Macro Highlights This Week
The December FOMC delivered a widely expected 25bp rate cut and announced roughly USD 40bn in short-term Treasury purchases to replenish system reserves. However, the market’s real focus was not the policy action itself, but Chair Powell’s rare admission that current job growth may be “systematically overstated.”
The latest dot plot shows a sharply widened divergence in views on the 2026 rate path. While the median still points to one 25bp cut, the range now spans from no cuts to as much as 150bp of easing, highlighting a lack of consensus among policymakers on the economic outlook.
At the corporate level, Oracle’s earnings triggered the market’s first broad-based concern over AI capex returns. Weak cloud revenue combined with sharply higher capital spending drove the stock down nearly 11%, weighing on the Nasdaq and prompting investors to reassess the structural mismatch between upstream compute investment and downstream demand monetization.
As a result, valuation premiums across the broader tech sector came under pressure.
Big Tech saw heavy selling on Friday. As of the Dec 4 close, more than half of the large-cap tech names were down over the past week, with Nvidia and $Alphabet(GOOG)$ both falling around 4% by Friday’s close.
Big Tech Weekly Core View — After Oracle and Broadcom earnings dragged Nvidia lower, does Nvidia's leading postion still hold?
$NVIDIA(NVDA)$ was dragged lower by $Oracle(ORCL)$ and $Broadcom(AVGO)$ earnings headlines. Despite this, Bank of America reiterated its bullish stance, maintaining a Buy rating, reaffirming Nvidia as a Top Pick, and reiterating a $275 target price.
According to Reuters, demand from China for Nvidia’s H200 chips has exceeded current production capacity, and Nvidia is evaluating an increase in H200 AI chip output. Nevertheless, NVDA still saw a sharp sell-off on Friday.
BofA highlights Nvidia’s valuation at 25x / 19x 2026E / 2027E PE, with a PEG of just 0.5x, which they view as significantly undervalued versus the Mag 7 and broader growth stocks (average PEG ~2x). The firm believes Nvidia’s technological and commercial leadership in AI compute continues to strengthen.
The Blackwell architecture is expected to deliver a 10–15x performance improvement over Hopper. Models trained on Blackwell are expected to roll out in early 2026, potentially widening Nvidia’s performance lead further. The next-generation Rubin architecture remains on track for 2H26, with CPX accelerators entering mass production in 4Q26.
On the commercial side, Nvidia sees dual visibility on both demand and supply supporting at least $500bn in revenue potential for 2025–2026. Additional letters of intent (LOIs) from OpenAI and Anthropic/Microsoft represent potential upside.
Compute utilization remains extremely high. Even five-year-old Ampere GPUs are still operating near 100% utilization, reinforcing a 5–6 year GPU lifecycle and signaling that AI capex is still in an expansionary phase. Gross margin guidance remains unchanged at the mid-70% range.
Bernstein’s latest observations on AI data center architecture note that training workloads require significantly more memory than inference.
Training mid-sized models can consume ~1TB of memory, while inference primarily relies on KV cache and temporary tensors, with far lower resource requirements. Tight HBM/DRAM supply continues to drive price increases.
TPUs offer cost advantages in energy efficiency and scalability, but GPUs retain an edge in general AI workloads due to their mature ecosystem and software support. High-bandwidth flash and other emerging non-volatile memory solutions are forming a new layer between DRAM and SSD, improving inference-side performance and energy efficiency.
Since the launch of ChatGPT, Nvidia’s quarterly revenue has accelerated sharply—from $5.9bn in 3Q22 to $57bn in 3Q25, nearly a tenfold increase in three years. Spillover demand effects from large-scale AI training and inference continue to strengthen.
Overall, Nvidia maintains a strong position across architecture roadmap execution, ecosystem stickiness, supply chain control, and demand visibility. Short-term competitive or regulatory noise has not altered its core role in global AI infrastructure.
That said, the close relationship between OpenAI, Oracle, and Nvidia has raised market concerns that negative developments at Oracle or OpenAI could spill over to Nvidia—explaining why Oracle’s earnings had such an outsized impact on NVDA and the broader AI sector this week.
Why Hold Big Tech? Why Does the “TANMAMG” Portfolio Consistently Beat the Market?
An equal-weighted portfolio of the Magnificent Seven (“TANMAMG”), rebalanced quarterly, has significantly outperformed the S&P 500 since 2015. Total return reached 3,112.63%, versus 303.16% for SPY—an excess return of 2,809.47%, while still operating at elevated valuation levels.
Year-to-date, Big Tech returns hit new highs at 25.21%, outperforming SPY’s 18.62%.
Comments