Author | Chen Wen

Source | Insight New Research Society

A financing frenzy, described by industry insiders as the 'night before the clear-out,' is unfolding in the global large model sector.

Before May even ended, three funding rounds totaling over $7 billion poured into the Chinese market: Kimi secured about $2 billion in early May, Jieyue Xingchen was reported to be nearing $2.5 billion in financing, and DeepSeek’s valuation soared to between $45 billion and $50 billion after its first external funding round.

In the U.S. and European markets, OpenAI, Anthropic, and SpaceX (post-merger with Elon Musk’s xAI) are expected to go public this year, with a combined valuation exceeding $3 trillion.

This trans-Pacific tidal wave of capital is surging toward the final top players in the large model sector at an unprecedented speed and scale. However, not all companies will receive funding—quite the opposite. For the vast majority, the music has stopped.

But for the select few that do secure funding, this could be their last ride to the next level.

01 The 'Musical Chairs' Game Nears Its End

Since the beginning of the year, China’s large model sector has delivered two major answers to the capital markets.

First, on January 8, after six years in operation, Zhipu Huazhang officially listed on the Hong Kong Stock Exchange, earning the title of 'world’s first large model stock' with an issue price of HK$1.16 billion per share. On its debut, the stock closed up 13.17%, reaching a market cap of HK$57.9 billion.

Just a day later, MiniMax, founded in early 2022, also listed on the Hong Kong Stock Exchange, surging 109.09% on its first day and breaking the global record for the fastest AI company to go public, with a market cap exceeding HK$100 billion.

Not only that, but their stock prices continued to rise post-listing. By May 15, Zhipu’s share price had soared from its issue price of HK$116.2 to a peak of HK$1,229, a more than tenfold increase in four months. MiniMax followed a similarly vertical growth trajectory.

JPMorgan Chase maintained an 'overweight' rating for both companies in a recent research report but offered a sobering assessment: the market’s valuation of them already implies that Zhipu’s annual recurring revenue (ARR) will reach $1 billion and MiniMax’s will hit $700 million by the end of 2026.

The frenzy in the secondary markets quickly spread to the primary markets.

On May 6, Yuezhiyanmian (Kimi) was reported to be nearing a new funding round of about $2 billion, with a post-money valuation exceeding $20 billion. This round was led by Meituan Longzhu, with participation from China Mobile, CPE (CITIC Industrial Fund), and others. Longzhu invested over $200 million.

Including three rounds of financing since the end of last year, Yuezhiyanmian has raised over $3.9 billion in six months, with total financing exceeding RMB 37.6 billion, making it the most funded domestic large model startup.

Another star company, DeepSeek, which made waves globally with its DeepSeek-R1 model in 2025, had previously adhered to a 'no external funding' policy. But this spring, the winds shifted.

According to a May 7 report by The Wall Street Journal, DeepSeek is raising billions of dollars from government-backed investors, with the National AI Industry Investment Fund in advanced negotiations to participate.

Sources revealed that Liang Wenfeng himself plans to invest RMB 20 billion from his own pocket. Industry estimates suggest the post-money valuation could exceed $50 billion.

Additionally, Jieyue Xingchen is reported to be finalizing nearly $2.5 billion in financing, having dismantled its red-chip structure to prepare for a Hong Kong IPO. Its investor lineup includes consumer electronics supply chain firms like Huaqin Technology, Longcheer Technology, and ZTE.

Shengshu Technology completed two major funding rounds in 2026: over RMB 600 million in Series A+ and nearly RMB 2 billion in Series B, totaling nearly RMB 2.6 billion in less than four months.

AI infrastructure provider Infinigence also announced on May 7 that it had completed over RMB 700 million in Series B financing.

Looking beyond China to the other side of the Pacific, the protagonists of this capital feast are even more massive.

Based on current public information, SpaceX is set to list on Nasdaq in June with a target valuation of $1.75 trillion. If successful, it would surpass Saudi Aramco as the largest IPO in human history. OpenAI plans to go public in the fourth quarter with a valuation of about $852 billion, while Anthropic also aims to complete its IPO this year, with a secondary market valuation already exceeding $1 trillion.

In just the primary market financings completed in February and March, OpenAI and Anthropic each secured hundreds of billions of dollars in funding. The combined valuation of these three giants exceeds $3 trillion, far surpassing any past tech IPO combination.

The core fact illustrated by this string of soaring numbers is that capital is concentrating irreversibly toward a tiny handful of top players in the sector.

Looking back, during the 'Hundred Models Battle' in 2023, hundreds of startups competed. By 2025, however, media statistics showed that only 22 financing rounds were completed by AI model layer companies throughout the year, totaling RMB 9.4 billion in disclosed amounts. The share of large model financing in total AI investment plummeted from 51% in 2024 to just 14%, with over 90% of the industry having been eliminated.

Yet, when over $7 billion poured into three top companies within three days in May 2026, the industry’s message became clear: capital is no longer 'blood-transfusing' the entire sector but is instead 'topping off' the final contenders.

02 The Economics of Token Factories

This capital boom did not arise out of thin air—it is driven by the dual forces of technological paradigm shifts and market logic reshaping. To understand it, we must examine both internal and external factors.

The industry narrative fundamentally shifted over the past year.

Before 2024, the core story of large models was 'bigger parameters, smarter models.' Major players burned money to train models, competing on intelligence ceilings.

But in February 2026, the emergence of long-range agents like OpenClaw (nicknamed 'Lobster') shattered the 'Pandora’s box' of computing power consumption. An agent handling a complex task requires dozens or even hundreds of model calls, with Token consumption skyrocketing from thousands in traditional single-round dialogues to hundreds of thousands or even millions.

The industry no longer competes on 'intelligence ceilings' but on who can produce massive Tokens at lower costs and more stably—what Nvidia founder Jensen Huang calls the 'economics of Token factories.' This is a industrial revolution driven by explosive real demand, supply-demand imbalances, and global computing power competition.

Data from the National Data Bureau clearly illustrates this 'brutal' explosion: China’s daily Token calls surged from 100 billion in early 2024 to 140 trillion in March 2026, a more than 1,000-fold increase in two years.

Since 2026, the A-share AI computing power sector has surged over 55% in cumulative gains, with top large model companies exceeding RMB 1 billion in monthly revenue. Some firms surpassed their 2025 full-year revenue in just 20 days.

The structural imbalance in supply has caused Token pricing power to shift dramatically upstream.

HBM high-bandwidth memory is monopolized by Samsung, SK Hynix, and Micron, with expansion cycles lasting 24 to 36 months, leading to a 40% HBM shortage in 2026. Electricity costs account for over 60% of Token production costs, while power infrastructure construction for large data centers takes 3 to 5 years.

This leads to a 'first-principles logic' defining today’s large model industry: large models are no longer just software but hybrids of 'software + cloud computing + heavy-asset industry.' Behind every user chat, search, and response lies real-time GPU and electricity consumption.

When a model’s 'marginal cost' no longer approaches zero, whoever controls the most computing resources and can produce Tokens at the lowest cost gains pricing power—and these resources are won not through algorithms but through hard cash.

At the macro level, massive AI infrastructure investments by international tech giants have intensified the industry’s focus on current competition.

According to the latest capital expenditure guidance released during the April 2026 earnings season, Microsoft’s full-year AI infrastructure capital spending is expected to reach $190 billion. Alphabet raised its full-year capital expenditure forecast to $180–190 billion, further increasing from February’s guidance. Meta also raised its forecast to $125–145 billion on April 29, citing rising component prices and data center construction costs. Amazon maintained its spending at about $200 billion.

Calculated at the upper limits, the four giants’ combined capital expenditures in 2026 will reach about $725 billion. Clearly, this is not just industry spending but the completion of a power supply system for a new intelligent era, laying the 'power grid' of computing power for all AI applications.

Meanwhile, the listing window effect from some startups has accelerated financing rhythms in the Sino-U.S. VC primary markets. The explosive post-listing gains of Zhipu and MiniMax established reference benchmarks in secondary markets for 'how much large model companies are worth,' fueling anxiety among unlisted firms. If they fail to price themselves during this window, Valuation correction (valuation corrections) may occur once market sentiment fades.

Thus, Jieyue Xingchen completed its red-chip dismantling, joint-stock restructuring, and Hong Kong IPO sprint ( sprint : sprint listing) within months. Kimi’s valuation soared from about $4.3 billion to over $20 billion, reflecting both fundamental improvements and the capital scramble to secure tickets for the 'next listed company.'

03 The Future Decisive Battles

On one side is capital frenzy; on the other, a shift in competition focus. The industry widely believes future battles will center on three areas.

First, commercialization will become the 'top priority' for all companies.

It must be recognized that a fundamental shift is occurring in the large model industry in 2026: the 'AGI premium' is cooling.

Over the past two years, capital markets justified high AI company valuations on a key implicit premise: Scaling Law would remain effective, model capabilities would rapidly advance with computing power investment, and AGI was only a matter of time. Investors were willing to accept short-term losses and discount 'future efficiency revolutions' into today’s stock prices.

But by 2026, while AI continues to progress, its trajectory appears less uniform—OpenAI revised its principles to reduce direct mentions of AGI; DeepMind’s Demis Hassabis openly admitted that current systems still have significant gaps in continuous learning and long-term planning.

The market’s focus has shifted from 'who is closer to AGI?' to 'who can make customers pay? Who can reduce inference costs? Who can survive?'

In fact, commercialization signals from some top players are already clear. ByteDance’s Doubao, which long relied on a free model with 345 million monthly active users, quietly launched a paid plan of up to RMB 5,088/year on the Apple App Store. OpenAI significantly strengthened Codex’s paid enterprise capabilities while actively restricting top-tier usage for free users.

This marks the global large model industry’s transition from burning money for traffic to a rational maturity phase, where the core competition shifts from 'whose model is stronger' to 'whose model monetizes first.'

Second, computing cost becomes the ultimate KPI.

As the large model industry evolves, inference capabilities, long texts, and multimodality will no longer be moats. After DeepSeek V4 brought open-source models close to GPT-4 levels, the industry systematically realized for the first time that model capabilities are easier to catch up to than imagined.

As models gradually 'commoditize,' capital markets ask: What else do you have besides the model?

This has triggered a shift in industry narrative.

In 2023, the focus was on 'bigger parameters, longer contexts'; today, companies discuss which terminals they’ve locked in, which supply chains they’ve bound, and which user entry points they control.

JPMorgan Chase noted in a research report that the market’s valuation of Zhipu already implies an expected ARR of about $1 billion by the end of 2026. Under the new evaluation framework, a company’s value is judged not just by benchmark scores but by its clients, cash flow health, number of paid scenarios unlocked, and its irreplaceability among partners.

Third, the agent explosion and pathway divergence.

2026 is widely seen as the 'year of the agent explosion.' While we focus on the quantity and speed of agents released by vendors, the more critical issue is the future divergence between ToB and ToC pathways.

One path follows 'embedding into production processes,' betting on determinism (certainty) in productivity gains. The other heads toward personal real-life scenarios, betting on mindshare and long-term scale.

Neither path is right or wrong, but they differ drastically in capital consumption rhythms and requirements for business model maturity. Serving enterprise clients demands a 'iron triangle' of reliability, integration, and security—a long-term trust-building process. C-end scenarios rely on data flywheels and self-reinforcing user mindshare, burning money early but offering strong scale effects later.

Against the backdrop of soaring computing bills and financing concentration reaching new heights, whether companies can achieve closed loops and positive cash flow in their respective lanes will directly determine their rankings after the 'night before the clear-out' in 2026.

04 Conclusion

For today’s investors, the choice is no longer about 'which direction to invest in' but about 'final-round bets' on a limited number of top players. Technical routes, scenario choices, and capital endurance—these three variables will jointly determine who stays at the table and who is shown the door.

In an era where models are increasingly commoditized, the true decisive factors may no longer be technological capabilities alone but how to transform those capabilities into services customers are willing to pay for continuously, how to turn computing investments into verifiable outputs, and how to evolve a product into a healthy company.