The bond market spent three days trying to tell equities something, and equities mostly ignored it. By Friday's close, the S&P 500 finished the week up 0.6%, the Nasdaq 100 jumped 2.1%, and the 10-year Treasury yield closed at 5.17%, its highest sustained level since June 2007. Both of those things are true. Only one of them is a problem.

The market isn't ignoring the bond rout. It's betting that AI earnings power is large enough to absorb it. Micron's report Tuesday night is the next real test of that bet.

A Week That Argued With Itself

Monday set up beautifully. Stocks rallied sharply on hopes of resumed U.S.-Iran negotiations, with oil futures dropping about $4.52 per barrel, roughly 4.5%, to $95.78 as President Trump signaled openness to meeting Iranian officials at the UN General Assembly. The Nasdaq surged 2.26% to its first record close since June 2, led by semis and AI names across the board. It looked, briefly, like the tape had simply been on sale.

Then the bond market remembered it had a thesis to test. Thursday's 7-year Treasury auction wrapped up the week's supply slate with a thud, as demand looked soft with particularly disappointing foreign participation. That followed soft demand for Tuesday's 5-year note auction, which also pushed yields higher. The 10-year benchmark reached its highest rate since June 2007, touching 5.163%, while the 30-year surged to levels last seen in 2004, hitting 5.49%. Two consecutive auction failures in the same week, back to back, is the market pricing in a structural oversupply of duration at a moment when the Fed has said explicitly it is not finished.

Markets currently expect the Fed to raise the federal funds rate another 25 basis points at its next meeting, with the probability of that move standing around 66%. Swaps are pricing in three additional quarter-point hikes over the next year. The September 16 hike to 3.75-4.00% was, in retrospect, an opening statement, not a hawkish one-and-done.

Friday offered some relief. Oil prices fell after news that the U.S. and Iran moved closer to an agreement to lift naval blockades on tankers in the Persian Gulf and ease economic measures against Tehran. WTI crude dropped 2.33% to settle at $92.41, while Brent declined 2.14% to $104.32. The 10-year pulled back modestly to 5.17% following the three-session rout, with the decline in oil providing some relief to the bond market. The week ended green. The anxiety did not.

AI Chips + Hardware: The Memory Leg Steps Up

Semis held their ground through a week that punished anything rate-sensitive. Applied Materials rose 2.3% and Lam Research gained 2.6% on Friday alone. The equipment names are telling you something: the buildout isn't pausing because the 10-year crossed 5%. The capex commitments are contracted. The supply constraints are real.

The week's real AI chip story is what happens Tuesday evening. Micron Technology reports its fiscal Q4 2026 earnings confirmed for September 30, after the close. We own MU as the memory and HBM leg of our chip pillar, and the setup coming in is genuinely unusual. HBM supply for 2026 is fully booked, HBM4 is already shipping into NVIDIA's Vera Rubin platform, and DRAM spot prices have risen 52% since January. Record free cash flow was achieved last quarter, HBM3E and HBM4 products are fully booked through 2027 with demand extending into 2028, and the company secured $22 billion in strategic customer agreements including $18 billion in cash deposits.

The street's consensus revenue figure for the print sits near the September 30 report, with revenue forecast around $50.45 billion. But the number that matters more is the forward guide. With MU's stock up roughly 266% year-to-date, the setup is compelling but perfection is partially priced in. The most consequential question is whether Q1 FY27 guidance exceeds $55 billion in revenue, which would validate the structural-earnings-power thesis over the cyclical-peak narrative. That's the print that either confirms the HBM supercycle or gives the bears their first real piece of evidence.

Platforms + AI Software: Microsoft Plays Defense, or Maybe Offense

The week's most underappreciated move was Microsoft, which finished Friday up 3.7% after announcing plans to consolidate its consumer and workplace Copilot products into a unified offering for corporate customers. The company is offering discounts of 30% to 50% on Copilot subscriptions for large corporate customers, with discounts beginning as soon as October targeting customers committing to large seat purchases and additional usage-based payments. Microsoft is also preparing to launch a revamped Copilot "super app" combining AI coding tools and agent features, with some capabilities carrying extra usage-based fees.

Here's the non-consensus read on those discounts: the market initially heard "MSFT is cutting prices" and registered it as margin compression. The better interpretation is that MSFT is doing what every serious enterprise software company does when it has a product it believes in, pulling the adoption curve forward aggressively now to lock in seat counts before competitors consolidate the corporate desktop. At $30 per user per month at list, Microsoft is offering roughly 30% off for customers with more than 1,000 seats and up to 50% for those buying more than 10,000 seats. If 10,000-seat commitments become the standard enterprise deal size, the volume math eclipses the discount math. We own MSFT and the Copilot monetization story is the thesis. A price cut to accelerate scale isn't a thesis break.

We also own GOOGL. The company's cleared antitrust overhang on the ad-tech remedy (no forced AdX sale) was supposed to be a catalyst, and the market shrugged it off. That's fine. The honest GOOGL bear case remains that Search is the profit engine and AI chat is its most credible disruptor. The bull case, which we think is the more complete frame, is that GOOGL owns the silicon layer in the TPU, owns a frontier model in Gemini, owns cloud distribution, and can afford to be disrupted by itself. That thesis didn't change this week.

The Summit: Symbolism, a Two-Month Extension, and Jensen Huang at the State Dinner

Thursday's meeting was Xi and Trump's second summit this year, and the Chinese leader's first visit to Washington in more than a decade. The pageantry was elaborate, the outcome was thin. Treasury Secretary Scott Bessent said the two sides extended the bilateral trade truce, which was due to expire in November, by only two months. Rather than locking in a durable agreement, the extension pushes the central questions, including tariffs, Chinese purchases, rare-earth supplies, and technology restrictions, into early January.

The detail that matters most for this portfolio: top U.S. technology executives, including Nvidia's Jensen Huang and Tesla's Elon Musk, were present at the state dinner. The two leaders underlined the importance of discussions on AI. There were no announced changes to chip export rules. That's actually the constructive read. NVDA's forward guidance has already stripped out China datacenter revenue entirely. Any softening of export controls from here is upside the model doesn't need. Any tightening would have been telegraphed by the summit's language, and the summit produced no such signal. January is when this becomes urgent again.

The meetings between U.S. and Chinese leaders failed to forge any trade breakthroughs. The two-month extension is a deadline, not a resolution. Expectations were low partly because both nations are vying for the lead in what amounts to an AI arms race. January's expiration will arrive with either a real framework or another extension, and the market will need to price which is more likely well before then.

Hard Assets: The Repression Thesis Under Its Hardest Test

Gold closed the week near $4,311 per ounce and bitcoin near $83,958. Both are functioning as the fund's stated insurance against financial repression, and both are sitting uncomfortably against a 10-year yield that just touched its highest level since 2007. That tension is the honest story of this sleeve right now.

The repression thesis holds that the U.S. debt stock is too large to service at real rates, so the eventual policy response is curve management and engineered inflation tolerance, with hard assets as the beneficiary. The University of Michigan's consumer sentiment survey confirmed a sharp rise in inflation expectations in September. Strong U.S. economic data, worsening fiscal conditions, and rising government debt have also weighed on the Treasury market. Those are the conditions the repression thesis was built for. The problem is that the Fed, so far, is tightening into them rather than accommodating them.

What's interesting is the correlation structure. The trading session on September 18 placed IBIT ahead of GLD in both percentage price movement and trading value. Bitcoin is trading with gold, not against it. When two instruments with different volatility profiles, different liquidity bases, and different investor bases start moving together, it usually means the macro signal is the dominant driver, not the idiosyncratic one. Both are responding to the same question: will real rates stay positive long enough to break the debasement narrative? That question doesn't resolve this week.

Power + Security: The Discount Rate Doubles the Pain

Higher long rates hit the power pillar twice: once on the multiple expansion that powered the trade, and again on the financing costs for new generation and transmission capacity. With the 10-year now at 5.17%, that double-tap is real. VST is the position we hold in this pillar, and the firm-power, PPA-anchored thesis around datacenter demand doesn't change with the rate environment, but the valuation math gets more demanding.

CRWD is a different kind of rate story, which is part of why it's in the book as a standing position. Platform security consolidation doesn't depend on the yield curve. It sells whether AI ROI shows up or not, because the threat surface expands regardless of macro conditions. That's the function it plays in the portfolio construction.

What's on Deck: The Week That Actually Matters

Next week is one of the most data-dense weeks of the quarter. The calendar arrives at a moment when the market is genuinely uncertain whether the bond rout is (a) inflation, (b) growth, (c) fiscal deterioration, or (d) all three. The data will help sort that.

The single most consequential item on that calendar for our book is Wednesday evening. September 30 brings ADP employment change, August PCE prices, the second estimate of Q2 GDP, and Micron's earnings after the close. That is a remarkable amount of macro and earnings signal compressed into one day. If PCE is hot and Micron guides above $55 billion for Q1 FY27, the two signals will point in opposite directions and the tape will have to decide which one matters more. That's the kind of session where positioning clarity matters more than being right on any individual print.

Watch the ISM Manufacturing number Thursday. The 10-year hovered near 5.2% this week as traders refocused on hawkish comments from Fed officials, while a lack of concrete progress in U.S.-Iran negotiations continued to fuel inflation concerns. A second consecutive strong ISM above 57 would put a November hike firmly on the table, not just in the probabilities but in the conversation. A global bond selloff this week extended beyond the U.S., with Japanese government bonds, U.K. gilts, and German bunds all hitting fresh highs. This isn't a U.S. fiscal story alone.

Forward Thesis by Pillar

AI Chips / Hardware (NVDA, MU, AMD, TSM). The compute pillar is contracted, supply-constrained, and getting paid regardless of the macro backdrop. NVDA's guidance already excludes China datacenter revenue, so Thursday's summit producing no chip policy movement is neutral to constructive. MU is the near-term catalyst. We need the Q1 FY27 guide, not just a Q4 beat. What would change conviction: guidance below $50 billion, or a signal that HBM booking deferrals are starting. Neither is in the data right now.

Platforms + AI Software (MSFT, AMZN, GOOGL, NOW). MSFT's Copilot pricing move is a land-grab timed to the enterprise budget cycle. We read it as evidence of confidence in the product, not desperation on price. GOOGL's thesis is intact after the antitrust resolution. The capex anxiety around hyperscalers remains the honest bear case for this pillar broadly, but the concern is timing, not permanence. What would change conviction: a meaningful deceleration in cloud revenue growth from either MSFT Azure or AMZN AWS in their October earnings.

Security (CRWD). The platform consolidation thesis doesn't need AI ROI to show up. It needs enterprise security budgets to keep growing, and the threat surface keeps expanding whether or not AI productivity materializes. Rate-independent is a feature right now.

Power (VST). The higher-rate environment compresses the multiple and complicates the financing on new capacity. The fundamental demand story, firm power for datacenter PPAs, is unchanged. The thesis is intact at a lower valuation, which is a different problem than the thesis breaking.

Hard Assets (GLD, IBIT). The repression frame is being tested. Real short rates are positive. The Fed is tightening. Both of those facts are unfavorable for the debasement trade in the short run. The structural argument, that the fiscal trajectory eventually forces a choice between real rates and debt sustainability, remains intact. The honest answer is that this sleeve requires patience and a longer time horizon than any single quarter's rate moves. The January trade deadline and the post-NFP Fed posture will both matter for where this goes next.

Off-AI Diversifier (VRTX). Cystic fibrosis cash flows don't care about the 10-year. That's the point. No new catalyst this week; the pipeline optionality is unchanged.

Frontier (TSLA). Long-horizon autonomy optionality. Nothing this week changes the thesis. The time horizon on this name is measured in years, not weeks.

The macro frame, stated plainly. Wall Street can't agree on what's driving the surge in bond yields, whether it's stubborn inflation, rapid growth, or an exploding deficit, but the likelihood that higher interest rates are here to stay is unnerving investors who fear that eventually something has to break. That uncertainty is the honest state of the market. Our book is built for a world where repression eventually wins. The bond market is currently arguing it doesn't. Both scenarios are open. Efforts by Treasury Secretary Bessent to cap long-dated yields through increased Treasury buybacks are widely seen as having had limited impact. The repression frame needs that to change. September 30's PCE and October 3's payrolls are the next real data points.

Nova One Capital and the Nova X Convergence Fund hold positions in securities discussed in this letter, including NVDA, MU, AMD, TSM, MSFT, AMZN, GOOGL, NOW, CRWD, VST, GLD, IBIT, VRTX, and TSLA. This letter reflects the views and analysis of Nova One Capital as of September 26, 2026, and is intended for informational purposes only. It does not constitute investment advice, a solicitation, or an offer to buy or sell any security. Past performance is not indicative of future results. All investments involve risk, including the possible loss of principal. The information contained herein is believed to be from reliable sources but is not guaranteed as to accuracy or completeness. Nova One Capital and its affiliates may hold positions in any securities mentioned and may trade in a manner inconsistent with the views expressed. Recipients of this letter should conduct their own due diligence and consult a qualified financial advisor before making any investment decisions. This document is for informational purposes only and does not constitute investment advice. September 26, 2026.