In partnership with
Editor’s Note: We released June’s AI Upside Rankings last week and we have a new #1 pick.
THE BRRR’s BOTTOM LINE
I’m starting to believe that AI is the only trade that matters.
It’s driving nearly all of the US’ economic growth, and now that the government has deemed Anthropic’s latest model too dangerous for public usage, the sector comes with an implicit financial backstop.
Goldman’s baseline now implies $765 billion of annual AI capex in 2026, rising to $1.6 trillion by 2031 — roughly $7.6 trillion of cumulative buildout.
Meanwhile, just five companies — Alphabet, Amazon, Meta, Microsoft, and Oracle — have already raised $255 billion in debt and equity this year, and are expected to spend roughly $750 billion on AI data centers by year-end.
That is a Manhattan Project-style mobilization being accelerated by the most profitable and powerful corporations in the world.
Strip out the AI buildout and the economy looks a lot less impressive. Q1 real GDP was just +1.6% annualized, revised down from +2.0%, with the revision driven by weaker investment and consumer spending.
Real GDI rose only +0.9%, and the average of GDP and GDI was +1.3%. The St. Louis Fed’s AI tracker shows why this matters: AI-related investment categories contributed 0.97 percentage points to real GDP growth in the first three quarters of 2025 — 39% of total GDP growth, versus 28% for comparable IT categories at the peak of the dot-com boom.
In plain English: the economy is not booming; AI capex is booming. Without data centers, chips, software, R&D, and power-hungry infrastructure, the “resilient economy” story starts looking much thinner.
Then came Anthropic’s new model, Fable 5, released last week.
The release was met with awe, as the model comfortably topped all intelligence records and benchmarks.
However, on Friday night, the U.S. government forced Anthropic to disable Fable 5 after imposing export controls on national-security grounds, just days after launch.
That is the tell. Fable, without guardrails, had already been described by Anthropic’s own red-team work as capable of identifying and exploiting zero-day vulnerabilities across every major operating system and browser when directed by a user.
Fable 5 was supposed to be the sanitized public version — and Washington still treated it like controlled strategic technology.
Business Insider reported that Amazon raised jailbreak concerns, officials ran findings past the NSA, and the order’s “net effect” was that Anthropic had to abruptly disable the models for all customers. Translation: frontier AI is no longer normal software.
It is now a military-grade strategic asset with a government kill switch — and anything important enough to be restricted is important enough to be backstopped, accelerated, and defended.
Should the economy take a sudden turn for the worse or the technological progress decelerates enough to spook investors, the government will likely deem frontier AI labs too important to slow down on R&D, and will fund the buildout itself as it races against the rest of the world.
Let’s set AI aside for a moment to drill into the recent macroeconomic/ geopolitical news.
The market wanted one thing last week: permission to stop trading on the latest status update from the Strait of Hormuz.
By Sunday night, Trump had all but guaranteed a peace deal and the end of the Iran War, WTI was down to $81.36 (-4.15%), and the reported details of the U.S.-Iran MoU/near-deal chatter gave risk assets enough oxygen to look past the ugliest parts of the inflation tape.
Important caveat: this is still a headline market, not a signed-and-sealed peace regime.
Beirut/Israel/Hezbollah spoiler risk did not magically disappear because DJT wanted to host a UFC event at the White House on his birthday with good optics.
The surface tape was relief. SPX 7,430.86 (+0.51%), Nasdaq 25,889.74 (+0.34%), and Dow 51,202.60 (+0.71%) closed higher Friday.
Crypto bounced too, with BTC around $65.4K (+1.46%), ETH $1,721 (+2.42%), and SOL $70.40 (+2.16%), but Fear & Greed was still only 18 / Extreme Fear.
Translation: this was not euphoric animal spirits. This was the market buying a lower probability of immediate energy escalation and then immediately asking whether the Fed will ruin the vibe.
The Fed problem is real but is significantly softened by the end of the war. While the inflation data were not friendly, (May CPI printed +4.2% y/y) energy accounted for a disproportionate amount of the pain.
Energy rose +3.9% m/m, was +23.5% y/y, and accounted for more than 60% of the monthly all-items increase.
With the war ending and the Strait opening, energy prices will likely plummet in short order.
The deeper force, working in the Fed’s favor, remains AI/productivity/automation deflation.
The problem is the bridge to that future is not free. It needs data centers, power, cooling, transmission, chips, land, debt, equity issuance, export licenses, and enough political cover to keep selling frontier models across borders.
That is the week’s real trade. Peace is the market’s excuse. AI is now a financing trade and a sovereignty fight.
AI Deep Cuts
AI is still the structural deflation engine, but the path there now looks less like software magic and more like project finance with GPUs. That changes the investor map. Benchmark scores matter. So do credit spreads, long rates, grid access, customer prepayments, sovereign export controls, and whether public markets are willing to absorb the next wave of AI supply.
Oracle turned AI demand into a financing story: Oracle reported Q4 revenue of $19.2B (+21%), cloud revenue of $9.9B (+47%), and OCI/IaaS of $5.8B (+93%). The monster number was $638B of remaining performance obligations, up 363% y/y and $85B sequentially.
The catch: FY26 free cash flow was -$23.7B because OCI investment swallowed the income statement victory lap. Oracle raised $43B of debt + $5B of equity in FY26 and expects about $40B more debt/equity financing in FY27, partially offset by $75B of prepaid/customer-supplied hardware in large AI contracts. This is not “cloud growth.” This is AI infrastructure becoming credit plumbing.
Amazon confirmed the borrowing impulse: Amazon’s Jun. 8 SEC filing confirmed a $17.5B senior unsecured delayed-draw term loan with Citibank as administrative agent, with commitments expiring Sept. 30, 2026 and proceeds available for general corporate purposes. Reporting has tied the financing context to AI-driven capex, but the clean number is the SEC-verified loan.
I would not publish a combined “$31.5B in 48 hours” headline without resolving the Canadian bond-sale currency. The investable point does not need the hype math: hyperscalers are pulling AI demand forward with balance sheets, and the market now has to underwrite the cost of money.
AI became a sovereignty fight over the weekend: Anthropic’s Jun. 12 statement says a U.S. government export-control directive forced the suspension of Fable 5 and Mythos 5 access because foreign-national access could not be cleanly separated. The directive reportedly arrived at 5:21 PM ET.
TechCrunch/WSJ/Reuters reporting added that Amazon/Andy Jassy had raised model-security concerns before the crackdown. The immediate spillover was the point: India, described in reporting as Anthropic/OpenAI’s second-largest AI market after the U.S., suddenly had to debate dependence on U.S. frontier labs. That is not a product-launch wrinkle. That is sovereign AI becoming an actual access risk.
China supplied the mirror image: Meta is reportedly separating/unwinding its $2B Manus deal after Beijing demanded tighter control, with Manus cut off from internal systems and data sharing halted.
Put that next to the Anthropic/Fable cutoff and the direction is obvious: frontier AI access, training data, model security, and M&A are becoming geopolitical instruments. Investors used to ask who has the best model. The better question is who has the model, the permission, the compute, the market access, and the domestic political cover to keep operating.
Compute deflation is the uncomfortable bull case: The contrarian tension is that capex can inflate while unit costs collapse.
Cached reporting around Avataar/Varya cited a 5-second 720p clip generated in 45 seconds on H200 versus 1,230 seconds for Wan 2.2, with pricing around $0.005/sec versus $0.10+/sec peers — roughly a 20x cost gap.
Other reporting suggested smarter routing could put 80% of workloads on models that are up to 99% cheaper within 12–18 months, with Harvey/Fireworks routing cutting inference costs around 3x. These need direct source rechecks before final publication, but the investor implication is the whole ballgame: adoption can explode while pricing power compresses. Great for productivity. Messy for anyone underwriting frontier-model returns like monopoly rent.
Trade What BTC Does Next — Not Just Whether You Own It
Crypto never sleeps, and neither does Kalshi. Trade whether BTC ends the day above $100K, whether ETH pumps in the next hour, or whether the market closes green this week. No wallets, no gas fees, no exchange risk — just your read on the market. Institutional volume is up 800% in six months. There's a reason the smart money is here.
Trade $10, get $10 free to start.
Trade responsibly.
How was today's email?
Got feedback? Follow the writer on Twitter @frank_locascio and send a message.
The BRRR is meant for informational purposes only. It is not investment advice. Please consult with your investment, tax, or legal advisor before making any investment decisions.
