This week's Intel Desk | AMD's data centre doubles yet the stock falls 9%; non-farm payrolls tonight

AI Intel Desk · Issue 1 · one readable piece of market intel every week

Every issue of the Intel Desk pulls scattered information into a single weekly brief, and marks the moments worth watching ahead of time

Intel dispatched — help yourself

📖 Key events this week (3–9 Aug)

ADP private payrolls 44K (expected 70K, prior 98K) came in far short; the ISM non-manufacturing employment sub-index at 47.4 fell below 50 into contraction; and unit labour costs rose just +1.3% in the initial estimate (expected +2.1%) — a "cooling labour market plus easing inflation pressure" combination that leans supportive for rate-cut expectations. But ISM manufacturing PMI at 55.6 came in stronger than expected and offsets some of that, so the market is still waiting on tonight's (7 Aug) non-farm payrolls to confirm the direction.

📌 Picked Q&As this week

Drawn from real Longbridge AI questions this week — the three topics users asked about most.

1. AMD's post-Q2 slump

What the market fears is never "fast growth" itself, but whether that growth is just an illusion created by contracts landing all at once. The AI disproved that doubt in one move with a six-quarter QoQ acceleration table — no jump, no fallback, just a steadily rising curve. In one line: not fireworks from short-term contracts, but a ramp of structural adoption.

The real question: AMD's capex came in at $808M vs analyst estimates of $299M — is this grabbing AI share, or over-expansion?

The AI's core answer:

• Eight quarters of capex: $132M → $808M (6× in a year, 4× in a single quarter)
• FCF/OCF still at 34% and the absolute figure at a record high — "this isn't expansion out of desperation; they're spending more because they're earning more"
• Three destinations for the capex: Helios rack-scale systems, MI450 tape-out, and CoWoS advanced packaging — all "building capacity", not "buying market share"
• The key call: not over-expansion, but a slope that implies a longer investment period — MI450 only lands in H1 2027, while Vera Rubin has been in volume production for a year

Image from a real Longbridge AI response.

2. SpaceX's first quarterly report since listing

An investment narrative is at its most dangerous when two attractive premises stack up — "subscribers doubled plus cash flow improving" sounds exactly like an inflection point already reached. The AI didn't get carried along; it recalculated P/S of 36× back up to 45×, and turned "FCF improved to −$1.1B" back into "H1 deteriorated to −$24.3B". In one line: whether it's expensive is a valuation question, but whether the valuation is even real is a data question first.

The real question: Starlink subscribers doubled and free cash flow improved from −$9B to −$1.1B — at 36× P/S, is it cheap enough?

The AI's core answer:

• Corrected the premises first: P/S is actually 43–45× (not 36×), and H1 cumulative FCF is −$24.3B (not an improvement to −$1.1B) — the question was based on figures circulating in the market, and the AI's first move was to check them
• SOTP valuation: a fair value of $309.8B–$490.6B versus the current $1,416.0B — a premium of 2.9–4.6×
• "Subscribers doubling" does not equal "revenue doubling": Starlink subscribers +68%, but ARPU fell from $85 to $66 (−22%)
• The key call: not cheap — two of the three premises are wrong, the SOTP breakdown shows a 3–4× premium, and FCF is still widening. "36× P/S is cheap" is an illusion propped up by a miscalculated valuation and a misread cash-flow narrative

Image from a real Longbridge AI response.

3. Palantir's Q2 results

To judge whether growth is real, don't look at how bright the quarter's number is — look at the shape of the curve. Contracts landing in a batch make it jump; sustained acceleration makes it climb. Palantir's QoQ growth climbed steadily from 7.6% to 22.5% over six quarters, and that is the hardest kind of curve to fake.

The real question: Palantir's US commercial revenue grew +153% — is this AI commercialisation genuinely taking off, or just contracts landing all at once?

The AI's core answer:

• Corrected the premise first: it isn't +153% — that's TCV growth; revenue actually grew +149%. The AI cleared up the metric the user had conflated in one line.
• Key evidence: nine quarters of QoQ acceleration (7.6% → 22.5%, with no "jump then fall back") + $4.9B of RPO locking in 12 months of visibility + revenue per top-20 customer +67%.
• The counterweight: the US alone accounts for 81% of revenue, non-operating income up +292% is propping up profit, and Ark has been trimming while Burry is short.

Image from a real Longbridge AI response.

✍️ Next week's calendar (10–16 Aug)

🎁 Comment to win

In the comments, share a conversation you've had with the AI

Rewards:

• Every valid entry earns288 Task Coins
• 10 valid participants are then drawn at random, each receiving anAI Pro benefits card (30 days) worth HK$148

Event period: from now until 14 August 2026, 7:00 PM. Rewards will be distributed within 15 working days after the event ends

AI Intel Desk · Issue 1

Pulling scattered information into one weekly brief, and marking the moments worth watching ahead of time. See you next issue.

* Nothing in this post constitutes investment advice. Longbridge reserves the right of final interpretation.