What is the key variable that determines the AI bull market?

```

Oil prices are above $100/barrel, the Strait of Hormuz has not yet been reopened to normal traffic, inflation and interest rate pressures are resurfacing, and expectations for Fed rate cuts are becoming more vulnerable. According to the traditional macro framework, this is not the most comfortable environment for high-valuation tech stocks. Yet, U.S. stocks are hitting new highs, and AI-related investments continue to be chased by capital.

Hongta Securities macro analyst Song Xuetao pointed out in a May 25th report: “The current AI boom is in a phase of rational frenzy—there’s already a bubble, but it is not out of control.”The key in this statement is not “bubble,” but “rational” frenzy: Agentic AI is evolving from an assistant tool to an autonomous execution tool, letting the market clearly see for the first time how AI can close the commercial loop from “burning money” to “making money.”

The rational side is that Agent application proliferation is bringing rapid growth in token consumption, inference computing demand, and leading vendors’ ARR. The frenzied side is that valuations have already priced in growth expectations for 2027–2028. As of May 20, the forward P/E ratio for the U.S. “Magnificent 7” stocks was about 35x, vs. about 25x for the other 493 S&P 500 firms. The implied premium is not based on ordinary growth stock logic, but on AI penetration speed reaching 5–8 times that of previous technological revolutions.

But what truly determines whether the AI bull market can continue is not a single quarter’s earnings, nor one blockbuster app, but three variables: in the short term, it’s about liquidity shocks—especially oil prices, inflation, interest rates, and yen carry trade unwinds; in the medium term, it’s about industry delivery—whether AI penetration speed matches current valuations; in the long term, it’s about harder constraints like energy, power grid, employment, social resistance, and hardware tech breakthroughs.

Agent moves from “copilot” to “autopilot”—the market starts rewarding capital expenditure

In the last round of AI trading, the market’s biggest worry was that tech giants were spending too fast: huge investments in data centers, GPUs, and cloud infrastructure, but revenue recovery paths were unclear. The change with Agentic AI is that it’s no longer just a Copilot-like assistant, but is evolving toward an Autopilot-style autonomous execution tool.

This brings two outcomes.

First, token consumption is accelerating again. The first wave of demand after GPT came from model performance upgrades; the second wave after agents was from an explosion in inference compute. Autonomous task execution means longer context, more complex steps, and more frequent model calls; inference is no longer an afterthought post-training but is becoming the main battlefield that continually consumes computing power.

Second, revenue expectations have been revised higher. After the proliferation of flagship agent applications like Openclaw and Claude Cowork, model vendors’ annual recurring revenue (ARR) is also growing rapidly. Cited mid-year estimates show Anthropic’s full-year ARR expectations have been revised up from $9 billion at the start of the year to $44 billion, doubling on average every six weeks. If the trend continues, next year’s ARR could surpass $300 billion.

This explains why the market no longer simply punishes capex. As long as revenue growth is fast enough, capital expenditure becomes a moat rather than a burden. Nvidia, Broadcom, as well as the hardware chains of optical modules, storage, etc., thus regain support.

With oil above $100, why can AI assets still rise?

This AI sector rally against rising oil prices isn’t because macro risks have disappeared, but because several forces are temporarily outweighing the risks.

First is demand diffusion across the industry chain. Inference stage doesn’t just need GPUs—CPUs, optical modules, and storage are also drawn into the high prosperity logic. 800G/1.6T optical modules are in short supply, high-end storage demand is rising. LightCounting forecasts that in 2026, shipments of 800G transceivers will double, and 1.6T port shipments will grow from a small base in 2025 to tens of millions, with 1.6T chipset sales topping $2 billion in 2026 and continuing high growth over three years.

Second is tech giants’ strong performance. In Q1, S&P 500 EPS growth was about 27.1%, the highest since Q4 2021, with Meta, Alphabet, and Amazon contributing 70% of incremental index earnings. As long as these heavyweights keep making money, oil shocks will only delay downward index pressure.

Third is America’s increased reliance on AI infrastructure for growth. In recent quarters, AI infrastructure investment has contributed more than half of U.S. GDP growth. Macroeconomic totals like payrolls and retail are still okay; while labor structure has diverged, as long as the overall doesn’t weaken obviously, the market is unlikely to quickly shift to stagflation trading.

There’s also a more direct reason: Big tech is less sensitive to oil prices than airlines, couriers, railroads, chemicals, autos, and tourism. They care more about electricity prices than oil prices. When traditional industries are squeezed by oil prices, capital is more likely to cluster around AI assets, mixing “safe-haven” and growth trades.

Valuations have already priced in the good times of 2027–2028

The danger in the AI trade lies not in a lack of industrial support but in how quickly the market prices it in.

U.S. “Magnificent 7” have a forward P/E of 35; the other 493 S&P 500 firms are at 25. The value gap implies a very smooth future: AI infrastructure keeps expanding over 3–5 years, and demand for computing, cloud, data centers, and semiconductors stays strong; AI keeps penetrating ads, search, cloud services, office software, code generation, finance, customer service, research, content, etc.; revenue contribution and productivity gains are achieved together.

But tech revolutions are rarely so orderly. Electricity took about 40 years to go from invention to large-scale production lines, and the computer about 25 years. At today’s market-implied AI adoption speed, it would have to commercialize 5–8 times faster than those general-purpose technologies.

This isn’t impossible, but the margin for error is slim. If AI business adoption lags capex, if inference demand doesn’t catch up to training, or if depreciation and electricity start to erode profit margins, valuations will respond first. A correct industrial direction doesn’t mean stock prices can be infinitely forward-looking.

The biggest short-term risk: rates rising faster than ARR

The real short-term pressure comes from liquidity.

If the Strait of Hormuz remains closed for an extended period and oil stays above $100 or rises further, inflation will spread from energy to services, transport, and raw materials. In April, U.S. PPI rose 9.8% y/y, highest since October 2022. Once inflation is entrenched, the Fed’s policy path will be forced to change.

Swap markets are already pricing in 0.8 Fed rate hikes this year, and even more hikes (over 2 times) from the ECB and BOE. Meanwhile, Fed leadership change and internal FOMC disagreements are also undermining faith in future easing.

Japan is another gray rhino. Japan has long been the global funding pool for leveraged trades, but yen depreciation and inflation pressures are forcing the BOJ to send tightening signals—30-year JGB yields have risen above 4%. If Japanese funding costs keep rising and trigger a global carry trade unwind, high-valuation AI assets will be in trouble.

There was a dress rehearsal on May 15: 10-year U.S. Treasury yields broke 4.5%, 30-year broke 5%, crowded momentum trades cooled, the Philly Semi index fell about 4% in one day, and the Nasdaq fell about 1.5%. This is not conclusive evidence of trend reversal but shows how sensitive crowded trades are to rates.

The most critical short-term comparison is simple: Can ARR (annual recurring revenue) be revised up faster than interest rates rise? If not, capital may retreat to more certain hardware segments; if liquidity keeps worsening and AI income expectations can’t be raised further, valuation pressure will increase substantially.

Harder medium/long-term issues: organizations, energy, employment, hardware

The midterm test is industry delivery. General-purpose tech revolutions are usually not straight-line rises but “accelerate, decelerate, re-accelerate.” There’s first a capital wave, then organizational adaptation, and finally productivity is released. The early internet also saw investment booms, capex expansion, and asset bubbles; real productivity improvement appeared years later.

The tough thing about AI pricing now is it almost requires corporate structure to rapidly adapt, workers to rapidly retrain, business models to quickly work, and for society not to strongly resist. This speed is rare in human history.

Long-term constraints are harder.

First is energy and infrastructure. AI data centers need huge amounts of electricity and cooling water; grid expansion, transformers, and energy storage are not just items on PowerPoint slides but real bottlenecks. If AI infrastructure keeps pushing up social electricity costs, regulatory and societal backlash will intensify.

Second is employment and consumption. In the short term, AI can boost enterprise efficiency and reduce demand for engineers, customer service, etc.; but if tech-driven layoffs outpace new job creation, consumer spending power gets hit. Productivity gains on the B2B side ultimately depend on C-side consumption. If non-AI sectors fall into recession, AI alone can’t stay on top for long.

Third is social acceptance. There was a craze for Openclaw in China early this year, but in the U.S. there’s rising resistance to data centers driving up electricity prices and tech-induced job loss. This will affect AI penetration speed.

Fourth is hardware tech disruption. If there’s a breakthrough like a “DeepSeek Moment” drastically improving compute, storage, or transfer efficiency, today’s tightest hardware bottlenecks might suddenly become glutted. The hard tech sector’s prosperity logic isn’t unbreakable.

The AI industry’s long-term outlook remains optimistic. If we don’t consider social problems from tech layoffs and changed production relations, AI does have a chance to boost total factor productivity and help the economy escape stagflation. Even if financial markets deleverage mid-cycle, the resulting data centers, low-cost technologies, and validated applications could be the foundation for the next industrial expansion.

But stock pricing is not the same as industry vision. What this AI bull market needs most to verify is whether the currently bet-on ARR, ROI, and tech penetration can actually be delivered when oil, inflation, interest rates, and social constraints have all hardened. Direction explains why there’s a bull market; only delivery speed decides if the bubble will get out of control.

Risk Reminders and DisclaimerThe market carries risks—invest cautiously. This article does not constitute individual investment advice and does not take into account the specific investment goals, financial situation, or needs of any particular user. Users should consider whether any opinions, views, or conclusions herein fit their situation. Acting on this content is at your own risk. ```