Zhipu: A Long-Term Owner's Perspective

Hunting ten-year five-baggers among great growth stocks — pressing the upside question: "Can it get much bigger?"

• QHow high is its market ceiling? Is it enlarging an existing pie, or creating an entirely new market?7/10

  A

  The ceiling is very high, and closer to "creating a brand-new market," but that is exactly a double-edged sword: the higher the ceiling, the more today's price is prepaying for the most optimistic scenario in which "Zhipu ultimately dominates this new pie."

  First, the pie itself. The report cites Frost & Sullivan data: China's AI market grew from RMB 93.7 billion in 2022 to RMB 160.7 billion in 2024, and is expected to reach RMB 993 billion by 2030; within that, China's large-model market is expected to reach RMB 101.1 billion by 2030, a CAGR of about 63.5% over 2024–2030. This is a classic high-growth early-stage track, and long-term demand is not the problem. From this angle, general-purpose large models are indeed not fighting over an already-divided existing pie (replacing incumbent systems the way traditional ERP or databases do), but growing a "model-as-a-service" new market that did not exist before on top of generative AI. The report discloses that Zhipu's models have already empowered more than 8,000 institutional clients and cover about 80 million devices, and official results materials show its MaaS API platform's call volume grew about 400% over the past twelve months and ARR rose to about RMB 1.7 billion, all showing that demand is newly created rather than transferred from elsewhere.

  But "creating a new market" has two sides when tested against Baillie Gifford's ten-year-five-bagger bar. The upside: TAM is far from saturated, and the theoretical growth space is order-of-magnitude. The deadly side: a new market has no established profit-distribution pattern, and who dominates and how the profit pool concentrates are entirely undetermined today. The report puts it bluntly: the top five have a small revenue-share gap, about 9.4% for the leader, about 6.6% for Zhipu, and about 6.4%, 6.1%, 4.7% for the next few; this means the profit pool "has not become highly concentrated, but is still in a stage of multiple strong players coexisting and competing fast on price and performance." More critically, the real opponents are not fellow startups like SenseTime and MiniMax, but big-platform players that come with cloud, traffic, ecosystem, and customer relationships, and strong model providers like DeepSeek that disrupt pricing with low price and high performance. Zhipu's own 2025 blended gross margin slid from 56.3% to 41.0%, and verified financial data also confirm that "the track is getting bigger while unit prices are being pushed down by competition."

  Conclusion: on the ceiling dimension, Zhipu stands in a genuinely huge and brand-new market, which is the foundation of its story; but a "big market" does not equal "big returns." Whether Zhipu can become a major carver of this new pie has no answer today, while the current price has already priced in a yes.

  Jun 11, 2026
• QOver the next five years, can its revenue at least double? Is the growth driven mainly by volume, price, or new businesses?7/10

  A

  Almost certainly it can double, and very likely more than double: this is one of Zhipu's few strengths beyond doubt. The real question is not "can it double" but "can unit economics turn positive while it doubles," and the latter is far harder than the former.

  First, the certainty of the growth itself. The report and verified financials agree: revenue went from RMB 57.4 million in 2022 to RMB 124.5 million in 2023, RMB 312.4 million in 2024, and RMB 724.3 million in 2025 (about +131.9% YoY), a three-year CAGR of about 133%. Starting from the still-small base of RMB 724 million, layered on the track's 60%+ industry CAGR, the bar to double revenue over the next five years (only about 15% annualized) is genuinely low; at Zhipu's current size and momentum, tripling to quintupling is within a reasonable range. Baillie Gifford would score this highly.

  Next, the drivers: Zhipu has all three of volume, price, and new business. Volume is the main engine: official materials show its MaaS platform's token call volume grew about 400% over the past twelve months, API/Agent revenue grew about 292.6% YoY, and cumulative enterprise users and developers reached about 4 million; the report separately discloses 240,000 paying Coding Plan developers and 400,000 paying Claw Plan and product users, a classic usage-driven ramp. Price also shows a rare positive signal: the report records an 83% API price increase in Q1 2026, a counter-trend hike in an industry generally locked in a price war. New business is the structural shift from on-premise deployment (RMB 534 million in 2025, 73.7%) to cloud API/Agent (RMB 190 million, 26.3%), where the cloud curve, closer to recurring revenue, is growing markedly faster.

  But the subtext of this Baillie Gifford question is "a quality doubling." Here comes the cold water: Zhipu's growth is "the more it grows, the more cash it burns," not "the more it grows, the more cash it produces." Verified data show 2025 R&D of about RMB 3.18 billion (note: the report body and table once recorded it as "about RMB 4.1 billion"; on the company's results-announcement basis it should be RMB 3.18 billion, +44.9% YoY), still more than four times the RMB 724 million revenue; blended gross margin over the same period fell from 56.3% to 41.0%. The report also makes clear that net operating cash flow deteriorated all the way from -RMB 68.2 million in 2022 to -RMB 2.2449 billion over 2022–2024. In other words, I am highly confident revenue will double, but in the foreseeable future it is more likely to be accompanied by losses and cash burn scaling up in step, rather than self-funding. It can double, but not a "good doubling," and this is the layer that most needs to be distinguished in Zhipu's growth narrative.

  Jun 11, 2026
• QFive years from now, what will take over as the next growth engine? Does this "second curve" exist today?4/10

  A

  The second curve is already sprouting today, but it looks more like "an extension of the same main curve" (from selling models to selling Agents/platforms) than a truly independent new growth pole that can take over after the core business peaks; and neither curve has yet been proven to make money.

  Zhipu's current revenue consists of two parts: on-premise deployment (RMB 534 million in 2025, 73.7%, project-based, tied to the client's infrastructure and acceptance milestones) and cloud deployment (RMB 190 million, 26.3%, closer to API/Agent usage-based billing). If today's "main curve" is the larger but heavier, less repeatable on-premise deployment, then the one with the most second-curve character is the cloud MaaS/Agent platform: it grows fastest, with official materials showing cloud API/Agent revenue up about 292.6% YoY, MaaS platform ARR rising to about RMB 1.7 billion, and token call volume up about 400%. The report also pins its hopes here: "If Zhipu wants to widen its moat, it must prove... that the Agent/MaaS platform can form a scale network." This is a real, rapidly growing second leg.

  But strictly speaking, this is more a form-upgrade of the main business than a "second curve." A second curve in the Baillie Gifford sense is ideally a new growth source relatively orthogonal to the existing business, able to independently step up when the existing engine slows (like AWS for Amazon, or Services for Apple). Zhipu's cloud platform and on-premise deployment are both built on the same GLM model capability underneath, highly homologous on the demand, technology, and competition sides: what big-platform players and cheap open-source models hit at once is precisely this same capability. Once "pure model-capability leadership" is commoditized (the report explicitly warns that open-source models, low-cost inference, and domestic-chip solutions are all accelerating this process), both curves come under pressure together, rather than one shielding the other.

  A more realistic concern: it is too early to talk about a second curve taking over when even the first curve's profit model has not been proven. Verified financials show cloud gross margin, though rebounding sharply from 3.3% to 18.9%, is still far below the level of mature software companies; blended gross margin instead fell to 41.0%. Candidates that could become a true second curve are overseas markets (the report says the API/Agent platform already covers 218 countries and regions) and open-ecosystem plays like GLM-4.6, which is adapted to Cambricon and Moore Threads domestic chips; but these are still traffic and ecosystem positioning today, far from contributing meaningful, profitable, independent revenue. Conclusion: the seeds of a second curve are visible, but it is neither independent enough nor yet proven able to self-fund, and is far from a successor card already in hand.

  Jun 11, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?4/10

  A

  The core advantage is "technology and first-mover status," but that is exactly the kind of moat most easily eroded by time; over the next three to five years, the technical-lead layer will very likely narrow as open-source and domestic chips spread, and whether it can widen the commercial moat (switching costs, scale network) in step is the unresolved deciding factor.

  First, the advantage does exist. The report records: Zhipu launched the GLM framework in 2021 and in 2022 open-sourced China's first ten-billion-parameter-scale model, GLM-130B, and is a front-rank player: by 2024 revenue it is China's largest independent large-model company and China's second-largest general-purpose large-model company, with a market share of about 6.6%; as of 2025 its API/Agent platform covers 218 countries and regions, and deep users of its services include "9 of the top 10 internet giants." This is not a model company that "nobody uses"; its technical brand and first-mover recognition are real assets.

  But what Baillie Gifford really asks is "will this moat widen or narrow over three to five years," and Zhipu's current moat components happen to be the least durable. After a line-by-line breakdown, the report's judgment is clear-eyed: on switching costs, after an enterprise does deep API integration, on-premise deployment, and industry customization there are indeed migration costs, but they are "far less solid than ERP, databases, or payment networks"; on pricing power, the 83% API price increase in Q1 2026 is a highlight, yet verified financials show cloud gross margin is still only 18.9% and blended gross margin instead fell to 41.0%, so it looks more like "a bargaining improvement from a local product upgrade" than solid pricing power; on cost advantage, the results chart claims "domestic hardware deployment cuts cost by 50% with no loss of performance," but the financials have not yet proven it has translated into an industry-leading margin.

  The narrowing pressure is structural. The report notes: open-source models, low-cost inference, and domestic-chip solutions are all advancing, and will commoditize "pure model-capability leadership" faster, and Zhipu itself is accelerating this: GLM-4.6 was open-sourced under the MIT license in September 2025 and adapted to Cambricon and Moore Threads domestic chips. Open-sourcing is an above-board play to grab ecosystem share, but its flip side is: when top-tier capability can be obtained for free and run on cheaper domestic compute, the scarcity of "the model itself" as a moat is diluted by its own hand. What could truly widen the moat are the two things the report says are still in progress: the Agent/MaaS platform forming a scale network effect, and platform revenue's gross margin and cash flow improving markedly with scale. If both are achieved, the moat upgrades from "technical leadership" to "platform lock-in"; if not, in three to five years it will be just a model service provider with decent technology but squeezed by open-source and big platforms and unable to raise prices. On current evidence, I assign a higher probability to "narrowing" than to "widening."

  Jun 11, 2026
• QIf its core business is disrupted, does it have the gene to reinvent itself? How does it treat mistakes and bad news?4/10

  A

  The gene to reinvent itself, in the narrow sense of "fast iteration and the courage to tear down and rebuild," Zhipu very likely has; and its attitude toward mistakes and bad news, judging by the quality of prospectus disclosure, is fairly candid. But be clear-eyed: this company has never truly gone through a test of "its core business being disrupted and facing a life-or-death crisis," so the so-called reinvention gene is for now more an inference than a verified fact.

  First, the Baillie Gifford supplementary premise of "self-reinvention." Large models are a field where the paradigm may shift every few months, and a company that survives in the top tier must, at its core, have the ability to keep overturning its own previous-generation products. Zhipu's evidence here is positive: from the 2021 GLM framework and the 2022 open-sourcing of GLM-130B to the September 2025 release and open-sourcing of GLM-4.6 with early adaptation to Cambricon and Moore Threads domestic chips, its technical cadence has never fallen behind; facing DeepSeek disrupting the industry with low price and high performance, and facing compute constraints from being added to the U.S. Entity List, it chose to proactively pivot to an open-source ecosystem and domestic-compute adaptation rather than cling to its old playbook. This flexibility of "change the playbook as soon as the environment changes" is an essential survival skill for a startup-type AI company, and Zhipu has it.

  But the true touchstone of the "reinvention gene" is "when the core business is disrupted," and Zhipu has not met that moment so far. All its fast iteration today is still adding to an upward growth curve; it has never gone through a crisis of its core business being completely overturned by rivals, forcing it to cut off a limb to survive and redefine the meaning of its existence. Compared with companies that truly "died once and came back to life" in a recent industry upheaval, Zhipu's resilience still rests largely at the inferred level of "it should have it." On a track with strong cycles and strong technical substitution, this stress-untested gene cannot be scored directly as a cemented advantage.

  Next, "how it treats mistakes and bad news." Here the report's assessment leans positive: in the prospectus the company "disclosed fairly fully the U.S. Entity List, the EAR, compute dependence, equity incentives, losses, and the no-dividend policy," and does not avoid discussing core risks. On major bad news like being added to the Entity List, the company disclosed it truthfully and explained its response logic (not building its own data centers, pivoting to procuring compute from domestic cloud service providers) rather than glossing over it. But the report also leaves a caution: so far capital allocation is hard to call excellent, and the post-listing acquisition of a persistently loss-making office-property target (total consideration up to about RMB 360.52 million) is a move that is "not necessarily wrong, but clearly does nothing to strengthen the AI moat," testing not integrity but whether management can maintain discipline under high valuation and high pressure. On balance: candidly facing bad news earns points; but "whether it can successfully reinvent after being disrupted" can only be an open question today, and no credit can be advanced on it.

  Jun 11, 2026
• QDo management (especially the founders) have a long-term horizon with interests deeply bound to the company? Are they willing to sacrifice current profit for five to ten years out?6/10

  A

  On interest alignment and long-term vision, Zhipu clearly qualifies: the founding team holds control with real money, and the company has proven, through the concrete actions of "no dividend, pour all money back into R&D," that it is willing to sacrifice current profit for the distant future. What really deserves a discount is not "whether it is willing to sacrifice current profit for five to ten years out" but "whether the capital-allocation discipline of that sacrifice is rock-solid."

  First, interest alignment. The report discloses it concretely: Zhipu is a classic founder-team-led company, with Liu Debing, Tang Jie, Li Juanzi, Xu Bin, Zhang Peng and entities controlled by Liu Debing such as Beijing Lianpai, Huihui, and Zhideng forming a concert-party controlling-shareholder group; after the IPO, Liu Debing directly holds about 8.97%, Beijing Lianpai about 7.73%, Zhideng about 6.18%, and Tang Jie about 6.10%, with the other founders holding additional direct or platform stakes, and employee holding platforms totaling about 15.15%. This is not a structure where "professional managers pocket shareholder capital empty-handed"; both founders and employees have their personal wealth staked in the equity. Baillie Gifford would score this relatively highly.

  Next, long-term vision and "sacrificing current profit for the distant future." Here Zhipu is almost a textbook case of "daring to sacrifice the present": the company pays no dividend and states clearly it will not pay cash dividends for the foreseeable future; verified financials show 2025 R&D of about RMB 3.18 billion, more than four times the RMB 724 million revenue, willing to bear a RMB 4.718 billion statutory net loss to pour money into models and compute. Net IPO proceeds were about HK$4.1734 billion, of which 70% goes to general large-model R&D, 10% to the MaaS platform, 10% to ecosystem/strategic investment, and 10% to working capital, almost entirely toward the future rather than current returns. On the trait Baillie Gifford values most, "willing to sacrifice current profit for five to ten years out," Zhipu is not insufficient but rather "sacrifices" quite thoroughly.

  But the points to deduct are in the "quality" and "discipline" of capital allocation, not "willingness." The report gives a neutral-to-conservative assessment: first, large-scale losses and equity-incentive dilution are real shareholder costs; willingness to burn cash does not equal burning it efficiently, and there is no visible inflection today where these investments are turning into distributable cash flow. Second, and more noteworthy, after listing the company announced the acquisition of a target company that mainly holds office property and has been persistently loss-making for the past two years, for a total consideration of up to about RMB 360.52 million (explicitly not using IPO proceeds, paid with internal resources and bank financing), which the report judges as "not necessarily wrong, but clearly not the kind of capital-allocation move that immediately strengthens the AI moat." This is exactly what Baillie Gifford watches: a team that truly allocates capital on a ten-year scale would put every cent into reinforcing core competitiveness, rather than being distracted by optimizing asset structure. On balance: alignment and long-term orientation are clear pluses, but capital-allocation discipline is still to be verified, so this question is "full marks on willingness, execution pending."

  Jun 11, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and free of reliance on harming society and regulation?5/10

  A

  "If it disappeared tomorrow, how much would customers miss it?" They would miss it, but it is not irreplaceable; that is exactly the problem: in a market surrounded by open-source models and big-platform players, and where Zhipu itself is open-sourcing, it is far from "indispensable." And the social and regulatory sustainability of its growth model has, running across it, an objectively existing and potentially escalating geopolitical risk (the U.S. Entity List), an exogenous constraint in its story that it cannot dismantle itself.

  First, "indispensability." Zhipu does have real, deep customer usage: the report discloses that its models have empowered more than 8,000 institutional clients and cover about 80 million devices, that deep users of its services include "9 of the top 10 internet giants," and that it has 240,000 paying Coding Plan developers and 400,000 paying Claw Plan and product users, with official materials showing cumulative enterprise users and developers of about 4 million. Enterprise clients that have done deep API integration, on-premise deployment, and industry customization do face migration friction, and these clients would genuinely miss it. But the intensity of "missing it" is constrained by the abundance of substitutes: the report is clear that this barrier is "far less solid than ERP, databases, or payment networks," while the market has strong low-price, high-performance models like DeepSeek, big-platform players with their own cloud and ecosystem, and plenty of open-source options. Ironically, Zhipu itself is reducing its own indispensability by hand: GLM-4.6 is open-sourced under the MIT license and adapted to Cambricon and Moore Threads domestic chips: when your top-tier capability can be downloaded for free and run on cheaper domestic compute, the "it must be you" quality is naturally diluted. So the honest answer is: if it disappeared, customers would be inconvenienced and would migrate, but they could find substitutes and business would not grind to a halt.

  Next, the second layer of this Baillie Gifford question: "is the growth model sustainable and free of reliance on harming society and regulation?" On this layer Zhipu is broadly healthy: its growth comes from selling AI capabilities to enterprises and developers to lift productivity, not from harming users, regulatory arbitrage, or negative social externalities, and through open-sourcing it has positive spillovers for the industry ecosystem. But there is one unavoidable hard constraint: in January 2025 the U.S. added Zhipu and its subsidiaries to the Entity List, recorded in both the Federal Register document and Reuters. The company states in its prospectus that this has not yet caused a material adverse effect (because it does not build its own data centers and procures compute from domestic cloud service providers), but the report also stresses that "this remains an objectively existing and potentially escalating long-term geopolitical risk." This means its "social/regulatory sustainability" is not entirely in its own hands: export controls, AI content compliance, data, and other regulatory variables are all exogenous gates on its growth path that could tighten at any time. On balance: customers would miss it but have substitutes, the growth model itself is legitimate and sustainable, but layered with a geopolitical constraint it cannot dismantle itself, Zhipu does not pass the double gate of "indispensable + sustainable" easily.

  Jun 11, 2026
• QHow are this business's unit economics (gross margin, incremental returns)? Do they improve or worsen at scale? Where does the money earned go?3/10

  A

  Unit economics are currently weak, and getting worse in direction: the signs of "better" at scale are only local (cloud gross margin rebounding, API price hikes), while overall gross margin is instead declining and incremental input far exceeds incremental output. The money earned (and the money raised) is almost entirely poured into R&D and compute, and has not yet turned into distributable cash flow. This is one of Zhipu's hardest weaknesses.

  First, gross margin and unit economics. Verified financials show 2025 blended gross margin of 41.0%, down 15.3 percentage points from 56.3% in 2024; the report further gives a longer trajectory: 54.6% (2022) → 64.6% (2023) → 56.3% (2024) → 41.0% (2025), i.e., two consecutive years of decline after the 2023 peak, "not yet forming a mature SaaS-style high margin overall." Structurally, the 73.7%-share on-premise deployment is project-based, tied to milestones and acceptance, more a solutions business than high-repeat subscription; the 26.3%-share cloud is more "subscription-like," but the price war and inference costs are heavy: cloud gross margin was once only 3.3% in 2024, rebounding to 18.9% in 2025, improving in direction but still far below excellent software companies in absolute level. This is a mixed picture of "local improvement, overall weakening," not a clean scale effect.

  Next, incremental returns (how much comes back per extra dollar invested). This is currently negative, with a stark gap. 2025 R&D was about RMB 3.18 billion (note: the report body and table once recorded it as "about RMB 4.1 billion"; on the results-announcement basis it should be RMB 3.18 billion, +44.9% YoY), more than four times the RMB 724 million revenue; the report says bluntly that this means the company "is still in the 'the more it grows, the more it must invest' stage, not the 'the more it grows, the more cash it produces' stage." On cash it is even more direct: the report records net operating cash flow deteriorating from -RMB 68.2 million in 2022 and -RMB 648 million in 2023 to -RMB 2.2449 billion in 2024; the report's conservative estimate of 2024 Owner Earnings is about -RMB 2.3 billion, not "very low" but significantly negative. The report also honestly marks that the complete 2025 cash flow statement was not obtained from its directly verifiable sources, so 2025 Owner Earnings is deliberately left "unknown" and not fabricated. But "R&D at four times revenue + persistent large negative historical operating cash flow" is enough to judge: today, every unit of scale expansion swallows more cash than it earns back.

  Finally, "where does the money earned go." The answer: there is basically no "money earned" to distribute; what is spent is the money raised, and almost all of it goes to reinvestment rather than returning to shareholders. The report discloses the company pays no dividend and will not pay cash dividends for the foreseeable future; of net IPO proceeds of about HK$4.1734 billion, 70% goes to R&D, 10% to the MaaS platform, 10% to ecosystem/strategic investment, and 10% to working capital; the prospectus also discloses that one core reason for the sharp rise in 2024 R&D was about RMB 1.241 billion more in compute service fees paid to third-party compute providers, and this business is "neither asset-light nor cheap, and highly dependent on an external compute ecosystem." On balance: current unit economics are weak, incremental returns are negative, and cash is all reinvested into the future; scaling currently amplifies losses, not profits. To score this question higher, one must first wait for hard evidence of cloud gross margin rising and stabilizing and of an inflection in operating cash flow, none of which has arrived today.

  Jun 11, 2026
• QFor it to rise fivefold in ten years, what conditions must hold simultaneously? Are they realistic? What expectations does today's share price imply?2/10

  A

  For it to rise another fivefold from today's price, a string of extremely optimistic conditions must hold "simultaneously," and the current price has already prepaid a good part of them as established facts, so the realism of rising fivefold again is very low, and today's price implies the most optimistic script of "the king of the next decade."

  First, lay out the implicit premise of this Baillie Gifford question: rising fivefold in ten years means multiple gears must mesh together: (1) revenue keeps an extremely high CAGR on an already RMB 724 million base (the report's optimistic scenario sets it at 60%); (2) year-10 Owner Earnings margin jumps from today's significantly negative to the 15%–20% level of a mature software company; (3) the commercial moat is strong enough to fend off big-platform players and low-cost open-source models, lifting cloud gross margin from 18.9% into a healthy range and holding it; (4) valuation multiples do not revert substantially to the industry norm. The report's judgment is direct: "if even two or three of these fail to materialize, this investment is prone to failure," and none of these four is easy, so having them all hold "simultaneously" is multiplying several low-probability events together.

  Next, "what expectations does today's share price imply?" This is the crux of the question, and where that string of conditions is nailed to the ground. The report ran a three-tier Owner Earnings discount based on the company's 2025 revenue, assuming 10-year revenue CAGR of 35%/45%/60%, year-10 Owner Earnings margin of 10%/15%/20%, and discount rate of 12%/11%/10%, arriving at equity value of roughly "conservative within RMB 2.2 billion, base within RMB 22.5 billion, optimistic within RMB 125 billion," equivalent to per-share magnitudes of about HK$5–10, HK$40–70, and HK$250–350. The problem is the market price is far above all three tiers: the report records a snapshot price of HK$1,040, market cap of about HK$463.68 billion, and a Reuters P/S of about 554.55x, already clearly above the upper edge of the report's most optimistic scenario (HK$250–350). And more extreme still: after the report's snapshot the share price kept surging; according to a June 7, 2026 Sina Finance report, Zhipu was up about 15x from its HK$116.2 IPO price, quoted at about HK$1,421 that day, touched nearly HK$1,993 intraday, and its market cap approached HK$900 billion on May 29 (against RMB 724 million revenue, the P/S has risen to the thousand-times order, higher than the report's snapshot of 554x). In other words, today's price prices in not just "the optimistic scenario coming true" but "even more optimistic than the optimistic scenario, and almost no mistakes within ten years."

  A cross-comparison makes the extremeness of the pricing even clearer: the report records SenseTime's P/S at about 13.56x as of mid-May 2026 and 4Paradigm's on the order of 3–4x, with Zhipu's P/S more than 40 times SenseTime's and on the order of a hundred times 4Paradigm's; even granting that Zhipu is "a purer large-model play with a stronger story," the report judges this is no longer an "excellent-company premium" but more like a "prepaid price for a distant winner-take-all."

  Conclusion: the string of conditions needed for a five-bagger in ten years each requires extreme success on its own, and multiplied together the realism is very low; yet the current price not only leaves no safety cushion for "the conditions falling through" but prices in the most optimistic script in advance. The report's "worth re-studying" range is HK$40–80 with operating cash flow clearly turning positive, and the order-of-magnitude gap between that and the current price is itself the plainest footnote to "today's share price implies expectations that are too high."

  Jun 11, 2026
• QWhy has the market not yet realized all this? Is it that it cannot understand, looks down on it, or cannot see far enough? What will become the "narrative inflection point"?2/10

  A

  This question must be asked in reverse for Zhipu: Baillie Gifford's classic "why has the market not yet realized this stock's greatness" presupposes "an undervalued, overlooked growth stock"; but Zhipu's situation is exactly the opposite: the market has not "failed to realize" but has already over-realized and priced in the most optimistic script in advance. So the honest answer is: what the market currently undervalues is not its potential but its risk; what has genuinely not been fully priced in is the fact that "it is still a cash-burning company far from self-funding."

  Why say the market is "running too hot" rather than "cannot understand / looks down on it / cannot see far"? The evidence is in the price itself. The report's snapshot shows a market price of HK$1,040 and a Reuters P/S of about 554.55x, already more than 40 times SenseTime's (about 13.56x) and on the order of a hundred times 4Paradigm's (about 3–4x); after the snapshot it surged further, and according to a June 7, 2026 Sina Finance report, Zhipu was up about 15x from its HK$116.2 IPO price, quoted at about HK$1,421 that day, touched nearly HK$1,993 intraday, and its market cap approached HK$900 billion on May 29. For a stock up more than tenfold in half a year with P/S charging to the thousand-times order, there is clearly no issue of "the market not realizing it is good." The market's pricing of the "story" is ample and even excessive: it understands the track, respects the company, and sees far enough; what it systematically underestimates is the other side the report repeatedly stresses: free cash flow has not been established (operating cash flow deteriorated from -RMB 68.2 million to -RMB 2.2449 billion over 2022–2024), 2025 R&D of about RMB 3.18 billion far exceeds the RMB 724 million revenue, competition is extremely intense, and there is almost no margin of safety in valuation.

  So "what will become the narrative inflection point" (the most crucial supplementary premise of this question)? The inflection is not "the market one day suddenly discovers how good it is" but "the market one day rediscovers how fragile it is," with possible triggers including:

  First, the most concrete and nearest-term catalyst is the lock-up expiry. According to the aforementioned June 7 report, Zhipu faces a share unlock in July, with the lock-up expiring around July 8 and involving about 11.6% of shares, and early investors (Meituan, Alibaba, Tencent, Hillhouse, etc.) with paper gains over 15x have strong incentives to sell down. When very high paper gains meet a release of tradable float, it is the most realistic supply-demand inflection.

  Second, fundamental falsification. The report lays it out clearly: if revenue keeps growing fast but operating cash flow stays significantly negative, days sales outstanding keeps lengthening (already stretched from 23 days in 2022 to 112 days in H1 2025), cloud gross margin fails to approach the level of excellent software companies, and the company still needs financing or dilution to sustain R&D, then the valuation paradigm could "switch from 'distant platform king' back to 'cash-burning technical service provider.'" The report warns that this switch, compounded by cash flow that keeps failing to turn positive, makes a drawdown on the order of 70%–90% "not unimaginable."

  Third, exogenous shocks. An escalation of the U.S. Entity List, tightening of AI content/data regulation, or a rival like DeepSeek once again resetting industry pricing with low price and high performance could all become triggers for a sentiment reversal.

  On balance: applying this Baillie Gifford question honestly to Zhipu, the answer is: the market does not "fail to understand how good it is" but "has not yet fully priced in its risk and valuation fragility"; today it looks more like what the report calls "a high-odds but extremely-high-valuation long-dated option," with downside risk far greater than upside odds. For an investor whose primary concern is permanent loss of capital, this is precisely the kind of "information edge" to avoid, not to bottom-fish.

  Jun 11, 2026