August 18, 2026 (InvestinChina.asia) – The artificial intelligence industry must generate revenue on the order of one trillion US dollars by 2029 to cover its accounting costs and prove that the current capital-spending boom is not a bubble, according to a research note published Monday by Guojin Securities’ macro team led by chief economist Song Xuetao, with contact analyst Zhong Tian. The external financing gap, meanwhile, will not peak until around 2028—at a scale of roughly $700 billion to $800 billion per year—meaning concerns over free cash flow will linger for “at least several more quarters.”
The analysis reframes the AI bubble debate in hard numbers. Treating the AI industry as a single consolidated entity, Guojin estimates that by around 2030, nearly $1 trillion in annual revenue will be required just to cover depreciation, interest and operating expenses. The revenue test will not be passed by agent-based ARR alone; it must be met through a combination of business models, including other modalities not yet fully mature.
The Trillion-Dollar Threshold
Under Guojin’s neutral assumptions—capital expenditure growing 25%, 15%, 5% and 0% respectively from 2027 through 2030, with depreciation schedules of 8 years for short-lived IT equipment and 20 years for long-lived data center infrastructure—the AI industry would need to book $923 billion in revenue by 2029 and $1.18 trillion by 2030 to reach accounting breakeven.
The thresholds shift sharply with depreciation speed. Under an aggressive schedule (5-year/15-year depreciation), the 2030 revenue requirement jumps to $1.42 trillion. Even under the most conservative scenario (10-year/25-year depreciation), the industry still needs $1.06 trillion in 2030 revenue. Viewed from a CAGR perspective, assuming the combined ARR of major US model providers reaches roughly $200 billion by end-2026, the sector must deliver compound annual growth of 63.3%, 55.8% or 51.8% through 2030 depending on the depreciation scenario—figures that demand simultaneous breakthroughs on both the technology and application fronts.
Put simply: even under pessimistic capital-spending assumptions and extremely conservative depreciation, the AI industry must complete a “trillion-dollar leap” by no later than the end of 2030. Under more aggressive spending scenarios, that deadline pulls forward to 2028.
The $700–800 Billion Financing Wall
Guojin’s modeling of the external funding gap—derived by subtracting depreciation and stock-based compensation from annual capex, under the assumption that revenue exactly matches the accounting breakeven level—peaks around 2028 at a scale of $700 billion to $800 billion per year.
That estimate may be optimistic. The team notes that if actual revenue falls short of breakeven, the financing gap will be larger and its peak delayed. As a reality check: the idealized funding gap for 2026 stands at roughly $600 billion to $700 billion, and the market is already approaching that magnitude. Bond issuance in the first half of 2026 reached approximately $250 billion, with full-year bond financing tracking toward $500 billion. Adding nearly $100 billion in equity financing from Intel and Google—and excluding off-balance-sheet financing and direct credit—the tally already approaches $600 billion.
The implication is stark: the AI sector’s appetite for external capital is not a near-term phenomenon that will self-correct quickly. It is a multi-year structural demand that will test the capacity of global credit markets.
What Drives the Numbers
Guojin’s sensitivity analysis highlights three variables that materially move the breakeven target:
- Depreciation speed: The split between short-lived compute hardware (GPUs, CPUs, servers, networking—modeled at 65% of capex) and long-lived data center infrastructure (35% of capex) dictates how fast costs hit the income statement. Faster depreciation front-loads the pain and raises the revenue bar.
- Capex growth trajectory: Under the optimistic spending path (40%, 20%, 10%, 10% from 2027–2030), the 2030 revenue requirement rises to roughly $1.4 trillion. Under the pessimistic path (10%, 5%, 0%, -5%), it still demands $1.08 trillion.
- Interest rates: Perhaps counterintuitively, Guojin finds that rate levels themselves have a negligible impact on the AI bubble’s self-validation path. A 2-percentage-point swing in borrowing costs changes the cumulative interest burden by only tens of billions of dollars annually—a rounding error against the trillion-dollar revenue requirement. Rates are better understood as a symptom of industry sentiment than a cause of bubble resolution.
The Scale Economy Wild Card
The team also flags factors that could lower the eventual breakeven threshold: whether data center buildouts—combined with the medium-term easing of power-grid constraints as new generation comes online—produce scale economies, and whether policy interventions such as “high-quality compute load shifting” or “west-to-east compute transfer” (analogous to China’s West-East Power Transmission strategy) emerge to optimize asset utilization.
There are also deeper questions about the nature of AI itself: if compute is connected to the grid, does it necessarily train a “better” model? And how should “better” be defined—by parameter count, or by cost-efficiency within a specific capability envelope? The answers will determine whether compute-and-power consolidation produces genuine scale effects.
Reading the Signal
Guojin’s framing captures a market that has moved from fearing “0 to 1” to fearing “1 to 10.” The industry has already converted revenue from several hundred million dollars in 2023 to over $100 billion in 2026, with the technology evolving at extraordinary speed. Yet the sunk costs are accumulating just as the market pivots from asking “does AI work?” to “can AI pay for itself at scale?”
The note stops short of declaring the AI buildout a bubble. Instead, it argues that the trillion-dollar revenue threshold and the hundreds-of-billions financing gap are likely to “gradually become the baseline consensus for measuring AI bubbles” in the period ahead. For investors in the hardware chain—semiconductors, servers, power equipment, and data center infrastructure—the accounting breakeven level and the actual revenue growth path will serve as critical reference points for assessing the health of capital expenditure and the credibility of future earnings.
The bottom line is that AI’s ability to self-validate through visible revenue conversion is no longer optional. Whether through agent ARR, enterprise licensing, inference consumption, or new modalities not yet mainstream, the sector must collectively approach a trillion dollars in annual revenue by 2029. Anything less, and the financing gap—already at $600 billion this year—will widen further, extending the market’s free-cash-flow anxiety well beyond the near term.
Leave a Reply