Recently, OpenAI's IPO plans have thrust the lab, long operating under a 'capped-profit' structure, into the spotlight of the public market. Meanwhile, Alphabet, Google's parent company, launched an $80 billion financing plan, with Berkshire Hathaway alone subscribing to $10 billion.
The conservative investment giant's first foray into tech stocks signals that the AI capital game has reached a new peak. Only now can we say that the AI industry is undergoing a profound paradigm shift.
The most direct manifestation is that "shortage of capital" and "spin-offs" have become the two dominant narratives for current AI companies.
The reason for the former is unsurprising: computing power is a heavy asset. Google's capital expenditure for 2026 is projected to be as high as $180 to $190 billion, with Microsoft, Meta, and Amazon also investing hundreds of billions. An H100 chip, a transformer for a data center, a grid connection – each step requires real money.
The latter has become a major strategic move for large domestic companies recently. Kuaishou's Kling AI had an internal valuation of around $6 billion within the group. After spinning off independently, its pre-money valuation skyrocketed to $18 billion, a threefold increase. Baidu's Kunlunxin was spun off for a separate listing, with external estimates suggesting it could contribute nearly $30 billion in incremental market cap to Baidu, equivalent to over 60% of its current total market value.
This phenomenon reflects the capital market's redefinition of AI assets. Within a conglomerate's consolidated financials, the AI business is seen as a profit-consuming investment. Once independent, it's priced based on the scarcity of the sector, revenue growth rate, and future potential, where price-to-sales ratios of several dozen times are not uncommon.
These two threads, seemingly independent, actually point to the same core: AI is transitioning from a technology-narrative-driven landscape to a new competitive environment dominated by capital efficiency.
The End of the Computing Power Race: The Breakdown and Reconfiguration of Financing Logic
Behind the "capital shortage" lies a fundamental logical chain. Today's large language model competition is essentially no longer a product competition but a heavy-asset race of computing power scale. OpenAI has committed to about $600 billion in future expenditure for computing power expansion. Even after completing a $122 billion financing round, this capital is expected to be depleted within three years.
More directly, OpenAI's CFO, John F. Ray, previously disclosed that while annualized revenue for 2025 has exceeded $20 billion, it's still insufficient to cover massive losses. The company incurs about $1.22 in losses for every $1 of revenue it generates.
The crux of the problem is that the cost curve of the AI business is fundamentally different from that of the traditional internet.
Adding one more user to WeChat brings marginal costs close to zero. But the more popular ChatGPT is, the more calls are made, and the higher the inference costs. User growth becomes pure benefit but also a cost pressure. This "anti-internet" business model means that economies of scale not only fail to bring profits but actually amplify cash flow pressure – user growth no longer directly equates to value growth.
Deeper still is the phenomenon of "circular accounting" in the AI era: Microsoft's $13 billion investment in OpenAI was not delivered in cash but in the form of "cloud credits." OpenAI uses these credits to train models, while Microsoft books them as new cloud revenue.
This closed-loop operation of "using investment to buy cloud services" appears on the surface as healthy revenue growth but is essentially paying oneself with one's own money and then classifying it as sales revenue. Estimates suggest OpenAI's annual cloud service bill has ballooned to over $60 billion, more than double its actual revenue of $25 billion.
This is the essential contradiction of the "capital shortage": the disconnect between valuation bubbles and actual cash flow. When investors start caring about "free cash flow" rather than "paper profits," the valuation system previously propped up by mutual investment promises and circular orders faces the risk of valuation deflation.
OpenAI plans for a $14 billion loss in 2026, expecting profitability only by 2029. Google's projected $180-$190 billion capital expenditure for 2026 indicates that the current AI "capital shortage" is not a cyclical liquidity issue but a dilemma of the entire business model at the capital structure level.
Why Is One Financial Statement Worth Three Times More?
One of the most noteworthy signals for 2026 is the concentrated spin-off of core AI assets by major tech companies.
Kuaishou's AI video product, Kling, plans a Pre-IPO round with a valuation of $20 billion, close to 70% of Kuaishou's parent company market cap. Concurrently, Baidu is pushing its AI chip company, Kunlunxin, towards a dual listing on A-share and Hong Kong markets, with 2025 revenue expected to exceed 3.5 billion RMB and achieve breakeven. Alibaba is reportedly planning to spin off Pingtouge, and ByteDance's Doubao could follow the same path at any time.
Think about it: before the spin-off, Morgan Stanley valued Kling at only around $6 billion. After the spin-off, targeting a $20 billion financing, the same assets, same revenue, same team – just by changing the financial statement – the valuation instantly differed by over three times.
The change in valuation logic here reveals a structural divide at the mechanism level: the primary market is different from the secondary market. Its game rules follow the unconventional pricing mechanism of "consensus determines value." The primary market looks at the future, sector positioning, imagination, and whether there will be buyers in the next round, but pays little attention to current profits and revenue.
The core logic for Kling's $20 billion valuation lies in the scarcity of such top-tier assets. After Sora's closure, the number of leading players in the AI video generation sector can be counted on one hand. The label of "AI infrastructure for the content industry" itself commands a premium.
So, what qualifies as a top-tier asset currently? In the current AI landscape, it's companies that possess self-developed foundation models (whether language, video, or multimodal), not wrapper or fine-tuning models; have proven at least one vertical scenario with large-scale users or revenue (not a demo, not proof-of-concept); and have "takeover expectations" for subsequent financing – either a strategic buyer (major tech company) or an IPO channel (US, Hong Kong, or A-share markets).
Companies meeting these three criteria globally can be counted on two hands: OpenAI, Anthropic, xAI, Google DeepMind (if independent), China's Zhipu AI, Moonshot AI, MiniMax, ByteDance's Doubao (if independent), Kuaishou's Kling (in spin-off process), Baidu's Kunlunxin (chip side). Each is a scarce target, each in a state of "buyers queuing, sellers raising prices."
The underlying logic of this "revaluation" is the cognitive shift of AI assets within major tech companies from "cost centers" to "value centers."
Within a conglomerate, the AI business is treated as part of the group's operations. Inside a major company, AI business is typically categorized as "strategic investment," meaning its costs (computing power, R&D, data annotation) are co-mingled with the group's mature cash flow businesses (like ads, e-commerce, gaming). The group CFO looks at consolidated statements. As long as the AI business is burning cash, it's constantly asked to explain "when it will contribute net profit."
In this context, AI teams are forced into short-term ROI justifications, and the valuation logic is naturally suppressed under the shadow of the group's overall P/E multiple – mature internet companies typically only get 10-15x P/E. Even high-growth businesses only enjoy a 20% premium, not the 3-5x P/S ratio of an independent sector.
Once spun off independently, the independent financial statements can redefine the boundaries of "costs" and "revenue." For example, the computing power costs previously consumed internally by the group can now be repriced at market rates as "related-party transaction revenue"; model training previously booked as R&D expenses can now be capitalized as "intangible assets" and amortized over periods.
In other words, these assets gain the pricing model of a "growth enterprise." The spun-off AI company can proceed with financing and strategy more flexibly, avoiding the constraints of internal resource allocation within the group, and receive independent pricing in the capital market based on its own growth prospects.
Simultaneously, this involves a further differentiation of valuation systems. The growth potential and forward-looking valuation of existing businesses in major companies, now tagged with AI, are beginning to command new premiums in the secondary market.
This also explains why traditional internet giants (like Baidu at $47.5 billion, Kuaishou at $27 billion) are being caught up with or even surpassed by AI newcomers in absolute market cap terms – Zhipu AI's latest market cap translates to approximately $58.6 billion, already surpassing Baidu to become China's ninth-largest AI tech stock.
From "Model Worship" to "Value Realization": The Industry Narrative Has Structurally Shifted
Some professionals suggest that the rapid development of the current AI era is reminiscent of the previous mobile internet explosion. This analogy is accurate, but the key difference lies in the essential difference in cost structures.
The mobile internet explosion relied on smartphone proliferation and continuously declining bandwidth costs – marginal costs trended downward. The AI explosion faces hard constraints like rising computing power costs, surging electricity consumption, and long data center construction cycles.
One observation is that the current AI industry is in a state of "85-degree water temperature – about to boil but not yet boiling."
The direction of technological breakthroughs (agents, multimodal) is clear, computing infrastructure investment is unprecedented (large US hyperscale cloud companies' 2026 capital expenditure will reach $805 billion, nearly double predictions from a year ago). But true commercialization, monetization, and adoption scale are still at a critical point of about to begin.
Currently, only a small fraction of CFOs saw actual financial value from AI in 2025, and even fewer Chinese enterprises achieved revenue growth through AI. This tension of "high investment, low return" is precisely the growing pain signal of the industry shifting from hype to practical implementation.
Many may not have noticed that the weight in the AI value chain has shifted from the GPU side to the entire system side. The latest Morgan Stanley research points out, "Agent AI marks a structural shift from computing to orchestration." In agent workflows, CPU-side orchestration time can account for 50% to 90% of total latency, leading to projections of an incremental $32.5 to $60 billion CPU market space by 2030.
This means the industry's core contradiction is shifting from "insufficient computing power" to "insufficient system efficiency." Correspondingly, the investment logic will expand from a "single-chip computing power race" to "full-stack system engineering." GPUs determine "if it can be done," but CPUs and systems determine "if it can be profitable."
If the mobile internet explosion was driven by "connectivity" as the core, then the AI explosion will likely have "intelligence" at its core, and the breadth of its value chain will likely surpass that of mobile internet, covering the entire chain of computing power, models, applications, and data.
Some economists point out that 2026 is becoming the singularity year for AI's leap from "assisting thinking" to "autonomous execution." The core contradiction at this stage is shifting from "who can train the strongest model" to "who can be the first to transform AI capabilities into tangible commercial value and user benefits in the most economical way, fastest speed, and widest coverage."
"It's not just about redefinition, but also about revaluation." Everything happening in the AI industry in 2026 – giants short on capital, frenzied financing, corporate spin-offs, IPO clusters – is essentially the concentrated release of the same capital logic: when the path of "burning money for growth" reaches its end, the industry must answer a most fundamental question: How much is this technology really worth?
The answer to this question will determine the power structure of the AI industry for the next decade. And 2026 is precisely the moment when this game of capital and technology fully unfolds.
This article is from WeChat public account "New Eyes" (ID: xinmouls), author: Li Xiaodong.
