Internal Memo: The System Is Working Exactly As Designed
Wednesday, March 4, 2026
To: Everyone still holding enterprise software
From: The Market
Re: You were warned
Let's reconstruct the last ten days with the dispassion of a forensic accountant, because that's what this moment requires. Not panic. Not cheerleading. Just a cold-eyed read of what happened, in sequence, and what it means for the next six months.
Event one. On February 20th, the Supreme Court ruled 6-3 in Learning Resources Inc. v. Trump that the International Emergency Economic Powers Act does not authorize the President to impose sweeping, open-ended tariffs. The majority opinion, authored by Chief Justice Roberts, leaned on the major questions doctrine — the legal principle that consequential policy decisions require explicit congressional authorization. Decades of executive branch mission creep, condensed into a single ruling. Markets initially celebrated: the dollar softened, the euro climbed toward $1.18, sterling held $1.35, and suddenly the word "refund" entered the vocabulary of 2,000 importers who filed suits at the Court of International Trade within days. FedEx, Costco, Revlon, Reebok — all in line. Between $130 billion and $175 billion in tariff revenue is now potentially at stake, with the Federal Circuit fast-tracking the question of repayments.
The celebration lasted roughly seventy-two hours.
Event two. The White House, within hours of the ruling, invoked Section 122 of the Trade Act of 1974 and imposed a new 10% global tariff — subsequently raised to 15% — under a provision that requires congressional approval to extend past 150 days. The tariffs must end in 150 days unless Congress votes to extend them, putting a vote squarely in the calendar a few months before the 2026 midterms. The administration had lost the legal battle and immediately reframed it as a tactical retreat to higher ground. Markets, correctly, did not treat this as a resolution. The DXY retreated to 97.65 as investors priced a "policy risk premium" — the legal foundation of U.S. protectionism was shifting beneath them in real time.
Event three. Nvidia reported. Fiscal Q4 revenue came in at $68.1 billion, up 73% year over year, with data center revenue hitting $62.3 billion — a record, up 75% from the prior year. First quarter guidance was set at $78.0 billion. By any normal measure, this is an extraordinary business. Net income for fiscal 2026 was $120.1 billion on 71% gross margins and 60.6% operating margins. These are pharmaceutical-level margins on semiconductor hardware, which shouldn't be possible, and yet.
The stock fell anyway. Revenue and guidance both beat analyst predictions, and the stock declined. The market wasn't judging the quarter. It was judging the architecture of what comes next.
Here is the structural tension that nobody in the financial press is explaining clearly enough, so let's do it properly.
The AI trade, as it has been constructed since 2023, rests on a simple sequential logic: chips → infrastructure → software → productivity → earnings. Nvidia owns step one completely. The hyperscalers — Alphabet, Amazon, Microsoft, Meta — are pouring capital into step two at a scale that is genuinely staggering. Combined capex for the four major hyperscalers could approach $700 billion this year. That number should stop you cold. $700 billion. Annually. Into data centers, fiber, cooling systems, and the chips that fill them.
And yet, simultaneously, the companies that were supposed to benefit from all that infrastructure — the enterprise software layer, the SaaS platforms that sit above the compute — are in free fall. Salesforce has dropped 21% year-to-date, ServiceNow 26%, Adobe 22%, and Intuit 37%. Wall Street has taken to calling it the "SaaSpocalypse," which is a terrible word but an accurate diagnosis.
The bear case is simple: if agentic AI can execute tasks that used to require a Salesforce license, a DocuSign subscription, a ServiceNow workflow — why pay for the software? The agents will just do the thing. The bull case — championed loudly by Nvidia's CEO — is that software companies will adapt, that AI will expand the total addressable market rather than cannibalize it, and that the selloff is a category error by investors who are confusing disruption with obliteration.
Both arguments contain truth. Neither resolves the uncertainty. And in the absence of resolution, $700 billion in annual capex is flowing into the infrastructure layer while the application layer bleeds out. Someone is right. Someone is going to look very stupid in about eighteen months. The problem is that both positions are currently being held at scale, by serious people, with serious money.
Then the Iran war started, and all of the above became a secondary concern for a week.
But here's the thing about wars and markets: the immediate price action is noise, and the structural repricing is the signal. The noise — Kospi -12%, XLE +8%, Bitcoin liquidations, gold whipsawing around $5,300 — that's already happened. The signal is what gets embedded.
The signal, in this case, is energy as a permanently higher cost input for the industries the AI buildout depends on. ISM prices paid surged 11.5 points to 70.5 in February — the sharpest single-month jump since March 2022 — before a barrel of Middle Eastern crude became a geopolitical instrument. Now layer an oil shock on top of that. Historically, readings above 70 in ISM prices paid presage CPI spikes three to six months out. We are looking at a potentially painful summer for consumers, which means a politically constrained Fed, which means rates staying higher for longer than the equity market has priced.
The probability of a June rate cut has already fallen below 50%. The futures market doesn't fully price a cut until late Q3. If the oil shock persists — and the Strait of Hormuz situation is not obviously resolving — that timeline drifts further.
This is the compounding problem. It's not one thing. It's the Supreme Court ruling creating tariff uncertainty, layered on top of a prices-paid surge driven by Section 232 steel and aluminum duties, layered on top of a Middle East war premium on energy, layered on top of a software sector in an existential identity crisis, layered on top of a Fed that cannot move in either direction without making something worse.
Each of these is manageable in isolation. Together, they form a system under stress.
There's a fiscal dimension that deserves more airtime. The SCOTUS ruling doesn't just create legal uncertainty — it creates a fiscal hole. Yale's Budget Lab estimates remaining tariff policies will raise about $1.2 trillion over the next decade, a figure that drops dramatically compared to what IEEPA authorities would have generated. The White House's trade revenue assumptions for the budget were almost certainly built on a more aggressive tariff regime. Now Section 122 — a 150-day stopgap with a congressional vote landmine embedded in it — is doing the heavy lifting.
Treasury Secretary Bessent has said this will result in "virtually unchanged tariff revenue in 2026." This is a remarkable statement that is either genuinely true, in which case the Supreme Court ruling was a political loss with limited economic consequence, or it's aspirational spin, in which case the fiscal math for the One Big Beautiful Bill Act gets harder just as midterm pressures mount.
The 150-day clock starts now. Mark your calendars for late July, when Congress will be asked to vote on extending tariffs weeks before the August recess and months before November elections. That vote — whichever way it goes — will be its own market event.
The DXY at 97.65 is worth sitting with. The dollar has been structurally weakening for months, and the SCOTUS ruling added another layer to the narrative: U.S. policy unpredictability is itself a dollar headwind. When the legal scaffolding of the world's largest economy can be dismantled by a 6-3 court decision on a Tuesday afternoon, the "safe haven" premium embedded in dollar assets takes a haircut. Not a catastrophic one. A quiet, persistent one.
The euro at $1.18, sterling at $1.35, a yuan that Beijing has been allowing to gradually firm against the dollar — these aren't dramatic moves. They're a slow repositioning by currency markets that have decided the dollar's structural premium needs recalibrating.
Gold above $5,300, central banks still buying, mBridge processing tens of billions in non-dollar settlements. The direction is not subtle, even if the speed is gradual.
What I keep coming back to is the $700 billion number. $700 billion in annual hyperscaler capex, flowing into infrastructure for an application layer that is currently in a bear market. Either the application layer recovers — software companies adapt, find their AI revenue, stop bleeding — or the $700 billion is, at some point, questioned. Not abandoned. Questioned. And when the market starts questioning the capital cycle of the most important investment theme of the decade, the Nvidia multiple — 40 times earnings on a $4.3 trillion market cap — becomes a conversation.
Jensen Huang says the software selloff is "the most illogical thing in the world." He may be right. He's also the man who benefits most from the capex continuing, so there's a natural incentive to say exactly that.
The market, right now, is holding two incompatible beliefs simultaneously: that AI infrastructure spending is rational and will compound into future productivity, and that the application layer those investments are meant to serve is in existential decline. One of those beliefs will have to break.
The war buys a few weeks of everyone looking at oil instead of NVDA's forward multiple.
That window won't last.