The AI buildout runs on two clocks

@kawishwaqar|October 15, 2026 (2d ago)3 views

Disclosure: I work at Jazz as Principal Evangelist Cloud & AI. This article is written as personal industry analysis. Views are my own.

Last week I wrote about a number that finally went public: the AI buildout's committed spend, sitting in a filing I could read. This week the same trade split into two clocks running at different speeds, and the gap between them is the whole story.

One clock is Nvidia paying its supply chain years in advance. The other is the companies that would run all that hardware getting repriced by their lenders right now.

The supply clock: Nvidia is buying 2028 now

Nvidia's quarterly filing, for the quarter ended July 26, puts its supply-and-capacity commitments at $279bn, up from $119bn the quarter before. The Wall Street Journal covered it as a $279bn supply-chain gamble. Look at how it lands by year, because that is the part that matters: $92bn in what is left of fiscal 2027, $87bn in 2028, and $88bn in fiscal 2029. That last column is Nvidia placing orders now for components it will not sell for years.

Some of that money already has a name on it. On March 2, Nvidia put $2bn each into Lumentum and Coherent, the two biggest makers of the lasers used in optical links, alongside multi-billion purchase commitments and capacity rights. Coherent's own SEC filing shows the mechanics: 7.79 million shares at $256.80, $2bn in cash, plus access to five more product families tied to co-packaged optics.

That is the bet in one line. The next generation, Feynman, ships in 2028, and its largest system links 1,152 GPUs, sixteen times the 72 in today's top rack. At those speeds copper gives out around 2.5 metres, so the connections move inside the chip package as light. Nvidia is pre-paying its supply chain to have the parts ready on time.

I will state the caveat Nvidia states itself, because it matters. The filing says these agreements may be "cancelable, rescheduled, or adjustable," and that its AI-cloud partners "can unilaterally stop providing to us and sell to third-party customers at more advantageous rates." Committed is not the same as ironclad. But $279bn is the number the company is willing to put in writing, and $88bn of it points at 2029.

The financing clock: CoreWeave's 15%

Now the other clock, which runs faster and points the other way.

On October 13, the yield on CoreWeave's 9.625% senior notes due 2032 touched 15% intraday and closed near 14.5%. Its 2031 notes traded around 14.9%. A bond yielding five points above its coupon tells you the market has repriced this credit hard. The stock fell about 16% the next day to somewhere near $65, and it now trades more than half below its high. Chamath Palihapitiya posted the yield chart, Elon Musk replied "yikes," and the CEO told staff in a memo that the market was acting irrationally.

The operating numbers behind the bond explain why the repricing is not nothing. CoreWeave paid $640m of interest in the second quarter, against $267m a year earlier, on $2.58bn of revenue. Interest more than doubled in twelve months and now runs larger than adjusted operating income. Total debt is around $52bn by one count.

Keep the nuance I have kept all along: that 15% is on unsecured holdco paper. The same company's project-level debt for two Applied Digital facilities refinanced at A3, investment grade, with a springing guaranty. So this is a corporate-credit problem, not yet a project-finance collapse. But it is the corporate credit that signs the next rack order.

What the two clocks say together

Put them side by side and the shape of the risk changes.

The supply clock says the buildout is locked in years ahead. Nvidia is paying suppliers for 2028 and 2029 capacity today, and its competitors, Google, Amazon, AMD, are all moving to the same optics and much of the same supply chain, so those orders hold even if Nvidia loses share.

The financing clock says the marginal buyer is being repriced right now. The companies that exist to rent out these racks, CoreWeave, Nebius, IREN, are the ones the credit market just marked down. Nebius is close to 40% below its June high.

That is the real question in this thesis, and this week made it sharper. Does the 2028 commitment story defer the reset, because the money is already committed and the parts are already ordered? Or does the financing crack first, and starve the very orders Nvidia is placing? The supply side of this trade is being underwritten for a date four years out. The demand side is being repriced this quarter.

Certain, probable, unresolved

Certain, because it is in filings: Nvidia's $279bn of supply commitments and the $88bn for fiscal 2029. The Lumentum and Coherent investments, confirmed by Nvidia's own releases and Coherent's SEC filing. CoreWeave's bond move, its interest expense level, and its debt, corroborated across CNBC, Bloomberg and the wires.

Signal, not proof: that the CoreWeave repricing is the first domino rather than a single-company problem, and that financing stress will reach the order book. The A3 refinance is the counterweight, and I will keep saying so.

Unresolved: which clock wins. Nobody can date this. What I can say is that both clocks are now visible in public documents, which was not true a month ago.

What I am watching

Three things.

Nvidia's next commitment table. If the fiscal 2029 number keeps growing, the supply clock is still winning. If it flattens, or the language around cancellability hardens, the financing clock is starting to bite.

CoreWeave's next earnings: whether adjusted operating income starts to close the gap with that $640m quarterly interest bill, and whether the bond yield settles below 15% or keeps climbing.

And the November 23 air-permit decision for Project Jupiter, which I flagged last week. Permits are where a schedule either holds or slips, and slippage there shows up in leasing economics a quarter or two later.

Last week the story was that the spend had a number. This week it has a counter-number. The buildout is buying 2028 with both hands while the companies that would run it get a harder look from their lenders. I am not calling which clock wins. I am watching both.