A Wartime Subscription Trap

If You Only Read One Thing

The expensive part of a dependency often appears when you try to leave it. Ukraine’s satellite predicament exposes the hardware cost of replacing Starlink; America’s inflation surprise raises the financing cost of commitments already made. These are separate shocks with a shared investment question: how much freedom remains after a military or a business has built around somebody else’s service?

Ukraine Cannot Switch by Subscription

Ukraine’s satellite disruption makes military procurement a question of continuing permission. Buying terminals and paying for connectivity does not give a customer control over the conditions under which that connectivity works.

Ukrainian officials reported sudden restrictions on military satellite service on October 14. By late evening, the defense ministry said connections had improved for drones moving at up to 160 kilometers per hour, but restrictions remained, according to RBC-Ukraine’s direct report. That is the terminal’s movement speed, not its internet download speed. The dispute concerns military use, including Starshield, rather than an established nationwide shutdown of ordinary Starlink access.

The motive is unresolved. Axios reports that a senior US official denied administration involvement. The operational dependency matters even without proving retaliation.

Switching costs here are physical. Think of replacing a phone carrier when the new network also requires a different phone, antenna and power supply. An alternative subscription is useful only after the equipment works in the same mission. For Ukraine, the relevant test is whether a replacement can fit on a drone and maintain the needed connection while moving. This is a hardware exit bill.

Reuters’ interviews with satellite specialists identify that bill: OneWeb faces terminal size, power and performance constraints, and existing Starlink terminals cannot simply migrate to its network. Eutelsat says it already provides connectivity in Ukraine. That establishes a supplier presence, not an interchangeable frontline service.

The strongest counterargument is that SpaceX’s performance advantage made dependence rational. A less capable alternative could reduce military effectiveness today. But operational superiority and dependable access are separate procurement attributes. October 12’s refinery bargaining concerned what Ukraine would agree to stop attacking; this development exposes a service dependency affecting what it can operate.

Over 12–36 months, I have moderate confidence that European satellite and compatible-terminal suppliers gain a stronger case for publicly financed orders. Eutelsat is a candidate beneficiary, but the near-term opportunity is funded adaptation, not immediate substitution or assured shareholder returns. Development costs could absorb the revenue. The commercial test is a paid procurement award specifying deployable replacement terminals; another sovereignty announcement would not meet it.

Inflation Raises the Exit Bill

September’s inflation report makes existing capital commitments harder to finance even where underlying demand remains strong. That distinction matters more than treating every falling technology stock as a verdict on AI adoption.

The Bureau of Labor Statistics release puts monthly consumer-price growth at 0.6% and annual inflation at 3.7%. Gasoline rose 9.1%, contributing more than half the monthly increase. Prices excluding food and energy rose 0.3%, with shelter up 0.4%. The evidence contains both an energy shock and continuing underlying pressure; it does not establish that every increase has the same cause.

Higher financing costs change the value of waiting. Consider a company that has promised to build a facility before customers begin paying. If borrowing becomes dearer, the company must provide more cash during construction or accept a smaller eventual return. Strong orders do not erase that interval. Existing fixed-rate debt does not reprice immediately; the exposure is new borrowing, refinancing and any floating-rate debt. A business able to postpone an uncommitted project has an option that a business already paying contractors has lost.

This updates October 8’s analysis of the Fed’s spillover concern: fresh price data now tests the concern recorded in September’s minutes. It is also the financial counterpart to Ukraine’s hardware exit bill. Changing direction becomes expensive after a system has been built around an assumption.

An immediate rate increase is not inevitable. ING economist James Knightley still expects December rather than October, citing soft wages and housing. His October 14 note put the market-implied probability of an October increase at 45%, up from roughly 20% before the release. That was a contemporaneous pricing estimate, not the Fed’s own forecast.

The investment read is moderately favorable to internally financed capacity owners relative to businesses that must repeatedly raise money before earning revenue, over the next 6–18 months. This qualifies the standing scarce-capacity argument: owning a bottleneck can produce attractive operating margins while financing its expansion destroys part of the return. ASML’s results below show why supplier strength and financial-market weakness can coexist.

The read weakens if financing becomes cheaper despite persistent price pressure. The next two inflation releases provide a separate operating test: whether monthly core increases cool below September’s 0.3% while fuel remains expensive. That would weaken the case that the energy shock is becoming a durable, economy-wide pricing problem.

The Contrarian Take

Everyone says: Falling chip shares mean the AI investment cycle is running out of customers.

Here’s why that’s incomplete: ASML’s filed results say planned supply will not fully satisfy customer demand in either 2027 or 2028. The sharper distinction is between demand for production and the price investors will pay for the resulting profits. A lower share price can reflect more expensive capital, expectations that were already excessive, or genuine deterioration; one trading session cannot distinguish them. The demand thesis would face a much stronger challenge from canceled equipment orders or delivery postponements than from a lower valuation alone.

Under the Radar

Google is contracting around the electricity queue. AEP announced a roughly $4 billion, 702-megawatt Wyoming fuel-cell project backed by a 20-year Google power-purchase agreement, with full operation expected by the end of 2028. The useful detail is that a named buyer supports a dedicated generation investment beyond AEP’s traditional service territory. Bloom supplies the technology; AEP carries a development role. That extends the power-bottleneck pattern behind October 7’s nuclear agreement, but with different equipment and execution risks. Neither the contract price nor realized project returns are disclosed in the announcement.
The factory boom can crowd out other factories. The Kansas City Fed’s October Beige Book reports that defense and data-center demand is displacing other production, while manufacturers extend hours because labor is scarce. These are regional business reports, not a national causal estimate. Still, the mechanism matters: new investment can rearrange existing capacity before it enlarges total supply. Suppliers with scarce skilled crews gain negotiating power; unrelated customers can face longer waits without experiencing any increase in their own demand.

Quick Takes

OpenAI faces eight separate state fronts. Iowa’s attorney general announced its suit alongside actions by seven other states, alleging consumer deception and harms from ChatGPT. These are allegations, not findings. Separate state cases mean a federal policy preference for voluntary safeguards does not settle the company’s exposure. The next competitive cost is defending product claims across jurisdictions, before any court orders a remedy. (Source)
Anthropic supplies an actual quarter. Reuters and other outlets report Q3 revenue of $16.8 billion and profitability on an adjusted basis, citing Anthropic’s disclosures to investors. That is stronger demand evidence than an annualized run rate. It still leaves the adjustments and future compute obligations unresolved; a profitable adjusted quarter does not establish that the next expansion finances itself. Public financial statements remain the missing test. (Source)
ASML earns from machines already installed. Q3 sales reached €11.7 billion and gross margin 57.5%. Management attributes the margin upside to strong demand for productivity upgrades, with service and field-option sales around €3.3 billion. Scarcity therefore rewards improving customers’ existing machines as well as selling new ones. That is a concrete way a capacity bottleneck produces current earnings before the next factory is completed. (Source)
TSMC’s profit is still accelerating. The chipmaker reported Q3 net income of about NT$763.1 billion, up 69% from a year earlier, Reuters reports this morning. That challenges an immediate collapse-in-demand interpretation of the semiconductor selloff. It does not establish that shares are cheap: investors are valuing future margins and investment needs as well as the quarter just completed. Current supplier earnings and future financing risk require separate judgments. (Source)

The Thread

Reversibility has a price that ordinary performance measures miss. A network can work well while being costly to replace; a factory can attract orders while being costly to stop building. My inference is that the freedom to change course becomes a separate economic asset once geopolitical conditions or financing terms move unexpectedly.

For valuation, that favors businesses whose next commitment is optional over otherwise similar businesses whose next payment is unavoidable. It also explains why governments might finance an inferior backup network: part of what they purchase is the ability to refuse a future condition. The return includes a choice preserved, which commercial revenue alone will not measure.

This argument weakens if Ukraine can restore dependable mission access through durable service terms without new hardware, or if committed developers can defer construction cheaply without losing customers. Those outcomes would show that contracts preserve more flexibility than today’s disruptions suggest.

Predictions

I predict: The Federal Reserve will leave its target rate range unchanged at its October meeting. September’s price data increases the risk of a later rise, but the countervailing wage and housing evidence supports waiting for another observation. This is a timing judgment, not a claim that inflation is solved. (Confidence: medium; Check by: 2026-10-29)

2026-10-15 · 03:25 AM ET