NVDA’s Financing Loop Raises Questions; Berkshire’s Buybacks Send a Clearer Signal
US Stocks · Options · News · Views
Yields rise, and tech feels it first
The new week started on a soft note. Major indexes finished slightly lower on light volume. Crude oil jumped more than 6%, while energy led the S&P sectors with a gain of roughly 4.6%. Real estate, utilities, and technology lagged. Money rotated toward inflation-sensitive and real-asset exposure, while long-duration growth moved to the back of the line.
The bigger pressure point was the bond market. Treasury yields rose across the curve, and US10Y retraced its decline since last Tuesday. Small- and mid-cap growth stocks, along with tech, took the brunt of the move. Fundamentals did not suddenly break; discount rates simply moved against them first. Wednesday’s CPI is the next major pricing catalyst. Markets are roughly split on a September hike, while FOMC voter Harker struck a hawkish tone on full employment and the possibility of multiple hikes. That setup does not favour blindly buying every dip.
NVDA: a $500B financing loop cuts both ways
Reports said a group of institutions would work with NVDA to provide roughly $500 billion in financing for AI infrastructure. The company later confirmed arrangements with six major Wall Street firms. The lenders will offer preferential-rate financing to NVDA customers. Jensen Huang described the strategy as going beyond chip sales by helping customers build “AI factories.”
The mechanics are simple. Customers get access to credit, then use that capital to buy GPUs, boosting near-term orders and revenue. In the short run, it acts as a demand accelerator. Over a longer horizon, however, it creates a financing loop: some demand is being pulled forward with credit instead of being funded by proven AI cash flows. Cisco used a similar playbook before 2000. When the cycle turns, that kind of demand elasticity can hit valuations hard.
That makes the market reaction more nuanced than a simple bullish or bearish headline. Financing can support shipments today. But if AI monetization falls short and borrowers run into trouble, guarantee risk and reputational risk can come back to the chip supplier. Semiconductors and NVDA sold off as investors questioned how much of the demand is truly organic.
Technically, NVDA remains inside its rising channel after breaking above 216. I would keep holding rather than overreact to a one-day narrative shift. The exit signal is a clear break in trend. Large pools of capital are still committing real money to AI infrastructure. Until the trend actually breaks, I would rather follow the capital than try to call the top.
MSFT: custom silicon strengthens the moat
The Information reported that MSFT plans to ramp production of its next-generation custom AI chips. After Maia and its follow-ons, attention is turning to a “300-series” design. Microsoft is reportedly in talks with TSM to secure delivery of more than 300,000 custom chips in 2027. Google, Amazon, and Microsoft are all developing their own silicon to reduce dependence on external GPUs and optimize workloads for their cloud platforms. For MSFT, this is a multi-year moat and efficiency story, not a day-trading catalyst.
It is not automatically bad news for the NVDA ecosystem either. Hyperscaler chips can take on specialized and inference workloads, while the most advanced training GPUs remain in tight supply. For Microsoft, custom silicon reduces supply-bottleneck risk and makes the long-term fundamental story cleaner.
The stock is still struggling below 516, with the daily chart overbought and showing bearish divergence. There are two clean ways to reset the setup: a high-volume breakout above 516 toward roughly 534 followed by a successful retest, or a direct pullback that works off both conditions. On a pullback, the main area I am watching remains 469–487, roughly 5% lower. This is still a core holding, so I would not flip in and out while the fundamental thesis remains intact.
BRK.B: accelerating buybacks are the clearest signal
BRK.B reported roughly $12 billion in Q2 after-tax operating earnings excluding FX, up about 6% year over year. The earnings growth itself is not remarkable. The more important signal is the pace of repurchases: about $235 million in Q1, roughly $4.5 billion in Q2, and another $3.4 billion in July. Berkshire’s rule has not changed. Management buys back stock when it trades below intrinsic value and the company still has ample cash, comfortably above its $30 billion floor.
Berkshire paused buybacks in May 2024 when valuation looked stretched. The sharp acceleration in Q2 and July now suggests that Buffett’s team sees the 470–500 area as meaningfully undervalued. Actual cash going into buybacks is a stronger signal than management simply calling the stock cheap.
My forward valuation ranges are roughly 406–674 for 2026, with a midpoint near 540, and 321–698 for 2027, with a midpoint near 560. At around 529, the stock does not look expensive relative to the 2027 midpoint. Earnings growth is slow, though: roughly 3% in each of the next two years and about a 7% five-year CAGR. I would not expect fireworks. Technically, the break above 510 improved the setup, while 529–538 is the next resistance zone before new highs. I see Berkshire as low-beta portfolio ballast, not a momentum trade.
TSM: already a large position, so adds stay disciplined
TSM sits alongside GOOG and TSLA in the third portfolio sleeve at roughly 6% each. My 2026 valuation range is about 253–467, with a midpoint near 360. For 2027, it is roughly 321–593, with a midpoint near 457. At 410–420, the stock still offers long-term upside versus the 2027 midpoint, but the near-term multiple is rich and the portfolio already has heavy semiconductor exposure.
Any additions stay on a preset ladder with a clear cancellation level. I would make a small add at 358–375, then add more aggressively on a panic move into 320–350, near the 200-day moving average. If the stock reaches 429 first, the add plan is cancelled. I am not chasing it higher. TSM still has pricing power, but crowded trades rarely reward late entries.
What I’m watching into Wednesday’s CPI
Higher yields have already hit tech once. The NVDA story has shifted from “pure demand” to “demand plus financing,” which adds volatility, but an intact rising channel is not a reason to panic-sell. For MSFT, the focus remains the long-term moat and the conditions around a 516 breakout. BRK.B is sending a tangible valuation signal through buybacks. TSM stays on the written buy ladder, and a move to 429 cancels the add plan.
If CPI pushes real yields higher again, growth stocks could take another hit. If inflation comes in cooler, tech has room to rebound. This week is less about guessing the next headline and more about sticking to structure: respect the trend on offense, use valuation as the defensive anchor, and let price come to the levels already written down.