Q3 2026, year over year: revenue vs headcount.
Meta: revenue +26%, headcount -15%
Alphabet: +25% vs +8%
Amazon: +12% vs +4%
Among these three, Meta shows the widest gap between revenue growth and headcount growth. These numbers alone don't show the cause.
1
14
Amazon's quarterly interest expense tripled: $538M to $1.69B (Q3 2025 to Q3 2026).
It raised $115.1B of long-term debt in the first nine months of 2026. Interest is still only 6% of Q3 operating income, but borrowing costs are now a visible line on the income statement.
1
20
CoreWeave grew revenue 168% in Q3.
But its financing costs tell another story.
$207M in GAAP operating income.
$846M in net interest expense.
And management expects $1.05B–$1.15B in interest expense next quarter.
The bull case: a $132.5B backlog and contracted demand support the buildout.
The bear case: financing costs could absorb much of the upside.
For $CRWV, what would convince you the economics are improving?
Faster operating profit growth, lower financing costs, or stronger free cash flow?
And why?
1
3
57
Alphabet reports $924.4B of purchase commitments, 3.5x its $265.1B of cash and marketable securities.
These are not near-term spending: $213.4B is due within 12 months, most is multi-year through 2031, and some supply contracts are stated at maximum price and volume.
30
Q3 2026 YoY, operating vs net income:
Alphabet: +30% vs -24%
Amazon: +61% vs -2%
Microsoft: +21% vs +32%
The drivers differ:
Alphabet: an unrealized loss now (mgmt: SpaceX)
Amazon: $9.5B pre-tax Anthropic gains a year ago
Microsoft: a $3.1B OpenAI hit to net income a year ago
32
Microsoft's commercial backlog (RPO) is $677B, 7.3x its last-quarter revenue.
By schedule: roughly 30% is expected to be recognized within 12 months, and that part grew 29%. The part beyond 12 months grew 98%.
That is contract timing, not realized conversion speed.
17
US investment in information processing equipment is 2.8% of GDP, yet its one-year rise equals ~15% of nominal GDP growth.
BEA Q3 2026, annual rate: $933.9B vs $663.3B a year ago; GDP +$1.83T.
Note: GDP subtracts imports; BEA's release doesn't say how much of this was imported.
21
Three hyperscalers added $208B of long-term debt in nine months.
Dec 31, 2025 -> Sep 30, 2026:
Amazon: $65.6B -> $173.1B
Alphabet: $46.5B -> $125.1B
Meta: $58.7B -> $80.9B
Nine-month capex rose 1.7x-2.1x at each. Alphabet says its new notes are for "general corporate purposes."
21
Similar growth headlines. Different growth engines: Eaton reported 28% Q3 sales growth; Hubbell reported 24.6%. The acquisition bridge changes the comparison.
For the quarter ended September 30, 2026:
• Eaton: 19 percentage points organic growth, 8 from acquisitions, 1 from FX.
• Hubbell: 11.3 points organic, 13.2 from acquisitions, 0.1 from FX.
Roughly 68% of Eaton's reported growth came from its organic component, using the rounded bridge. At Hubbell, acquisitions supplied about 54% of the dollar increase in sales.
That doesn't make acquired growth bad. It means headline growth isn't a clean measure of underlying demand.
The electrical businesses still show strength: Eaton's Electrical Americas grew 22% organically, Electrical Global 27%, and Hubbell's Electrical Solutions 22%.
My takeaway for the AI-infrastructure watchlist: separate existing-business growth from purchased growth before comparing suppliers.
These are diversified businesses, not pure AI revenue streams. Organic growth includes pricing, and the companies' definitions aren't perfectly interchangeable.
A weekend earnings comparison—not a fresh market-moving announcement.
Sources:
sec.gov/Archives/edgar/data/…
hubbell.gcs-web.com/news-rel…
sec.gov/Archives/edgar/data/…
51
Read this as a capital-allocation signal, not just a product obituary.
Musk says the Cybertruck line ends in ~8 months not because the truck is unprofitable, but because, in his words, the value of incremental Cybercab production vastly outweighs the opportunity cost of Cybertruck's complexity. That's his framing; it doesn't settle how much sales trends or the truck's own economics mattered.
Musk also says orders run well ahead of production, and that Tesla will have to return deposits to a lot of buyers because it can't build all of them before the planned shutdown.
The metrics we'd watch now: how many Cybercabs Tesla actually builds, and paid miles per vehicle. These are our analytical yardsticks, not Tesla-reported KPIs.
The Cybertruck order rate far exceeds our production rate. We will try to make as many as possible before shutting down the line in about 8 months, but will have to return deposits for a lot of people.
This is not a case of Cybertruck being unprofitable, as has been misreported by legacy media, but rather that the opportunity cost of the complexity of Cybertruck is vastly outweighed by the value of incremental Cybercab production.
1
1
113
Revisiting the Nov. 5–6 announcements—not Sunday breaking news: Crusoe reported a Wyoming site acquisition and the completion of its pre-development engagement.
Crusoe says Google acquired its southeast Wyoming site. Crusoe also says its pre-development engagement was completed, covering site selection, governmental approvals and a power strategy developed with Tallgrass Energy.
Google’s Nov. 5 announcement describes its first Wyoming data center but does not mention Crusoe. The acquisition account comes from Crusoe’s Nov. 6 statement, not confirmation by both companies.
These statements do not establish that development rights, permits or construction risks transferred. This is a single case, not evidence of an industry-wide trend.
Terms weren’t disclosed in these announcements. The acquisition price, Crusoe’s profit or loss, transaction financing and allocation of subsequent construction risks remain unknown. Completing pre-development does not establish that the project’s remaining risks have been eliminated.
Sources
crusoe.ai/resources/newsroom…
blog.google/innovation-and-a…
69
$ABNB offers an AI datapoint beyond capex.
In Q3, about 40% of all support issues were resolved without a human, up from one-third in Q2. Support cost per booking fell over 17% YoY, largely driven by its AI assistant, the company says.
That’s measurable unit-cost progress—not proof that AI alone drove overall margin expansion.
73
Gold ETFs added $2.7B in October, yet assets fell 6% to $542B, says the World Gold Council. Holdings rose to a record 4,275 tonnes. Lower prices outweighed inflows. Falling AUM is not the same as investor selling—and persistent buying is not proof of a price floor.
1
2
55