Morning Watchlist: Saturday Edition
A quick note from Behind the Markets
Policy headlines are not earnings.
Shanghai can lower a down payment. Washington can promise domestic minerals. Hyperscalers can announce another giant AI buildout. None of those facts tells you who ultimately earns a return.
This morning is about separating activity from economics — the difference between money moving through a system and a business actually capturing it. And one number frames the whole issue: a Wall Street Journal analysis published this month found that nine large technology companies carry roughly $3 trillion in off-balance-sheet commitments tied to artificial intelligence. That's about triple their combined reported leases and long-term borrowings.
Enormous activity. The question is always the same: who owns the risk, and who gets paid?
1) Shanghai Is Cutting Down Payments. That Does Not Fix China's Property Balance Sheet.
Shanghai will ease property rules — lower minimum down payments for some second homes outside the outer ring, changes to housing-fund rules, and subsidies for buyers who sell an existing home and purchase a new one — in an effort to revive demand.
That can unlock transactions. It does not repair developer balance sheets, restore household confidence, or finish an unfinished project. And notice the shape of the policy: this is a targeted attempt to restart one segment of one city, not a national rescue. Governments announce national rescues when they think a single large action will work. They announce down-payment tweaks for second homes outside the outer ring when they're trying to find something that moves without committing to the balance-sheet problem underneath.
The distinction that matters for investors is between businesses that earn from turnover and businesses that need a boom. Property managers, renovation suppliers, appliance and furnishing brands, mortgage servicers, and distressed-asset specialists can all earn from a transaction happening at all. Developers need prices and volumes to rise together.
If you go looking, the diagnostic questions are collections, inventory, presales, and — most importantly — how much cash is trapped in projects the company cannot finish or sell. A subsidy can create a sale. Only cash flow proves a business is healing.
The direct beneficiaries are Chinese-listed and Hong Kong-listed property and consumer names, where a U.S. investor takes on policy risk, disclosure risk, and currency risk simultaneously, in a sector that has destroyed a great deal of foreign capital over the past five years.
Bottom line: China is trying to buy housing activity, not announce a miracle. Follow the companies that collect cash from transactions and repairs, not those selling hope about a developer rebound.
2) The AI Boom Has an Off-Balance-Sheet Problem
A Wall Street Journal analysis of securities filings, published this month, examined nine companies — Meta, Alphabet, Amazon, Microsoft, Oracle, Nvidia, Broadcom, SpaceX, and AMD — and found roughly $3 trillion of off-balance-sheet commitments tied to AI. That figure covers uncommenced data-center leases, GPU and server purchase obligations, special-purpose-vehicle financings, and residual-value guarantees. It is roughly three times the combined outstanding leases and long-term borrowings those same companies report.
Some specifics worth holding onto:
Uncommenced leases alone totaled about $1.2 trillion — roughly four times the figure disclosed a year earlier.
Microsoft reported $329.1 billion of leases that had not yet commenced as of June 30.
Meta carries roughly $420 billion off-balance-sheet — nearly three times its reported debt.
Earlier estimates were materially lower: about $1.65 trillion (Nikkei Asia, July) and $2.3 trillion (Bank of America). The number keeps growing as disclosure improves.
Here's the mechanism, because it's legal, ordinary, and worth understanding rather than being scandalized by. A hyperscaler partners with an outside capital provider to form a special-purpose vehicle that owns the data center. Meta's "Project Beignet," a $27.3 billion SPV with Blue Owl for its Hyperion campus, is the template. Because the tech company doesn't hold a controlling interest, the SPV's debt never appears on its balance sheet — only a contractual commitment in the footnotes. Add a residual-value guarantee — if the tenant walks away and the asset sells for less than a guaranteed minimum, the sponsor covers the shortfall — and the SPV's debt can earn an investment-grade rating while the liability stays contingent and off the books.
This is not automatically bearish. Leasing power and capacity is rational when demand is visible. But it changes who carries the risk, and that's the entire investable question. Who owns the asset, who guarantees the lease, and what happens if model prices fall faster than usage rises?
Two things to add to that list.
First, follow the counterparties. Nvidia's August memoranda of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR aim to mobilize over $500 billion of third-party capital into compute financing platforms. Nvidia isn't the lender — those six firms are, often channeling insurance and other long-duration money. Nvidia may provide residual-value support of up to 25% of an opportunity, which could expose it to as much as $125 billion if fully drawn.
Second, notice where that money is coming from. Insurance floats and private-credit vehicles are funding a meaningful share of this. That same market is now a principal counterparty to trillions of dollars of AI commitments whose value depends on utilization forecasts nobody can verify. Two opaque systems have become load-bearing for each other.
When you look at any AI-adjacent operator, find out whether its growth depends on one anchor customer and a lease that only works at peak utilization.
Bottom line: AI capex is becoming a financing story. Follow the companies that underwrite demand honestly — not the ones hiding leverage behind a lease.
3) Lockheed Is Shopping for the Minerals Beneath the Missile
Lockheed Martin is negotiating to secure U.S. supplies of scandium and germanium as Washington pressures contractors to cut Chinese dependence. Specifically, per Reuters, it is in talks with NioCorp Developments for scandium, and Teck Resources and 5N Plus for germanium.
The scale of the dependence is easy to underestimate. The United States has not mined scandium since 1969. Rio Tinto is the only North American producer, with capacity of roughly nine metric tons a year. Global germanium consumption runs around 60 metric tons annually per the USGS — the entire world market is about the weight of a loaded semi-truck — and the U.S. imports more than half of what it needs. The policy pressure is real: last month an executive order made it harder for defense contractors to obtain the waivers that let them keep buying from China, with tighter procurement rules taking effect in 2027.
Now here is why this section is titled the way it is — this is a qualification business, not a mining story.
Look at the actual NioCorp arrangement. It's a preliminary agreement to supply roughly 15 metric tons of scandium a year from the Elk Creek project in Nebraska — a mine slated to open in 2028, with the agreement still requiring finalization. So the sequence is: a letter of intent, for a material the country hasn't produced in fifty-seven years, from a mine that doesn't exist yet, to be delivered two years from now, under procurement rules that haven't taken effect.
Every part of that is genuine progress toward supply security. None of it is revenue.
And that's the tension: the customer wants security of supply; the shareholder needs a return on capital. Those are not the same objective, and government support can fund the first while leaving the second unresolved. Reuters notes the hurdles plainly — Chinese suppliers have long been cheaper, and U.S. mining and processing capacity remains limited.
The most interesting structural detail is the one closest to the draft's thesis: 5N Plus received Pentagon funding this year to process germanium from recycled feedstock. No mine required. That's the processing-and-qualification bottleneck earning money, which is what the section says to look for.
Of the three named suppliers, one is a pre-revenue developer whose mine opens in 2028, one is a large diversified miner where germanium is immaterial to earnings, and one is a Canadian small-cap. There is no clean U.S.-listed way to own this bottleneck today — which tells you the qualification gap is real, and that the market has not yet produced a business that closes it profitably.
Bottom line: Strategic minerals are a qualification business, not just a mining story. Own the processing and verification bottlenecks before you chase the next deposit — and notice when no such business exists yet.
4) The Consumer Split Is Getting Harder to Hide
Two retailers reported within a day of each other and told opposite stories. Target raised its full-year outlook. Britain's JD Sports cut its profit forecast on weak U.S. sales. Same consumer, same country, different outcomes — which is exactly the point.
The Stock: Target (NYSE: TGT)
Target's second quarter, reported Wednesday (event risk cleared), was genuinely good — and it contains the single cleanest consumer datapoint this month.
Net sales rose 5.3% to $26.53 billion, ahead of the ~$26.15 billion expected. Comparable sales rose 3.8% against a 2.4% consensus. Store comps grew 2.7%, digital comps 8.7%, and same-day delivery more than 25%. Non-merchandise revenue — advertising through Roundel, Target Circle 360 memberships, and the Target+ marketplace — grew 20.1%, which is the highest-margin revenue in the building. Return on invested capital improved to 15.4% from 14.3%. This is the second consecutive quarter of comparable-sales growth after thirteen straight quarters of weak or negative comps.
Now the datapoint. That 3.8% comp was driven by a 3.6% increase in traffic — while average ticket stayed roughly flat.
Read that carefully. More people are coming through the doors, more often. They are not spending more per visit. That is precisely what a household does when its real income is falling — and we know it is, with inflation at 3.4% against wage growth of 3.2% for four consecutive months. Target has cut prices on more than 10,000 frequently purchased items over the past year. It is buying traffic with price, and the traffic is showing up.
Now the part that has to be separated out: a $994 million tariff refund. That one-time benefit contributed $752 million to net earnings and $1.65 to earnings per share. It's why headline net earnings doubled (up 100.8%) and operating income rose 94.4% to $2.56 billion.
Strip it out and the business still improved — adjusted EPS rose about 20% excluding the refund, and the underlying guidance midpoint rose $0.75. But be precise about the headline: full-year EPS guidance of $9.90–$10.90 includes that $1.65 of refund. Excluding it, the range is $8.25–$9.25. And the full-year operating margin guidance of roughly 6% includes about 90 basis points of refund benefit; ex-refunds, the margin is guided to about 50 basis points above last year's adjusted 4.6%. That's real progress. It is not a 6% margin business yet.
Three more things to weigh. SG&A grew faster than sales, with the expense rate rising to 21.6% from 21.3% — the cost of buying that traffic. Capital spending rose 27% to $1.4 billion. And Target repurchased no stock in the quarter despite $8.3 billion of remaining authorization — which, applying our usual buyback discipline, tells you management is prioritizing reinvestment over per-share engineering. That's defensible, arguably admirable, and worth knowing if you were counting on buybacks.
CEO Michael Fiddelke was careful with his words: "two strong quarters is not the goal," with apparel and home still needing work.
Verdict: a real turnaround with a genuine traffic signal, a headline flattered by a one-time refund, and a management team saying so themselves.
Bottom line: The consumer is not one trade. Underwrite the margin after promotions and refunds — not the traffic headline.
Before You Go
Activity is easy to mistake for progress.
A housing subsidy moves a transaction without fixing a balance sheet. A lease funds an AI buildout while moving the debt somewhere you can't see it — about $3 trillion of somewhere, at last count. A mineral announcement makes a supply chain sound secure when the mine opens in 2028. And a retailer can double its reported earnings while the actual improvement is a fifth of that.
None of those are frauds. Every one is a legitimate business decision. But every one also puts distance between the headline and the cash — and closing that distance is the entire job.
Ask who pays, who owns the risk, and what has to be true for the story to compound. Then find the number that would prove it.
Found this helpful? Share it with others.
Written by Behind the Markets