Nvidia Closed Green on the Day the Jobs Beat Sank Everything Else — the AI Capex Trade Has Stopped Waiting for the Fed

Stocks that discount distant cash flows are the first to break when the discount rate jumps. The Sept 4 setup was textbook: a hot payrolls print re-priced the front of the curve, traders moved toward a September Fed hike, and long-duration equities sold off. The S&P 500 gave back 29 points, the Nasdaq 77, the Dow 272. NVIDIA did the opposite — the stock extended its post-earnings rally, closed at $230.36, and finished within $5.38 of its $235.74 record set May 14. The VIX ended the session at 14.53, gold fell 1.4% to $4,476, and crude held above $91 — a risk-off tape with one conspicuous green tick.

The catalyst behind the move is a print the market treated as proof that demand is still outrunning supply. NVIDIA reported $96.22B of Q2 FY27 revenue on Aug 26, 2026 — up 106% year over year, ahead of the $92.07B consensus. Data Center alone delivered $89.0B (+117% YoY, +18% sequentially), with adjusted EPS of $2.22 versus $2.09 expected. The market responded with the second-largest single-day market-cap gain on record the next session: shares added roughly $400B on Aug 28, 2026, a 6.8–8% jump, after Reuters and CNBC reported the print.

The guide is what mattered: $108B for Q3 alone is roughly $432B annualized before the FY28 step-up. On the call, management said Vera Rubin will represent about 20% of Data Center revenue in Q3, putting next-generation capacity on the run-rate within two quarters of shipping. The bull case is no longer a forecast — it is bookings already in the system. And because NVIDIA's outlook excludes China entirely, the September-quarter number has no upside cushion from any thaw in export controls.

The BIS Annual Economic Report 2026 puts the five largest hyperscalers at more than $1 trillion of AI-related capex across 2025 and 2026 combined. That is the denominator NVIDIA is selling into. The deals on top — OpenAI's 10-gigawatt letter of intent (first gigawatt shipping in H2 2026 on Vera Rubin, per OpenAI), NVIDIA's August 17 commitment of up to $105B in credit and compute for an Ohio OpenAI campus, and Anthropic's $35B Lambda compute deal announced Aug 31 — turn that denominator into multi-year backlog rather than discretionary spend.

If the AI capex cycle is rate-insensitive, it is also capacity-constrained. Every NVIDIA accelerator that ships has to be paired with high-bandwidth memory and advanced packaging — and both have been locked in years ahead. SK Hynix and NVIDIA on Jul 25, 2026, announced a $500B+ strategic initiative covering AI factories and next-generation memory, including the NVHBM stack that the company says delivers 30% higher bandwidth and 15% lower power than commodity HBM4e. Micron's entire 2026 HBM output is sold out under multi-year fixed-price contracts, with HBM4 ramping in Q2 CY26 — meaning every gigawatt of OpenAI or Amazon build-out through 2026 is already booked into a supplier's P&L, not an open order book.

TSMC's monthly CoWoS capacity is heading toward 120,000–140,000 wafers in 2026, per TrendForce, and the foundry is adding another 50,000–60,000 wafers of new capacity through year-end. Even at that run-rate, NVIDIA is estimated to hold roughly 60% of the wafer book — about 595,000 wafers for the full year — with Broadcom-anchored programs (Google TPU, Meta MTIA) capturing most of the rest. TrendForce expects the CoWoS supply-demand gap to narrow from roughly 20% to 10% by the end of 2026 as the new lines come online — still a shortage, not a glut.

Short-term (days to quarters): the next test is the Federal Reserve's September decision, where traders are now pricing roughly a 62% chance of a hike. A surprise cut would compress the decoupling narrative (everyone rallies); a hike would re-test whether the Sept 4 divergence holds. The next NVDA catalyst is the Q3 FY27 print in late November, where $108B is the new floor and Vera Rubin's ramp to ~20% of Data Center revenue is the bar to clear. Anthropic's IPO filing, reportedly imminent after Labor Day, is the next event-driven read on the equity stakes embedded in NVIDIA's $10B commitment.

Long-term (one to three years): the structural argument is that AI capex has decoupled from the rate cycle because the binding constraint is no longer capital cost but physical capacity. NVIDIA trades at a 28.9x trailing P/E and 24.9x forward, against the analyst consensus target of $327.13 — implying roughly 42% upside even if Q3 prints in line. The bottleneck above the foundry — TSMC CoWoS and SK Hynix/Micron HBM — is the place where supply tightness turns into multi-year revenue. The risk that re-couples the trade to the Fed is not a slower AI buildout; it is a credit event that disrupts the hyperscaler debt-funded capex. That is the line to watch, not the front of the Treasury curve.