This is one of those days where the entire macro and tech complex is rolled into a single 12-hour window, and most people are going to miss the actual signal because they’re staring at the wrong line item.

By the time markets close on Wednesday, you will know:

Whether the Fed is still pretending it has the luxury of patience.
What four of the five biggest spenders in the global economy plan to do with another $625 billion of AI capex.
Whether the OpenAI revenue miss leaked to the WSJ on Tuesday actually changes the picture.

If you only watch the headlines, you’ll see “Fed holds, Powell vague, tech beats or misses.” That isn’t the story. The story is the structural shift in where wealth is being built right now, and what your portfolio should be doing about it.

What the Fed will actually say (and what it won’t)

CME FedWatch is pricing today’s decision at roughly a 100% probability of a hold in the 3.5–3.75% range. So the rate itself is irrelevant. The signal is in the language and the dot plot.

CPI re-accelerated to 3.3% year over year in March, up sharply from 2.4% in February, mostly because the Iran war supply shock pushed energy back into the inflation print. That number gives the Fed exactly the cover it wants to do nothing while the political pressure to cut is screaming. Powell is in his home stretch as Chair. He is not going to spend that political capital on a meeting nobody expected anything from.

What to actually listen for:

Whether the Committee softens its language on labor. Cracks in the labor data are the only thing that gets a cut on the table this year.
Whether the median dot still implies one cut in 2026. If even one dot moves out, the curve reprices and bond duration takes a hit.
Whether Powell is willing to call the inflation reacceleration “transitory.” If yes, expect risk assets to rip. If no, expect a flat reaction and a focus on earnings.

The wealth read: this Fed is no longer the protagonist of the market story. AI capex is. The Fed is now the slow-moving constraint, not the catalyst.

The number nobody is repeating enough: $625 billion

Microsoft, Alphabet, Amazon, and Meta are projected to spend a combined ~$625 billion on AI and cloud infrastructure in 2026 alone. Add Oracle and the figure for the top five U.S. hyperscalers reaches around $720 billion. About three quarters of that is AI-specific, putting pure 2026 AI infrastructure spend at roughly $450 billion.

To put that in scale: that single year of capex from five companies is larger than the entire annual GDP of countries like Norway, Argentina, or Israel. It is bigger than the global venture capital industry’s total deployment in any year on record.

When this much money moves into one supply chain, it does not matter much whether the consumer-facing AI app of the moment hits its quarterly user target. The capex commitments are structural, multi-year, and contractually locked. The companies receiving that money — chip designers, foundries, data center developers, power generators, network equipment makers, cooling systems, real estate — are the actual beneficiaries.

This is the picks-and-shovels logic, and it is not a secret. But it is being repeatedly mispriced because every time a frontier lab stumbles, the supply chain gets sold off as if the buildout has been cancelled. It hasn’t. (Related reading: Anthropic Just Hit $30B ARR in 15 Months.)

The OpenAI miss is a feature, not a bug

The Wall Street Journal reported Tuesday that OpenAI missed its own internal user-growth and revenue targets, with reports that CFO Sarah Friar has expressed concerns internally about funding the company’s massive compute agreements. Oracle dropped about 4%, Nvidia 1.5%, Broadcom and AMD harder.

Two things are true at once.

True thing #1. OpenAI is the highest-profile AI lab in the world, and its growth curve is the closest thing to a referendum on consumer AI demand. A miss matters at the headline level.

True thing #2. Anthropic — its closest competitor — has been adding enterprise customers paying $1M+ per year at a rate that has roughly doubled, and Google’s Gemini has been gaining share in late 2025 and into 2026. Cheaper inference and better models from China (DeepSeek V4, Moonshot Kimi) are making AI more usable, not less.

What that combination tells you: the AI economy is rotating, not collapsing. The frontier model layer is becoming more competitive, which compresses margins for any single lab and pushes the value capture down the stack — to chips, to clouds, to specialized agents, and to companies embedding AI into existing software people already pay for.

If you are long the names that get paid every time an AI workload runs — regardless of which model wins that day — you are positioned correctly. If you are long a single brand, you are exposed to exactly the kind of headline that hit Tuesday. (See also: DeepSeek V4 Is About to Drop — And Nvidia’s Moat Just Got a Hairline Crack.)

What to actually watch in tonight’s earnings

Forget the EPS beat-or-miss line. Watch these four numbers:

Capex guidance. Did Meta, Google, Amazon, and Microsoft hold or raise their 2026 AI infrastructure spend? If any of them raise, NVIDIA prints a new high before the open. If any of them cuts, the entire AI supply chain gets a 5–10% reset.
Cloud growth rates. Azure, GCP, AWS — are they still showing 25%+ YoY growth on bigger bases? That growth is the real proxy for AI commercial demand, much more reliable than any frontier lab user count.
Operating margin. Capex this large is not free. Margin compression is the cost of the buildout. Markets have been forgiving of that. If they stop being forgiving — say, on a Meta or Amazon — the rotation out of names with thin AI ROI is going to be sharp.
Buybacks. Mature mega-caps with massive cash flow are buying back stock alongside spending hundreds of billions on capex. That is not a contradiction. It is what wealth-creation at scale looks like.

The wealth read for your portfolio

You don’t need to predict who wins the AI race. The mistake most retail investors make is trying to pick the next OpenAI. The smarter move is to be exposed to the buildout regardless of which lab pulls ahead.

A simple way to think about it:

Picks-and-shovels: semiconductor names (NVDA, AVGO, TSM), networking (ANET, CSCO), power (Constellation, Vistra, NRG, GE Vernova), data center REITs (DLR, EQIX). These get paid as long as the capex flows.
Distribution layer: the hyperscalers themselves. They control where the AI workloads run. The capex is the price of admission to keep that monopoly.
Application layer: companies embedding AI into software people already pay for. This is where consumer/enterprise revenue actually shows up. Think productivity, security, finance, healthcare, legal.

The structural thesis is unchanged: AI capex this large always means a multi-year supply chain super-cycle. The Fed is not going to derail it. An OpenAI revenue miss is not going to derail it. A bad Mag 7 print might cause a 10% drawdown, but the buildout is contractually locked through at least 2027.

What you actually have to do is be exposed and stay exposed. Track your progress. Stop trying to time the entry.

Want to track your own path to financial independence and see exactly how much progress you’re making toward your FU money number? The Wealtharian Wealth Tracker lets you monitor your net worth, FU money progress, and investment milestones in one place. Try it free →