassed massive C-end user bases, then monetized through ads, e-commerce, and gaming. Technical capabilities served business models.

Large model-native players like Zhipu take a different path: first establish a generational advantage in model capabilities, attract developers, build ecosystems, and then seek monetization. OpenAI and Anthropic follow this model, securing sky-high valuations through technical leadership before clear business models emerged, then gradually building commercialization loops via enterprise services and developer payments.

Whether this path works in China remains unanswered.

A company with RMB 700 million in annual revenue and heavy losses supporting a HKD 1 trillion market cap must prove three things in the coming years: model capabilities can sustainably approach the global top tier; API and enterprise service revenues can scale rapidly; and high R&D investment can eventually yield higher margins and stronger repurchase rates. Otherwise, today’s valuation cannot survive on narrative alone.

Yet the market has already priced it at HKD 1 trillion. This at least signals one thing: investors believe the large model era’s business models will not organically grow from old-era profit pools. They require a company unburdened by history to cultivate them anew.

China’s internet companies’ identity crisis ultimately stems from invalidated coordinate systems. Capital markets sense this first.

The crack between the two coordinate systems is the collective strategic confusion of China’s internet companies. Everything measured by old coordinates is depreciating, while new coordinates are held by a few. Most still stand on old shores, gazing at distant masts, unsure whether to set sail.

Zhipu’s trillion isn’t one company’s victory but the silence of countless others. An old map folds away as a new one unfurls. Old-era giants are stranded ashore, watching others’ masts vanish beyond the horizon. No one can tell them whether sailing leads to a new continent or deeper seas.