42 Chapters AI Industry and Infrastructure: Capabilities, Delivery, Capital and Value Allocation
This article is a source-constrained thematic compilation: it preserves the frameworks, disagreements and questions open to further verification from 42 Chapters interviews conducted between 2023 and 2026. References to "at the time" must not be rewritten as "now"; figures on operations, financing, performance and markets that have not been independently verified are not presented as statistical facts, and no securities trading advice is provided on that basis. For company engineering cases, see the related hardware/frontier research articles.
I. Breaking down one grand narrative into five propositions
Improved model capabilities, user adoption of applications, customer willingness to pay, corporate profitability, and high capital valuations are five distinct things. In March 2024, Mo Jielin split the narrative into Scaling Laws, competitor positioning, and application explosion: heating up in the first two does not automatically deliver the third. Discussions around DeepSeek and Manus in March 2025 further noted that model contributions and application delivery cannot be evaluated on the same axis; third-party models do not render applications valueless, but packaging an application launch as a foundational model breakthrough inflates false expectations. [S55][S26]
This article frames the causal chain as:Changes in capability boundaries → Tasks become completable → Delivery is sufficiently reliable → Sustained usage and payment → Returns are distributed across the value chain. Capital investment can support every link in advance, and can also price in expectations before actual returns materialize; therefore, stock prices or financing fervor are not suitable acceptance criteria for completion of the chain.
At the end of 2023, Mo Jielin distinguished actionable practices from external conditions using industry "bright lines" and "dark lines" such as interest rates, inventory and demand; in 2026, Zhu Ning further emphasized that real technology value, corporate profitability and valuation reasonableness are mutually independent. Knowing that technology is advancing does not resolve issues of price, timing and bearing consequences. His four conditions—"new concepts, liquidity, policy support, inexperienced participants"—are explicitly necessary but not sufficient, and cannot be turned into a formula for calling a top once all boxes are ticked. [S29][S50]
A November 2025 dialogue with Mo Jielin further laid out maturity mismatches: the long-term buildout of data centers and power capacity, short-term application revenue, and quarterly market expectations are not synchronized. One should map the funding chain from end customers → applications → inference/training procurement → hardware/power, marking payments, commitments, timelines and gross margins, and avoid double-counting repeated circular cooperation announcements. Speaker identification in that episode was severely inaccurate; the positions that Scaling Laws had "failed" versus "not failed but hard to evaluate" are opposing views and cannot be synthesized into a claim that pretraining is over. [S17]
II. Stage narratives are useful, provided stages are not treated as a calendar
| Interview timing | Key observations/predictions at the time | Test conditions to retain |
|---|---|---|
| 2023, Zhang Jinjian and Chen Yu | Early-stage teams may leverage resources to complete learning; models, middle layers and applications carry multiple layers of value | Technical feasibility, team accumulation and customer demand must each be confirmed separately; one cannot look only at financing [S46][S48] |
| Early and mid-2024, Qu Kai | Where controllability is insufficient, sell outcomes first / constrain tasks; going all-in on learning differs from immediately committing all capital | Manual rework, full-chain delivery and cash burn; a small closed loop does not equal large-scale PMF [S3][S24] |
| 2024, Wang Hua | Three layers: interaction restructuring, complex task automation, AGI | Confidence and evidence differ across layers; a good experience at the first layer cannot be extrapolated to full autonomy [S8] |
| Early 2025, Qu Kai | Agents move from information services to executing outcomes; multimodality moves toward participation | Billable outcomes, liability for failure, procurement trust and real completion [S28] |
| August 2025, Mo Jielin | Shift from a narrative of model intelligence to Token consumption and industrialization | Separate training and inference demand, trial revenue from stable PMF [S52] |
| 2026, Zhang Jinjian and Qu Kai | Strong reactions from big tech, cold application financing, growing model momentum | Continuous demand, iterative implementation; financing stories cannot be passed off as completed technology [S11][S7] |
These are not a sequential, definitive industry chronology, but observations from different coverage circles. The 2024 claim that "generation beats recommendation" was an extreme thought experiment that explicitly excluded factors such as community effects; the 2026 line that "applications are gone" was a rhetorical question, not a statistic on application counts. Comparing them should prompt asking which constraints changed and which demand remains, rather than retroactively enlisting every new event as having been predicted all along. [S3][S7]
In August 2024, Qu Kai used the Character.AI incident to reason toward "model commoditization, the model wars nearing an end," only to observe in 2026 that model momentum had grown stronger; this is precisely the evolution of views that should be preserved, rather than deleting the earlier prediction to stitch together a narrative of continuous correctness. His CAI analysis also distinguished user value, product operations, self-developed models and exit outcomes: partial personnel and licensing arrangements do not equal an equity acquisition of the entire company; whether an exit allows some people to profit also does not prove product profitability or the failure of companionship demand. Character psychology, regulatory motives and transaction amounts discussed in the episode were not independently verified. [S39][S7]
III. Model training, data and costs: look at the objective function, and also at operating conditions
In 2025, Zhu Yibo of StepFun explained Infra as the combined system of compute, communication, storage, scheduling and frameworks; the same model can have different time-to-first-token, output speed, stability and per-task costs. High throughput with a long wait for the first token is not necessarily suitable for chat. Small-parameter models are not necessarily cheap either, as communication, memory access, operators and hardware adaptation can change the bill. [S36]
Training and inference should not be lumped together as "compute demand." In August 2025, Mo Jielin argued that pretraining demand is cyclical while inference depends on application usage, and that large-scale RL generation may in turn increase training-side demand. That was the explanation at the time and does not mean the two must necessarily rise together going forward. Zhu Yibo used decoding to illustrate objective reordering: RL feedback also depends on large volumes of outputs, and inference optimization affects training iterations; optimizing MFU, training speed and launch costs can still conflict. The right question is "which bill are you optimizing for," not which metric should always be maximized. [S52][S36]
In 2026, Meng Fanqing further put data value back into the closed loop: correct formatting, appropriate difficulty and absence of cheating are only initial filters; data is effective only if target tasks improve after training without old capabilities degrading. Simulated environments such as Feishu/Notion must first behave consistently with real environments, and scoring must also be checked; otherwise, more trajectories may amplify false progress. The example of 99 one-minute traces and one 100-minute trace shows how tail tasks slow down an entire batch; data, algorithms and scheduling jointly determine effective throughput. [S33]
Self-improvement is also layered: fixing outputs, modifying harnesses, generating training data and training new models are not the same achievement. Vertical tasks may have strong models build processes while smaller models execute them, but quality, privacy, cost and maintenance still require real-world testing. Meng Fanqing disagrees that prices will necessarily trend toward zero; Wang Hua's 2024 prediction of "GPT-4-level models costing a few cents per user" and Qu Kai's prediction that near-zero domestic Token prices would open up large DAU markets are conditional forecasts. Placing these disagreements side by side is safer than budgeting based on the cheapest forecast. [S33][S8][S52]
Data volume is not the same as usage rights. In 2023, Hu Yuanming of Meshy already pointed out that being able to see 3D models on a platform does not mean they can be used for training; the late-2023 macro discussion claim that "China has more scenario data" cannot bypass ownership, quality and process issues. Real work trajectories, personal conversations and outputs from external teacher models likewise require independent confirmation of permissible use; research value does not imply authorization. [S32][S29][S33]
In August 2026, Lu Yuxin's low-budget post-training case further broke down the metrics: being number one on Hugging Face Trending reflects popularity, not overall performance number one; a budget of a few hundred dollars covers part of the data and training for a personal project, and does not include from-scratch pretraining, labor, failed experiments or long-term serving. Raising performance on a target benchmark may be accompanied by declines in other capabilities. The data pathology of "declaring a task complete when it is not" requires checking against actual tool states, rather than only training polite phrasing; strong teacher trajectories may also exceed the capacity a student can absorb. Small-model routes must be compared on data, deployment, language, safety and regression performance, not judged by a single score gain. [S19]
Local AI likewise has genuine disagreements: guests emphasize sensitive scenarios and creative consumption, while the host and producer press on convenience and the gap to SOTA. Not sending data to model vendors only removes one outbound path; plugins, logs, syncing and application telemetry still need to be checked; local deployment does not automatically guarantee privacy, and owning a graphics card does not equal zero operating cost. This case supports bringing more options into comparison, but does not constitute a conclusion that everyone should train or deploy locally. [S19]
IV. The value of chips and hardware comes from system position and adoptability
In 2024, Ji Yu framed NVIDIA as competition at the system level of chips, software and developer ecosystem: CUDA, frameworks, operators and habits mean substitution is not just a matter of comparing point performance. New components can first supplement existing CPU architectures, migrate gradually, and then expand in value; a real product definition must state which tasks it serves, what it is willing to give up, and why incumbents are constrained by their installed base from making the same trade-offs. Being "more all-around" also means more trade-offs, and does not guarantee that no separable contradictions will ever emerge. [S56]
This both supports and has boundaries relative to Zhu Yibo's point on model–hardware co-design: deeper adaptation can create efficiency, while binding to a specific architecture may limit future changes; popular open-source models help ecosystem adoption but can also lock in investment directions. Neutral third parties that rely only on brief optimization leads may be squeezed by upstream and downstream, but neither speaker proved that third parties have no opportunity. Customer deployment freedom, migration burden, reliability and operating costs must be compared together. [S36][S56]
Consumer hardware cannot directly apply the chip ecosystem narrative. In 2024, Xie Yujia emphasized that independent terminals must first answer what scenarios/user groups they supplement relative to phones, before passing the three gates of capability, cost and willingness to use. R&D cycles of 12–18 months and the fact that tooling and hardware cannot be freely rolled back make wrong definitions more costly than in software; OTA can update content and algorithms but cannot fix all structural, power supply and safety issues. Reasonable gross margins must both cover services and provide room for error correction, rather than serving as an excuse for high prices. [S9]
Crowdfunding also requires distinguishing testing interest, concentrated launch marketing, and financing execution demonstrations. Millions in sales may come at a loss; early backers do not equal the mass market; small-scale ROI cannot be directly extrapolated to scaled volume. Channels, CPA, returns, delivery and subsequent repurchases are closer to a complete operating picture than launch revenue alone. This set of conditions makes the inference that "hardware is viable once models are good enough" untenable. [S9]
V. Application value accrues to complete delivery, not to company labels
Chen Yu's 2023 Zilliz case illustrates a transferable fulcrum: the team pivoted from GPU-accelerated databases to the vector demand exposed by customers' image-search use cases, while retaining compute and database engineering capabilities; prototypes are easy, but production performance and global commercialization are harder. He also used PingCAP to show how cross-platform neutrality can become an independent company's value. Identical underlying capabilities do not mean identical deployment scope, customer trust or delivery responsibility. [S48]
Meshy in late 2023 again provided two quality lines: professional and UGC. A generated image that looks 3D does not mean the mesh topology, textures, editability, rigging and game budget are all usable; ordinary users may also not know how to consume the files. Mid- and long-distance props and UGC game costumes may deliver early surprises, but that does not prove high-spec character production has been replaced. Research indicates that fit with existing production pipelines is an indispensable layer between model capability and revenue. [S32]
Enterprise delivery is the same. At the end of 2025, Jenny noted that 95% accurate investment banking document analysis may still require people to review from scratch, making the net benefit small; serving only two or three large customers with 60–70% of revenue concentrated may also be a custom business, and cannot be lightly declared standardized PMF. In 2024, Qu Kai proposed AI + human selling of outcomes as a path to bypass instability at the time, but rework and review must be costed, and "human fallback" must not be allowed to obscure gross margins. [S27][S3]
When foundation model companies build applications, and application companies fill in gaps in reinforcement learning or infrastructure, it shows that industry boundaries shift along with bottlenecks. Mo Jielin discussed Context using ServiceNow’s workflow delivery and Figma’s cross-organizational collaboration; whether these assets remain durable depends on users and delivery tasks, not on the “application layer” label. In 2026, Qu Kai proposed that continuously serving the same category of demand and adjusting solutions as models change gives operators a stable axis: demand does not need to change frequently to match capital’s trendy buzzwords. [S52][S7]
VI. Agent infrastructure is not only about computing power and model interfaces
Zhang Jinjian’s 2023 “bot buying wine” scenario already included demand understanding, merchant supply, communication traceability, payment permissions, deception, and dispute liability; in 2026 he again emphasized identity, representational authority, transactions, and accountability. Both periods together illustrate that being able to generate transaction suggestions is different from being able to legally and reliably represent a user to complete a transaction. Identity and liability remain issues to be built out; one cannot infer existing legal personhood or unlimited payment authority from the word “subjectivity.” [S46][S11]
The editorial distillation for product teams is to draw two maps at the same time: a technology dependency map (models, execution environments, Context, tools, and data) and a responsibility map (who authorizes, who holds data, who audits, and who recovers from failure). A working technical interface only validates part of the former. The 2025 vision of outcome-based payment still requires definable outcomes, accountable failure accounting, and customer willingness to migrate workflows; one cannot skip trust costs by “selling labor.” [S28]
VII. Constrain your own judgments using changes in sources
A reusable industry record sheet should include: publication date, role and interest relationships at the time, observed samples, claims, alternative explanations, what evidence would overturn the claim, and the next checkpoint. For financing volume, ARR, valuation, Token growth rate, cost, and crowdfunding ROI, retain the denominator, period, and stage for each; without a consistent caliber, do not participate in horizontal comparisons. In 2023, Zhang Jinjian re-read old newspapers to recover the information available at the time, and at the end of the segment he also warned against repeatedly persuading others to reinforce one’s own beliefs; Zhu Ning reminded that understanding bias does not make one immune to it in action. [S46][S50]
For product teams, the most actionable industry research conclusion is not “applications or models will win,” but to fix a set of real tasks and inspect quality, time to first token, end-to-end duration, cost per successful task, and human review; place users, vendors, and team budgets in the same scenario. Scale learning investment sustainably, and expand personnel, computing power, and capital in stages along validated links. Only by retaining failures, unfulfilled predictions, and both market and technology clocks can one avoid collecting only news that supports an existing direction. [S24][S36]
Source index and verification boundaries
The S numbers in the text correspond to the original programs below. Dates retain the time when the program was released; links contain official guest introductions and chapter navigation, and doubtful points in oral accounts and transcripts are not settled by guesswork.
S3: 2024 and 2025 will be the two most critical years for the next wave | AI year-end review (2024-01-13),Original program text.
S7: After 4 years of AI development, have applications disappeared? | AI mid-year review (2026-07-19),Original program text.
S8: Three development stages of AI, similarities and differences with the mobile internet, and future evolution paths | Conversation with Wang Hua of Sinovation Ventures (2024-08-10),Original program text.
S9: The Dao, Fa, and Shu of AI hardware | Conversation with Xie Yujia, founder of Sanermao (2024-04-20),Original program text.
S11: The first three years of going all in on AI | Conversation with Zhang Jinjian, partner at Oasis Capital (2026-01-10),Original program text.
S24: How bad was the AI market in the first half of the year? Don’t try to sprint right after starting | My response to Mo Jielin (2024-07-13),Original program text.
S26: How did the world end up with “the East rising and the West falling”? Discussing the secondary market and the DeepSeek+Manus boom | Conversation with Mo Jielin (2025-03-22),Original program text.
S27: The real differences between Chinese and American AI venture capital | Conversation with Jenny, partner at Leonis Capital (2025-12-27),Original program text.
S28: Why we are extremely optimistic about AI in 2025 | AI year-end review (2025-01-04),Original program text.
S29: Why are we starting to be optimistic? Visible and hidden threads of economic development, AI, and best practices | Conversation with Mo Jielin (2023-12-02),Original program text.
S32: From graphics to AI-generated 3D: How far are we from a 3D version of Midjourney? | Conversation with Yuan Ming, founder of Meshy.AI (2023-12-16),Original program text.
S33: From distillation to synthetic data to RSI, what is the next focus of model competition? | Conversation with Meng Fanqing, co-founder of Evolvent AI (2026-08-08),Original program text.
S36: Everything about AI Infra | Conversation with Zhu Yibo, co-founder of StepFun (2025-08-02),Original program text.
S39: Fatalism and the accursed Scaling Law | Hot take on CAI being acquired (2024-08-17),Original program text.
S46: The person investing most aggressively in AI | Conversation with Zhang Jinjian, partner at Oasis Capital (2023-07-29),Original program text.
S48: Revealing the investment logic behind Minimax, Zilliz, and PingCAP | Conversation with Chen Yu, partner at Yunqi Partners (2023-08-12),Original program text.
S50: Four necessary but insufficient conditions for a bubble | Conversation with Professor Zhu Ning, economist (2026-06-13),Original program text.
S52: Silicon Valley’s big AI turn and the secondary market bull run | Conversation with Mo Jielin (2025-08-23),Original program text.
S55: NVIDIA, Sora, and three core narrative logics of AI | Conversation with Mo Jielin (2024-03-02),Original program text.
S56: NVIDIA’s open scheme, flaws, and ambitions | Dialogue with Mackler Ji Yu, founder of Xingyun (2024-06-10),Original program text.
S17: “Do you think there is an AI bubble?” — Yes! | Conversation with Mo Jielin (2025-11-22),Original program text.
S19: How one person, in two weeks, with hundreds of dollars, trained a model that topped Hugging Face | Conversation with researcher Lu Yuxin (2026-08-29),Original program text.
Personal psychology of figures in sources, internal corporate motivations, market data, technical performance, and predictions are not automatically upgraded to verified facts. Official identities introduced in sources also represent only the time of the program; this manuscript does not extend public release authorization to private materials.
English version
AI Industry and Infrastructure Through 42 Zhangjing: Capabilities, Delivery, Capital, and Who Captures Value
This article is a synthesis bounded by its sources. It preserves frameworks, disagreements, and questions for further testing from 42 Zhangjing interviews published between 2023 and 2026. “At the time” must not become “today.” Business, financing, performance, and market figures that have not been independently checked are not treated as statistical facts or used to offer securities trading advice. Engineering examples from individual companies are discussed separately in related articles on hardware and frontier research.
1. Break a Grand Narrative into Five Propositions
Better model capabilities, applications attracting users, customers willing to pay, profitable companies, and high valuations from capital markets are five different things. In March 2024, Mo Jielin separated the narrative into scaling laws, competitors’ positions, and an application boom: growing momentum in the first two does not automatically deliver the third. A March 2025 discussion of DeepSeek and Manus similarly argued that model contributions and application delivery cannot be evaluated on the same axis. Using a third-party model does not make an application valueless, but presenting an application launch as a foundation-model breakthrough creates the wrong expectations. [S55][S26]
The editorial synthesis here expresses the causal chain as: a change in capability boundaries → a task becomes achievable → delivery becomes reliable enough → sustained use and payment → returns are distributed across the value chain. Capital can support each link in advance, while markets can also price in expectations before returns materialize. Share prices and enthusiasm for financing are therefore poor acceptance criteria for whether the chain is complete.
At the end of 2023, Mo distinguished the industry’s “visible threads” from “hidden threads” such as interest rates, inventories, and demand, separating practices a team could undertake from external conditions. In 2026, Zhu Ning further emphasized that technology’s real value, corporate profitability, and reasonable valuations are independent questions. Knowing that technology is advancing does not resolve price, timing, or the ability to bear the consequences. His four conditions—new concepts, liquidity, policy support, and inexperienced participants—were explicitly necessary but insufficient. They cannot be turned into a formula for identifying a market top whenever all four appear. [S29][S50]
The November 2025 conversation with Mo added a mismatch of time horizons: long-term construction of data centers and power capacity, short-term application revenue, and quarterly market expectations do not move in sync. A useful editorial proposal is to map the funding chain from end customers → applications → inference/training purchases → hardware/electricity, marking payments, commitments, terms, and gross margins rather than repeatedly adding together circular partnership announcements. Speaker identification in that episode is severely inaccurate. Claims that scaling laws had “failed” and that they had “not failed, but had become difficult to evaluate” are opposing views; they cannot be combined into a conclusion that pretraining was over. [S17]
2. Stage Narratives Help Only If Stages Are Not Treated as a Calendar
| Interview date | Main observations or predictions at the time | Conditions that must remain part of the test |
|---|---|---|
| Zhang Jinjian and Chen Yu, 2023 | Early teams might use resources to accelerate learning; models, middleware, and applications could all create value | Check technical feasibility, accumulated team capabilities, and customer demand separately rather than looking only at financing [S46][S48] |
| Qu Kai, early and mid-2024 | When controllability is inadequate, sell outcomes or restrict tasks; learning intensively is different from immediately committing all available funds | Human rework, delivery across the full process, and cash burn; a small working loop is not large-scale PMF [S3][S24] |
| Wang Hua, 2024 | Three layers: redesigning interaction, automating complex tasks, and AGI | Confidence and evidence differ by layer; a good first-layer experience does not establish full autonomy [S8] |
| Qu Kai, early 2025 | Agents would move from information services toward executing outcomes; multimodal systems would move toward participation | Billable outcomes, responsibility for failure, procurement trust, and actual completion [S28] |
| Mo Jielin, August 2025 | The narrative was shifting from model intelligence to token consumption and industrial adoption | Separate training demand from inference demand, and novelty-driven revenue from stable PMF [S52] |
| Zhang Jinjian and Qu Kai, 2026 | Large companies were responding strongly, application financing was weak, and models were gaining momentum | Continuity of demand and implementation iterations; a financing story must not be presented as technology already built [S11][S7] |
These are observations covering different circles, not a definitive industry chronology in which one stage necessarily follows another. The 2024 proposition that “generation beats recommendation” was an extreme thought experiment that explicitly excluded factors such as communities. The 2026 question “Have applications disappeared?” was rhetorical, not a statistical count of applications. Comparing these views requires asking which constraints changed and which needs persisted, instead of retroactively treating every new event as something already predicted. [S3][S7]
In August 2024, Qu Kai used the Character.AI event to project that “models are becoming commodities and the model wars are nearly over.” By 2026, he was observing stronger momentum behind models. This evolution in views is precisely what should be retained, rather than deleting the earlier prediction to assemble an uninterrupted story of being right. His CAI analysis also distinguished user value, product operations, proprietary models, and exit outcomes. Arrangements involving some employees and licensing do not mean the entire company was acquired through an equity purchase. An exit allowing some people to profit does not prove either product profitability or the failure of demand for companionship. Claims in the episode about individuals’ psychology, regulatory motives, and transaction amounts were not independently verified. [S39][S7]
3. Model Training, Data, and Cost: Examine Both Objectives and Operating Conditions
In 2025, StepFun’s Zhu Yibo described infrastructure as a combined system of computation, communication, storage, scheduling, and frameworks. The same model can have different time to first token, generation speed, stability, and cost per task. High throughput with a long initial wait may be unsuitable for chat. A model with fewer parameters is not necessarily cheaper: communication, memory access, operators, and hardware adaptation can change the bill. [S36]
Training and inference should not be merged into a single category of “compute demand.” In August 2025, Mo argued that pretraining demand was cyclical and inference depended on application usage, while the large volume of generation required by RL might increase demand on the training side. This was an explanation offered at that time, not a claim that the two must always rise together thereafter. Zhu used decoding to illustrate changing priorities: RL feedback also requires large amounts of output, so inference optimization affects training iterations. Optimizing model FLOPs utilization (MFU), training speed, and production serving cost can still produce conflicts. The right question is “Which bill are we optimizing?” rather than which metric should always be maximized. [S52][S36]
In 2026, Meng Fanqing placed data value back inside a feedback loop. Correct formatting, appropriate difficulty, and absence of cheating are only initial filters. Data is effective when training improves the target tasks without degrading previous capabilities. Simulated environments such as Feishu or Notion must first behave consistently with the real environments, and their scoring must also be checked. Otherwise, more trajectories may amplify false progress. The example of 99 tasks taking one minute each and one task taking 100 minutes illustrates how tail tasks slow down a whole batch. Data, algorithms, and scheduling jointly determine effective throughput. [S33]
Self-improvement also has distinct layers: correcting an output, changing the harness, producing training data, and training a new model are not the same achievement. A vertical task may use a strong model to build a process and a smaller one to execute it, but quality, privacy, cost, and maintenance still require measurement. Meng disagreed that prices must approach zero. Wang Hua’s 2024 prediction of “GPT-4-level models for a few fen per user,” and Qu Kai’s argument that near-zero domestic token prices could open markets with large daily active user bases, were conditional predictions. Keeping these disagreements side by side is more prudent than budgeting against whichever forecast is cheapest. [S33][S8][S52]
Data volume is not permission to use it. In 2023, Meshy’s Hu Yuanming already noted that being able to view a 3D model on a platform does not mean it can be used for training. The late-2023 macro discussion that “China has more data from real-world scenarios” cannot bypass ownership, quality, and process requirements. Real work trajectories, personal conversations, and outputs from external teacher models likewise require separate confirmation of permitted use. Research value does not imply authorization. [S32][S29][S33]
Lu Yuxin’s low-budget post-training case in August 2026 further separated the metrics. Ranking first on Hugging Face Trending indicates popularity, not first place in overall performance. The few hundred dollars covered part of one personal project’s data and training costs, not pretraining from scratch, labor, failed experiments, or long-term serving. A gain on the target benchmark might also accompany losses in other capabilities. The data problem he described—declaring completion when the task was unfinished—requires comparison against actual tool state, rather than merely training more modest wording. Trajectories from a strong teacher may also exceed what the student can absorb. A small-model approach needs comparisons across data, deployment, language, safety, and regression performance, not just one higher score. [S19]
Local AI involved real disagreement as well. The guest valued sensitive use cases and the compute consumed by creative work, while the host and producer questioned convenience and the gap from the state of the art (SOTA). Keeping data away from a model vendor removes only one possible leakage path; plugins, logs, synchronization, and application telemetry still need checking. Local deployment does not automatically eliminate privacy concerns, and owning a GPU does not mean operating costs are zero. This case supports including more options in comparisons, not a conclusion that everyone should train models or deploy locally. [S19]
4. Chip and Hardware Value Comes from System Position and Adoptability
In 2024, Ji Yu explained Nvidia’s position as system-level competition across chips, software, and the developer ecosystem. CUDA, frameworks, operators, and established habits mean that replacing it requires more than comparing isolated performance. A new component can first complement the existing CPU system, migrate gradually, and then expand its value. A genuine product definition should explain which tasks it serves, what it is willing to give up, and why incumbents’ installed bases prevent them from making the same trade-off. Being “more general-purpose” also means balancing more requirements; it does not guarantee there will never be a conflict that a focused entrant can exploit. [S56]
This supports Zhu Yibo’s discussion of model–hardware co-design, while also exposing its limits. Deeper adaptation can create efficiency, but attachment to a particular architecture can constrain later changes. Popular open-source models can encourage ecosystem adoption while also locking in investment directions. A neutral third party relying only on a temporary optimization lead may be squeezed by upstream and downstream players, but neither speaker proved that third parties necessarily have no opportunities. Customer freedom in deployment, migration burden, reliability, and operating cost must be compared together. [S36][S56]
Consumer hardware cannot simply inherit the chip-ecosystem narrative. In 2024, Xie Yujia stressed that a standalone device must first explain which use case or audience it serves beyond what a phone provides, then pass three tests: capability, cost, and willingness to use it. A 12–18-month development cycle, tooling, and hardware that cannot be freely rolled back make a mistaken product definition more costly than in software. Over-the-air (OTA) updates can change content and algorithms, but cannot repair every structural, power, or safety issue. A reasonable gross margin must cover service and leave room for correction; it is not an excuse for a high selling price. [S9]
Crowdfunding also needs to be distinguished by purpose: testing interest, concentrated launch marketing, or demonstrating execution for fundraising. Sales of a million may still lose money, and early backers do not represent the mass market. Small-scale ROI cannot be directly extrapolated to higher volumes. Channels, cost per acquisition (CPA), returns, fulfillment, and later repeat purchases provide a fuller view of operations than launch proceeds alone. Together, these conditions undermine the inference that “a good enough model makes hardware viable.” [S9]
5. Applications Capture Value Through Complete Delivery, Not Company Labels
Chen Yu’s 2023 Zilliz example showed a transferable technical foothold. The team shifted from GPU-accelerated databases toward vector requirements revealed by customers’ image-search needs, retaining its capabilities in computation and database engineering. A prototype was easy; production performance and global commercialization were harder. He also used PingCAP to show how cross-platform neutrality could create value for an independent company. Sharing underlying capabilities does not mean sharing the same deployment scope, customer trust, or delivery responsibilities. [S48]
At the end of 2023, Meshy offered two further quality thresholds: professional production and user-generated content (UGC). An image that looks 3D does not establish that mesh topology, textures, editability, rigging, and game performance budgets are all usable. Ordinary users may not even know how to use the file. Mid-distance and background props, or UGC clothing for games, might produce early positive surprises; that does not prove that high-specification character production has been replaced. Adapting a research representation to existing production pipelines is an indispensable layer between model capability and revenue. [S32]
Enterprise delivery follows the same logic. At the end of 2025, Jenny said that investment-banking document analysis with 95% accuracy could make people review everything from the beginning, leaving little net benefit. Serving only two or three large customers, with 60–70% of revenue concentrated among them, might also indicate custom services rather than standardized PMF. In 2024, Qu Kai proposed selling outcomes through AI plus human labor as a way around instability at that time. Rework and review must still be counted; a “human backstop” must not obscure the gross margin. [S27][S3]
Foundation-model companies building applications, and application companies adding RL or infrastructure, suggest that boundaries move with the bottlenecks. Mo discussed context through ServiceNow’s workflow delivery and Figma’s cross-organization collaboration. Whether those assets endure depends on users and the tasks being delivered, not the “application layer” label. In 2026, Qu Kai’s suggestion to keep serving the same category of demand while changing the solution as models evolve gave operators a stable axis: needs need not change whenever a new term becomes popular with investors. [S52][S7]
6. Agent Infrastructure Extends Beyond Compute and Model APIs
Zhang Jinjian’s 2023 idea of a “bot buying wine” already included understanding demand, merchant supply, traceable communications, payment permissions, deception, and responsibility for disputes. In 2026, he again emphasized identity, authority to represent someone, transactions, and accountability. Together, these episodes show that generating a transaction recommendation is different from legally and reliably completing a transaction on a user’s behalf. Identity and responsibility remained problems to be built for; the word “agency” does not establish legal personhood or unlimited payment authority. [S46][S11]
The editorial proposal for product teams is to draw two maps together: a technical dependency map covering models, execution environments, context, tools, and data; and a responsibility map covering who authorizes actions, who holds data, who reviews, and who restores the system after failure. Connecting technical interfaces validates only part of the first map. The 2025 vision of outcome-based payment also requires definable outcomes, failures whose costs can be accounted for, and customers willing to migrate their processes. Calling it “selling labor” does not remove the cost of earning trust. [S28]
7. Use Changes in Sources to Constrain Your Own Judgments
A reusable industry record should contain publication date, the speaker’s role and interests at that time, observation sample, claims, alternative explanations, evidence that could disprove them, and the next checkpoint. Financing counts, annual recurring revenue (ARR), valuations, token growth, costs, and crowdfunding ROI each need their own denominator, period, and stage. Without a defined basis, they should not enter cross-case comparisons. In 2023, Zhang Jinjian revisited old newspapers to reconstruct the information available at the time; the episode’s closing remarks also warned that repeatedly persuading others can reinforce one’s own beliefs. Zhu Ning reminded listeners that understanding biases does not make one immune to them in action. [S46][S50]
For product teams, the most actionable result of industry research is not deciding whether “applications or models will win.” It is fixing a set of real tasks and checking quality, time to first token, total completion time, cost per successful task, and human review, placing user, supplier, and team budgets within the same scenario. Learning can receive sustained investment, while staffing, compute, and funding expand in stages around validated links. Keeping records of failures, missed predictions, and the different clocks of markets and technology helps avoid collecting only news that supports an existing direction. [S24][S36]
Source Index and Verification Boundaries
The S references identify the original episodes listed below. Dates retain the episodes' historical timing. The linked pages provide official guest introductions and chapter navigation; uncertainties in speech or transcription are not resolved through guesswork.
S3: “2024 and 2025 Will Be the Two Most Critical Years of the Next Wave | AI Year-End Review” (2024-01-13), Original episode.
S7: “After Four Years of AI Development, Have Applications Disappeared? | AI Midyear Review” (2026-07-19), Original episode.
S8: “AI’s Three Stages of Development, Similarities and Differences with Mobile Internet, and Its Future Path | A Conversation with Sinovation Ventures’ Wang Hua” (2024-08-10), Original episode.
S9: “The Principles, Methods, and Techniques of AI Hardware | A Conversation with Sanermao Founder Xie Yujia” (2024-04-20), Original episode.
S11: “The First Three Years of Going All In on AI | A Conversation with Oasis Capital Partner Zhang Jinjian” (2026-01-10), Original episode.
S24: “How Bad Was the AI Market in the First Half? Don’t Try to Sprint Just After Starting | Answering Mo Jielin” (2024-07-13), Original episode.
S26: “How Did the World Become ‘East Rising, West Falling’? Public Markets and the DeepSeek + Manus Boom | A Conversation with Mo Jielin” (2025-03-22), Original episode.
S27: “The Real Differences Between AI Venture Investing in China and the US | A Conversation with Leonis Capital Partner Jenny” (2025-12-27), Original episode.
S28: “Why We Are Extremely Optimistic About AI in 2025 | AI Year-End Review” (2025-01-04), Original episode.
S29: “Why Are We Becoming Optimistic? The Visible and Hidden Threads of Economic Development, AI, and Best Practices | A Conversation with Mo Jielin” (2023-12-02), Original episode.
S32: “From Computer Graphics to AI-Generated 3D: How Far Are We from a 3D Midjourney? | A Conversation with Meshy.AI Founder Yuanming” (2023-12-16), Original episode.
S33: “From Distillation to Synthetic Data to RSI: What Is the Next Focus of Model Competition? | A Conversation with Evolvent AI Co-Founder Meng Fanqing” (2026-08-08), Original episode.
S36: “Everything About AI Infrastructure | A Conversation with StepFun Co-Founder Zhu Yibo” (2025-08-02), Original episode.
S39: “Fatalism and the Much-Maligned Scaling Law | Provocative Takes on CAI’s Acquisition” (2024-08-17), Original episode.
S46: “The Most Aggressive AI Investor | A Conversation with Oasis Capital Partner Zhang Jinjian” (2023-07-29), Original episode.
S48: “Inside the Investment Logic Behind Minimax, Zilliz, and PingCAP | A Conversation with Yunqi Partners’ Chen Yu” (2023-08-12), Original episode.
S50: “Four Necessary but Insufficient Conditions for a Bubble | A Conversation with Economist Professor Zhu Ning” (2026-06-13), Original episode.
S52: “Silicon Valley’s Big AI Turn and the Public-Market Bull Run | A Conversation with Mo Jielin” (2025-08-23), Original episode.
S55: “Nvidia, Sora, and Three Core Narratives of AI | A Conversation with Mo Jielin” (2024-03-02), Original episode.
S56: “Nvidia’s Open Strategy, Vulnerabilities, and Ambitions | A Conversation with Xingyun Founder Mackler Ji Yu” (2024-06-10), Original episode.
S17: “‘Do You Think AI Is in a Bubble?’—Yes! | A Conversation with Mo Jielin” (2025-11-22), Original episode.
S19: “One Person, Two Weeks, a Few Hundred Dollars: Training a Model That Tops Hugging Face | A Conversation with Researcher Lu Yuxin” (2026-08-29), Original episode.
Statements in the sources about individuals’ psychology, internal corporate motives, market data, technical performance, and predictions do not automatically become verified facts. Official roles in source introductions apply only to the episode’s historical setting. This article does not extend publication authorization to private materials.