Summary

The State Council’s new outbound-investment rules are not simply about corralling money back home; they put how domestic earnings leave the country, where they are invested and what assets they hold into a traceable framework. 李丰 calls it a rare “upper-level law”: it covers individuals, companies and related entities, with ministries to follow by drafting detailed rules. The Evergrande and Wahaha family-trust disputes, the ODI suspension, the suspected Manus data transaction and tighter approval for VIE companies listing in Hong Kong are all different entry points into the same cross-border chain of funds and assets.
This round of regulation may be happening now because the dollar is siphoning global liquidity without the US itself doing much monetary easing. A roughly $70T US equity market needs fresh money to make new highs, and 10+ years of cheap yen in Japan have become the main liquidity reservoir; after taking over, 沃什 favored Fed balance-sheet reduction, while the US also needs buyers for rolled-over Treasuries, deficits and rising interest costs. Yet when the 30-year Treasury yield broke above 5%, capital still went first into stocks. 李丰’s view is that China at minimum needs to stop unauthorized money from joining this liquidity drain.
China’s dollar-denominated property debt is the episode’s most contested domestic case study of the dollar liquidity cycle. 李丰 argues that before 2020, roughly one-quarter of China’s external debt was concentrated in property; by deliberately severing the bank-developer chain in 2021 and allowing offshore bonds to move to near-total default by the end of 2023, China avoided using state funds to bail out external creditors. 李翔 maintains that corporate debt is not sovereign debt: China has state-owned banks, state-owned core assets and capital controls, so it cannot be compared with Argentina or South Korea. “The rescue is for the building, not Evergrande” is the narrow point on which they agree.
The AI capital frenzy is shifting from a valuation story to a direct bid for cash in the public markets. The figures discussed were roughly $75B for SpaceX, $80B in new Google stock and a possible additional $70B-$80B from Anthropic—more than $200B across the three—against roughly $500B-$700B in cumulative AI capex commitments from leading companies this year. The “last available liquidity” has to support US equities while also being absorbed by data centers and mega-IPOs.
What determines whether the AI edifice stands is not chip pricing but whether each token can ultimately be sold at a price that does not lose money. Search engines retrieve webpages that already exist; large models “have to compute every interaction again.” If capital markets demand current-period break-even, even 2 cents per question could send large volumes of low-value demand back to search. Doubao beginning to charge is therefore not a side story: it puts front and center the fact that consumers, enterprises or advertisers must ultimately pay for the entire AI capex chain.
China’s K-shaped recovery may turn tech market caps into more broadly distributed employee wealth, with housing demand in core cities showing it first. 李丰 argues that chips, robotics and technology manufacturing need equity to bind hundreds of employees; if they rise to 30%-50% or even 50%-60% of the capital market, the number of beneficiaries could be dozens of times larger than under the traditional manufacturing model centered on founders. 李翔 cautions that even leading companies such as Unitree still have few employees and may not be enough to lift urban housing prices. Tech newcomers queuing for luxury homes in Hangzhou and Shenzhen, and Changxin Storage employees receiving close to RMB1B in stock, are directional samples rather than sufficient evidence.
A-shares are being forced into an extreme seesaw by gains concentrated in narrow sectors, fund-style drift and Huijin selling broad-market ETFs. Public-fund assets have broken through RMB39T for the first time, but most new money is going into fixed-income and fixed-income-like products; meanwhile, 2 Huijin entities cut their ETF holdings by at least RMB500B in 6 months, with shares down by more than one-third, while funds sold large-cap banks and consumer stocks to chase “optics, chips.” If US AI assets correct, Huijin could return to broad-market ETFs while style-drifting funds sell high and buy low, sending capital back in the opposite direction with the same concentration.

Deep dive

1. The State Council Elevates Fragmented Rules into an Outbound-Investment Framework

李丰 says the most unusual feature of the latest document is who issued it. In the past, implementation notices came separately from the NDRC, the financial regulator, the banking regulator and the securities regulator. This time, the State Council set the direction, scope and boundaries first, leaving lower-level agencies to draft the detailed rules.
The framework he calls an “upper-level law” covers individuals, companies and related entities at the same time. It regulates not just a single securities account but the entire outbound-investment process: the source of funds, the route out of the country, the investment purpose and the destination of the assets.
That also explains why the policies involving Futu and Tiger Brokers are not isolated events. The regulatory focus is not on any particular platform, but on how domestic operating income enters an overseas account, whether the source can be verified and what happens afterward.

2. Family Trusts, ODI, Data Transactions and VIEs Point to the Same Cross-Border Chain

The Evergrande and Wahaha family-trust disputes expose the same set of questions: how trust assets were created, whether money left the country through genuine or fictitious investments, and whether an overseas trust can be pursued when domestic debt or asset-division disputes arise.
The equity-investment industry runs directly into regulation through Document 37 and ODI. 李丰 says a major municipality issued window guidance early this year: ODI procedures still unresolved since 2023 were put on hold because the new policy had not been clarified. Recent rumors about tracing accounts also largely point back to 2023.
When Meta acquired Manus, 李丰 speculates that Meta took the intellectual property, data and core team while leaving the company operating to avoid antitrust issues. What may actually trigger Chinese regulation is the transfer into the transaction of data generated onshore without approval.
Reports circulating in March and April that approvals for VIE companies seeking Hong Kong listings had tightened have not been officially confirmed. But where the principal business scope and operations are onshore while the equity structure sits offshore, the companies do fall within the same set of boundaries that remain to be clarified.

3. A Larger Hong Kong IPO Pipeline Makes “Where Should the Money Stay?” Unavoidable

After a company lists in Hong Kong, whether through a domestic or red-chip structure, its IPO proceeds, follow-on issuance proceeds, secondary-sale proceeds and founder cash-outs are all funds raised in Hong Kong or offshore. Whether those funds should return to China and how they should be filed is precisely what the new rules are meant to sort out.
The reverse flow exists as well. More state-owned capital is participating as cornerstone or anchor investors in Hong Kong IPOs, requiring ODI approval before RMB can be converted into HKD or another foreign currency. The channel therefore covers more than private capital.
李丰 cited a recent report saying Hong Kong has overtaken Switzerland as the leading offshore wealth-management center. His explanation is that Switzerland damaged its credibility by freezing Russian-linked assets, while Hong Kong and Singapore have risen sharply since 2023.
The broader backdrop is that Chinese households and companies have accumulated enough offshore assets through overseas investment, factory construction and partnerships that regulators must answer how the money left, what it did abroad and whether it came from Chinese operating income.

4. Tracing Lawful Sources Is the Main Line; Suppressing IPO Speculation Is a Side Effect

Social media most readily interprets the policy as a ban on residents buying US or Hong Kong stocks, or as an attempt to force money back into domestic markets. 李丰’s correction is that the main line is “lawful purpose, lawful channel and traceability,” with tax and financial management included.
Hong Kong accounts can already be linked with mainland tax information through CRS. The new framework is meant to track the full conversion of domestic wealth into overseas cash, equity, trusts or data assets, rather than focusing only on a single securities trade.
Hong Kong has seen a boom in 1- to 5-month IPO flipping. Some capital uses gray-market FX, onshore collateral against offshore borrowing, and routing through Singapore, the Cayman Islands or BVI. If gray-market activity is explicitly defined as illegal, the difficulty and cost of moving money out—and the potential legal penalty—will rise.
There are still no detailed rules on whether individual stock trading requires approval, how quotas will be set or how filings should work. 李丰 therefore downgrades the impact on Hong Kong IPO speculation to a “small factor” and no longer treats it as the policy’s main cause.

5. “Plugging Every Hole” Usually Means the Reservoir May Be Filled Next

李丰 uses a reservoir analogy: if regulators are only dealing with old problems, they can plug holes and patch leaks one by one. When they unify every channel into a single system, it usually means more than closure—the direction itself is about to see a larger change.
His exact formulation was: “The usual reason for plugging every hole is that water is about to be released.” That may point to a larger policy shift ahead, but the program made no specific forecast.
The timing can also be explained by the dollar liquidity cycle. With China’s outbound-investment framework still incomplete and the US drawing liquidity from the world, China would at minimum restrict unauthorized or opaque money from joining that outflow.

6. A $70T US Equity Market Needs Fresh Liquidity Borrowed from Japan

李丰’s shorthand is that the US equity market is worth roughly $70T. To push the broad indexes to new highs, existing capital rotating between sectors is not enough: “New water has to come in for the water level to rise.”
There has been no synchronized global easing, so the team views Japan as the main reservoir in this cycle. After 10+ years of near-zero rates, investors can borrow yen, convert into dollars and earn a carry even by buying Treasuries.
The carry trade requires persistent yen selling and dollar buying, naturally pressuring the yen. Whether the money ends up in Treasuries, stocks or other dollar assets is not Japan’s decision: “The money is definitely there for the arbitrage.”
Japan recently spent more than ¥1T to stabilize the exchange rate, producing only about ¥1 of movement. After reviewing decades of intervention records, the team’s conclusion was that the marginal efficiency of Japanese FX intervention has fallen sharply.

7. A Freely Floating Exchange Rate Does Not Automatically Remove One-Way Depreciation Risk

李翔 argues that if Japan’s exchange rate were free to move, the market itself could use depreciation to cushion capital outflows. China, by contrast, wants both low interest rates and a stable RMB-dollar exchange rate, which appears to create greater pressure.
李丰’s rebuttal is that once rapid depreciation creates a one-way expectation, it attracts more capital betting on further declines. Corporate income, property and other assets denominated in the local currency would also be impaired, transmitting the risk from FX markets into the real economy.
Their disagreement ultimately comes down to the impossible trinity. Once capital is fully free to move, foreign-exchange reserves, central-bank credibility and political will alone are unlikely to stop a large outflow. A country that wants to retain the ability to intervene in its exchange rate must retain some capital controls.
China has never opened the capital account to fully free RMB convertibility. Combined with substantial foreign-exchange reserves, that gives it more room to adjust. 李丰 also revised his statement that Japan “lacks the will”: more precisely, Japan has less room, force and autonomy to act.

8. Fed Balance-Sheet Reduction Forces the US to Draw Liquidity Globally, but the Money Went to Stocks First

The premise of the discussion is that the dollar has been absorbing global liquidity since April, even though the US itself has not launched large-scale quantitative easing. After 沃什 took over as Fed chair in May, the prevailing preference remained balance-sheet reduction and preservation of Fed independence.
Balance-sheet reduction means the Fed buys fewer Treasuries, or even sells them. The US must also roll over old debt, cover a widening fiscal deficit and pay rapidly rising interest costs, so it needs more overseas capital to take the other side of dollar assets.
The anomaly is that the 30-year Treasury yield briefly rose above 5%, yet incoming funds did not rush into bonds. More went into higher-risk equities, creating the appearance of a renewed US stock-market rally since April.
李丰 uses that backdrop to explain the timing of China’s policy, while clearly labeling it as the part he is guessing: the US is not creating liquidity itself but still needs to support both bonds and stocks, giving China more reason to control unauthorized outflows.

9. Argentina Shows How Floating-Rate Dollar Debt Becomes an Asset-Transfer Chain

Looking back at roughly 60 years of Argentine history, colleagues on the team broke the dollar cycle into a repeated loop: during low-rate liquidity injections, dollars enter through investment and external debt; during tightening, the local currency depreciates, debt-service costs rise, and the asset and political structures are reorganized.
The most dangerous design is floating-rate debt. Debt accumulates when rates are ultra-low while operating revenue is earned in local currency. Once dollar rates rise, principal, interest and the exchange rate all tighten at once; the result is either repayment with assets or insolvency.
李丰 says Argentina has gone through a similar loop nearly 3 times, including one before the Milei government took office. Inflation, a collapsing local currency and sovereign-credit bankruptcy compounded; after a change in government, the country borrowed again.
Citing a description in The Oil Wars, he said that in some loans, less than 10% of the funds were genuinely available for the borrower to use freely, with the rest tied to designated purchases. That simultaneously entrenches sovereign debt, household poverty and control of resources by a small number of families.

10. 李翔 Rejects Blaming Every Development Failure on “Petrodollars”

李翔’s objection is that Argentina’s problems first came from its own development model. IMF and World Bank loans may have restricted uses because corruption in recipient countries was severe, not necessarily because the US designed an asset-stripping operation.
He acknowledges that the IMF and World Bank board structures create room for political bargaining, but insists that “oil is the result, not the cause.” East Asia, he notes, did not later follow the development prescription in full.
李丰 responds that Argentina’s first cycle was indeed centered on oil and natural resources, and that the restrictions on IMF-loan use before 2000 were closely aligned with US export demand. Only after one cycle did the domestic oligarchic structure become further entrenched.
The dispute remains unresolved. One side emphasizes institutions, corruption and economic structure; the other emphasizes the external constraints jointly created by dollar demand, loan terms and the interest-rate cycle.

11. The Dollar Cycle Captures Carry and Rewrites the Shareholder Lists of Core Assets

李丰 summarizes the cycle’s outcome as 3 things happening at once: “Ordinary people take on debt, wealth becomes highly concentrated, and US financial groups obtain equity and economic interests in core assets.”
Japan and South Korea were more typical examples after the Asian financial crisis. Large numbers of US financial groups appeared among the major shareholders of large companies and banks. The gains from the capital cycle can come from acquiring equity cheaply during a crisis or from trading profits after a bubble rises.
He also mentioned a Korean mega-company under pressure this year to monetize nearly all of its non-core equity investments in China by June 30, on the grounds that “they cannot use our money to support China’s competitive core technology industries.”
For the chain to work, 3 conditions must coexist: free cross-border capital flows, global demand for dollars, and financial institutions able to decide when to lend and when to call money back based on the interest-rate cycle.

12. China’s Dollar Property Debt Is the Closest—and Most Contested—Domestic Case Study

China has relatively little external debt overall because of its foreign-exchange reserves and capital controls. But 李丰’s estimate is that before 2020, roughly one-quarter of external debt was concentrated in property, mainly corporate bonds issued offshore by developers.
After August 2021, the property bubble was deliberately punctured. By the end of 2023, the related offshore dollar bonds had “almost all blown up.” Defaults also created a cross-default effect: once one bond triggered, other creditors could demand early repayment of principal and interest at the same time.
The “guarantee delivery of homes” program used government funds to pay directly for construction and delivery, preventing developers from using the money to repay debt. 李翔’s point was that “the rescue is for that building”; companies such as Evergrande could still go bankrupt.
Whether Shenzhen Metro should continue supporting Vanke became a boundary question. If state-owned enterprises and fiscal funds keep injecting capital, are they protecting the projects, the financial system or indirectly guaranteeing repayment of external debt?

13. The Two Did Not Agree on Whether Exposing the Risk in 2021 Was Necessary

李丰 believes that cutting off bank financing to developers early and allowing offshore debt to default, though extremely painful for China’s economy, prevented RMB loans, state-owned assets or fiscal funds from ultimately filling the hole left for external creditors.
李翔’s rebuttal is that corporate debt and sovereign debt are entirely different. China’s major banks and core assets are state-owned, and strict FX controls mean an insolvent developer cannot simply convert RMB loans into dollars to repay debt.
They also disagree on how banks would have behaved. 李丰 thinks that without a red line, banks might have continued rolling loans, allowing developers to repay external debt from operating cash flow until the public system absorbed the bad debt. 李翔 believes banks would not “so obviously” funnel money to an insolvent borrower.
They agree only that the risk was exposed deliberately, not that the pace was optimal. 李翔 believes 2021 may have allowed a gentler solution. 李丰 stresses that waiting until the Fed began hiking in March 2022 would have made the situation more difficult: rates rose by 50, 75 and 75 basis points in succession from May through July, reaching 2.25% by July.

14. The Key to the “House” Metaphor Is Not Calling the Top but Influencing the Liquidity Window

李丰 uses Japan from 1985 to 1991 to show that the same pool of capital can trade stocks and property during an upswing, then select the corporate equity it truly wants during a crisis. Assets and profits are not mutually exclusive.
李翔’s pushback is direct: even if a financial group becomes a major shareholder, selling can crush the price. No financial group can precisely control the next financing round’s tenfold increase or exit perfectly at the top.
李丰 concedes that hitting the absolute top is unnecessary. The key is whether one can influence a range: “I know the rise I pushed ran roughly from 5 to 10.” If the host country restricts liquidity at 7 or allows the debt to blow up directly, the arbitrage chain cannot close as planned.
China discussed foreign control of large companies such as Xugong and Yuchai around 2007-2010, but the idea was never implemented on a large scale. Along with capital controls, these institutional differences explain why China never completed the traditional dollar-liquidity cycle.

15. The K-Shaped Recovery Is Turning Tech Market Caps into More Distributed Employee Wealth

李翔 says property markets in major cities currently appear broadly stable, with Shanghai and Shenzhen perhaps close to stabilization, though the point remains contested. 李丰 believes the real medium-term support for demand could come from the shift from traditional equipment manufacturing toward technology industries.
First-quarter economic data, trade data and the PMI subindices for 5月 all show a “K-shaped recovery”: year-on-year growth and the share of technology-related industries are rising, while property remains a drag.
Unlike traditional manufacturing, which relies on 2 or 3 founders, chips, robotics and technology manufacturing depend heavily on talent and must use equity incentives to bind 100-200 or even several hundred employees. At the same RMB30T-RMB40T market capitalization, the number of beneficiaries could increase by dozens of times.
李翔 cautions that leading companies such as Unitree have only a few hundred employees and may not be able to pull up housing prices. 李丰 emphasizes that China has more than one Unitree and that technology manufacturing has a longer upstream and downstream chain. Luxury-home purchases by newcomers in Hangzhou and Shenzhen, and nearly RMB1B of Changxin Storage shares used for employee ownership, are merely the earliest samples.

16. 3 Mega-Financings Are Competing with the Entire US Equity Market for the Same Cash Pool

In the past, a $10B-$20B offering could be called a mega-IPO, with Meta and Alibaba’s 2014 listing as reference points. SpaceX’s target this time is $75B, putting the offering at a scale not previously seen.
Google also plans to issue $80B of new stock, with Berkshire Hathaway subscribing to $10B privately. Buffett’s roughly $300B cash pile has begun to decline, making the transaction a signal for watching the cycle as well.
李丰 conducted 2 “boring qualitative studies.” Berkshire’s cash peaking does not itself mean the stock market has peaked, but once the cash starts being redeployed, a market inflection may appear within several months to more than 1 year. After a mega-IPO, the inflection window runs from immediately to 10 months.
Facebook’s 2012 listing was an exception in which no inflection appeared. The post-crisis easing environment was still in place, and the European debt crisis pushed European funds back into the US. Today there is no comparable source of incremental dollar liquidity.

17. Google’s $80B Issuance Is the Answer to Capital-Expenditure Constraints

The balance-sheet figures cited for Google were cash and cash equivalents below $130B, long-term debt of roughly $60B-$70B, or about 50% of cash. The comparable figure is roughly 60% for Microsoft and close to 70% for Amazon and Meta.
Google’s cash flow was about $170B last year. If it fully executes this year’s roughly $180B capex guidance, capital spending will consume an entire year’s cash generation. Maintaining the current cash-to-debt ratio while borrowing another $50B-$60B would bring it close to its comfort limit.
It therefore chose a third path: get ahead of SpaceX and “draw the same-size tube of water first,” issuing $80B directly into the secondary market instead of continuing to consume cash or relying only on debt.
Anthropic, valued at roughly $90B according to the program, may also raise $70B-$80B. With SpaceX included, the 3 funding rounds total more than $200B, before any potential OpenAI listing.

18. AI Capex Supports Profit Guidance but Has Yet to Show Up in First-Quarter Physical Data

Google, Microsoft, Amazon, Meta and Oracle have committed roughly $500B-$700B of combined capex this year. Data centers are therefore amplifying demand layer by layer across chips, equipment, materials, hiring and pricing.
The team’s qualitative estimate for the S&P 500 is that roughly 40% of profit guidance may come from the transmission of this historic capex cycle. If the S&P 500 is a proxy for the US economy, perhaps one-third of growth also depends on these construction plans being executed as scheduled.
The estimate does not yet deduct constraints from power, land, environmental approvals or equipment supply. More puzzlingly, on May 28 the US revised first-quarter GDP growth down from 2 to 1.6, while capital-goods investment fell 1.1 quarter on quarter.
In China’s broadly defined electrical and mechanical equipment category, digital-processing equipment and some data-center-related equipment also showed no growth. Consumer confidence fell to a low. 李丰 guesses that the US-Iran war in February and March may have caused some disruption, but stresses that these are volatile data points that still require observation.

19. Doubao’s Paid Model Exposes the AI Pyramid’s Weakest Layer

The base of the AI edifice is data-center capex, supporting chips, equipment, models and applications as it moves upward. But every enterprise investment must ultimately be paid for by consumers, enterprise customers or advertisers.
Doubao’s decision to charge small businesses and consumers therefore matters more than a simple price-increase headline. It puts the question of whether end users will bear the real cost of inference at the center of the capex chain.
李丰 emphasizes the structural difference between large models and search. Search mainly retrieves static webpages already stored in a database, while a model “has to compute every interaction again.” No question can bypass cloud-based token consumption.
The market currently does not require model companies to recoup historical R&D spending through pricing, or even to cover current inference costs with current-period revenue. If the valuation logic cools, “every token must be profitable” will become the new discipline.

20. Full-Price Tokens Will Re-Test the Real Demand for Replacing White-Collar Work

If an ordinary question costs 2 cents, many users may avoid large models and return to search. Usage observed under today’s free or low-cost regime therefore does not equal effective demand under full pricing.
Meta once published a leaderboard of employee token usage, then removed it after discovering that employees were using too much. Microsoft saw something similar. Even in coding, it is difficult to tell how many calls create real value and how many are experiments or simply “burning tokens.”
If capital markets demand immediate profitability, model companies will cut data purchases and R&D. Leaders will advance more slowly, and there will be fewer new results for later entrants to distill. 李丰 speculates that the technology gap could therefore narrow.
Google may be relatively advantaged after the cooldown: it controls the model, applications, consumer gateway and advertising system, with cost control “like BYD.” But if every question-and-answer round must pay for itself, it remains unclear how much advertising value would need to rise.

21. Deposits Are Moving First into Fixed Income; Benchmark Reform Will Not Automatically Trigger a Major Reallocation

Fund benchmark adjustments are first about correcting style drift. A consumer fund cannot hold a large amount of chips while continuing to benchmark itself against a consumer index. Managers must either bring holdings back in line or change the benchmark; neither action naturally implies a one-off market-wide repositioning.
April financial data showed deposits rising, with deposits at non-bank financial institutions also increasing rapidly. Chinese public-fund assets broke through RMB39T for the first time, but the fastest growth came from bond and fixed-income-like products.
Deposit rates are low. Some banks no longer offer 3-year or 5-year term deposits at low rates, while “extremely mild inflation,” with CPI at least not below 0, means the huge stock of deposits is beginning to look for somewhere that can at least preserve purchasing power.
The current “deposit migration” is mainly from term deposits into fixed-income-like products, with a small amount moving into floating-return products. So far this year, neither insurance nor the stock market has absorbed a comparable volume of household capital.

22. Huijin and Style-Drifting Funds Have Turned A-Shares into an Extreme Seesaw

Only a handful of sectors, notably optics and chips, have produced standout fund performance this year. To chase returns, managers have drained banks, consumer stocks and other large caps to buy the hot sectors. That is why style drift has become widespread.
Over the past 6 months, the ETF shares held by 2 Huijin entities fell by more than one-third, implying sales of at least RMB500B. Their holdings had previously expanded from RMB100B-RMB200B to roughly RMB1.5T, with broad-market products such as the SSE 50 bought to stabilize the market.
李丰 guesses that once a few sectors rose too quickly and mechanically lifted the indexes, Huijin had to sell broad-market products to keep the market in a “slow bull” phase. But broad-market ETFs contain more banks and other companies “just standing there,” meaning funds and the national team were selling the same holdings at the same time.
A market joke divides holdings into those “standing in the light,” those “only standing in the light” and those “standing there naked.” The first group chases hot technology; the latter 2 hold cyclicals, banks, brokerages or consumer stocks. Those latter groups are taking the heaviest damage.

23. An Index Reversal Could Produce Synchronized Recovery, While China Leads the Embodied-Intelligence Boom

After the SSE 50 recently added some fashionable technology companies that had already rallied sharply, selling the SSE 50 produces a different stabilization effect. If the hot sectors keep rising rapidly, new tools will be needed to prevent a few heavyweight stocks from lifting the entire index.
If US AI assets fall at some point and transmit the shock to A-shares, Huijin may buy broad-market ETFs again. Style-drifting funds may also sell high-priced technology stocks and buy back the large caps they previously abandoned. “Everyone draws blood together, and everyone recovers together.”
The final short take is that both the US and China have an embodied-intelligence boom, but China’s is clearly hotter. The US excitement is concentrated more in pure software; China’s most likely hot area is “putting software into the hardware chain,” embedding models in robots and manufacturing.

Nebius Co-Founder on AI Infrastructure Bubbles | How Price Elastic is Demand for Compute A 4-Hour Interview with 阳萌 on the Future of Consumer Electronics