Author: Tidal Investment
I. The Market Changed Its Script
Lately, the market is both excited and a bit nervous. SpaceX completed a massive $750 billion IPO, and OpenAI and Anthropic are also rumored to be preparing for listings. At the same time, Alphabet plans for an $80 billion equity financing round, and Meta is arranging new funding.
Frankly speaking, seeing so many giants collectively asking the market for money, few can stay calm. But interpreting this wave as AI peaking is a bit too simplistic. This looks more like AI's story has turned to the next chapter.
Over the past two years, the market was buying demand explosion and industry imagination, concerned with whether AI would work. By 2026, the question has become: How long can such massive investment intensity last?
Wu Shaokang, founder of Tidal Investment, said: "The market always sees fast-moving variables, but what determines the direction of a cycle is often the slow-moving variables."
Standing in mid-2026, we remain optimistic about the AI industry chain. But today's optimism can hardly be sustained by imagination alone. Talking about AI two years ago, you could talk about models, about AGI, but talking the same way today, the market might not buy it.
II. Money is Still Being Invested, and More Aggressively
How to tell if a cycle is over? See if the money people are still paying. Looking through the books of the five major cloud providers gives a pretty clear answer.
-
Alphabet: 2025 Capex $90 billion, 2026 guidance raised to $180 billion.
-
Amazon: 2025 Capex $130 billion, 2026 guidance raised to $200 billion.
-
The remaining three are also moving in the same direction: Meta 2026 guidance raised to $140 billion, Microsoft to $190 billion, Oracle FY26 already near $60 billion.
These numbers are a bit scary when laid out. People used to think the strongest aspect of these internet giants was their solid cash flow and plentiful cash on hand. But now, even they are starting to proactively ask the market for more money in the face of AI. Besides that $80 billion equity financing, Alphabet has also issued a significant amount of debt over the past year. AI infrastructure has become so massive that even companies with the best cash flow are restructuring their capital.
Money is still flowing in, that's not in question. The problem is, how long can this continue?
III. Why This Investment Cycle Won't Stop Easily
What are people most afraid of? Afraid that Capex will peak, afraid this cycle will run its course in two or three years like past tech hardware procurement cycles, then enter a long digestion period. Servers, phones, PCs—many hardware cycles followed this pattern: demand rises, then capacity expands, inventory piles up; once downstream slows, the whole industry chain faces valuation cuts.
This concern was not wrong in past cycles. But this round of AI Capex is probably not so simple.
First, the money is flowing to too many places. The cloud providers' money is all called Capex on the surface, but breaking it down reveals they are completely different categories: compute, memory, networking, power. Each layer has its own expansion rhythm and its own bottlenecks. And with engineering projects, once started, pulling back halfway is more costly than pushing through.
More critically, bottlenecks are shifting from chips down to more physical constraints. Chip shortages can eventually be resolved by expanding production. But power, transformers, high-power-density racks—these don't ramp up as quickly. Just getting a grid connection can take years of waiting.
And Capex is far beyond just GPUs now. Clear signals are coming from the supply chain: Eaton, which makes power distribution equipment, saw data center orders surge 240% YoY in Q1 2026.
Activities like transformers, UPS, liquid cooling, thermal management, rack integration only appear in large volumes when cloud providers are determined to build campuses. The simultaneous explosion of these orders shows that behind this Capex wave lies solid construction progress.
Putting these together, you can see this investment cycle won't stop easily.
IV. What the Market Is Actually Worried About
We're optimistic, but we can't ignore two current market concerns.
Concern 1: Capex Growing Faster Than Revenue, Will ROI Be Realized?
The Capex growth rates of the five major cloud providers in 2025 all outpaced their revenue growth. Alphabet's depreciation rose from $15.3 billion in 2024 to $21.1 billion in 2025, a real 38% increase hitting the income statement. Amazon stated plainly in its earnings report that the FCF decline is due to AI investment pushing up PPE.
A popular market saying is that when Capex growth exceeds revenue growth, it signals peak ROI. This isn't wrong, but applying it to the cloud business is a bit simplistic. AWS, Azure, GCP all went through phases in the early 2010s where Capex far outstripped revenue, eventually monetizing through scale. The difference with this AI Capex round is the higher capital intensity; payback depends on whether future AI workloads can be monetized.
Of course, we're not blindly optimistic. To change our view, we'd need to see a few things: cloud providers starting to lower Capex guidance, order cancellations or delays, or AI product revenue and usage falling short of expectations. As of mid-2026, none of these have happened.
ROI risks certainly exist, but the current facts lean more towards optimism. If data truly starts to deteriorate, then it's time to revise judgment. We're not there yet.
Concern 2: Is This Another 2000?
How did the 2000 bubble burst? Back then, demand was also rising, with more people online and more traffic flowing every year. The problem was on the supply side.
There was a saying: Internet traffic doubled every 100 days. Telecom companies believed this curve and feverishly laid fiber along railways and roads. There was a cost advantage with fiber: once the trench was dug, adding more cable cost little extra, so they crammed in enough capacity for over a decade. Dozens of companies dug their own trenches simultaneously, resulting in supply far exceeding demand. Consequently, fiber prices crashed to the floor. By the time rising traffic filled them up, a decade had passed, and most of those companies couldn't survive that long.
There is certainly a bubble element in this round too. Any major cycle has some froth; some companies ride the AI wave, some money looks excessive in hindsight.
But on the supply side, it's the opposite this time, because AI doesn't just need a pipe buried. Transformers are customized heavy equipment, constrained by silicon steel and lengthy approvals. Grid connections can't be parallelized like digging trenches; they must queue behind the public grid for years. More importantly, electricity can't be pre-laid like fiber; you can't install electricity needed ten years from now and just leave it there.
So a 2000-style collapse is unlikely to repeat this time.
V. The AI Show Isn't Over
Just in the past couple of days, SpaceX has corrected sharply from its highs, even falling below its IPO closing price, making the market nervous again. Seeing so many giants collectively ask the market for money easily makes people tense, wondering if AI is peaking.
But we don't see it that way.
Giants are raising massive funds now because the show must go on, and the hurdles ahead are increasing. Look at the five cloud providers: none have lowered their 2026 Capex guidance; all have raised it. Looking further ahead, transformers take up to four years for delivery, data centers wait years for grid connection. These hurdles aren't easily overcome by just throwing more money at them.
So while a wave of financing looks scary, it's essentially just an intermission.
Don't rush to call the top. The AI show isn't over yet; it just switched scripts.
