Two data prints changed the conversation this week. Micron reported Tuesday night (Sep 30) that fiscal Q4 revenue hit $54.23 billion, up 379% from a year ago, with a Q1 guide of $61.5 billion against a Street consensus of $57 billion. Then, on Friday morning (Oct 2), the Bureau of Labor Statistics told us the economy added 29,000 jobs in September, roughly a third of what economists expected, and revised the prior two months down by a combined 60,000. The bond market, which had spent the entire week trying to reprice the world toward another Fed hike, promptly took a breath. The 10-year yield, which had peaked at 5.29% Wednesday and briefly touched levels not seen since 2007, closed Friday near 5.175%. The Nasdaq finished the week in the green. Almost nothing else did.

The bond market spent the week arguing for another hike. Micron proved the AI cycle doesn't need permission. Friday's jobs print gave rates a reason to step back. The question now is whether that step back lasts.

A September to Forget, and a Week That Argues About What's Next

September closed with the S&P 500 down roughly 0.5% for the month and the Dow off more than 4%. The Nasdaq outperformed, rising 1.9% during the month, the clearest illustration of a tape that is increasingly a story about six or seven names rather than the market underneath them. The S&P 500 equal-weight index trailed its cap-weighted counterpart by 4.6 percentage points in September, according to Bespoke Investment Group. This is not healthy breadth. Only 21% of S&P 500 constituents are trading above their 50-day moving averages, and just 40% are above their 200-day.

The week ending October 3 carried that same split forward. The Dow and the S&P 500 were headed for weekly losses, while the Nasdaq was the only major average to finish in the green. The S&P 500 advanced 0.49% to 7,722.93 on Friday alone, with the Nasdaq Composite gaining 1.19% to close at 27,190.86, mostly on the jobs print and the rate relief that followed. The underlying story is a bifurcated market where the megacap AI complex is doing its own thing while the rest of corporate America sits on hold, waiting to find out whether rates are actually going higher or whether Friday morning was the turn.

Micron: The Memory Cycle Is Not a Cycle Anymore

MU has been the most important earnings report in our book this quarter, and it delivered. Fiscal Q4 revenue reached $54.2 billion, up 31% sequentially, while gross margin hit 87% and non-GAAP EPS came in at $33.42, all exceeding the high end of guidance. For the full fiscal year, Micron achieved record revenue of $133.2 billion, up 256% year-over-year, with gross margins expanding to 81.1%, a 40 percentage point improvement from fiscal 2025, and EPS increasing 811% year-over-year to $75.52.

Those numbers are striking enough. The guide is the real signal. For the fiscal first quarter, Micron projected revenue of approximately $61.5 billion and adjusted EPS of $38.15, against analyst expectations of $57 billion in revenue and $35.40 in adjusted EPS. HBM revenue is growing faster than company revenue overall, and most 2027 HBM supply is already contracted at significantly higher prices; data center SSD revenue exceeded $10 billion in fiscal Q4, more than ten times the year-earlier level. When HBM is sold out a year in advance at prices above any prior cycle peak, the conventional memory-cycle framing no longer applies. This is a contracted infrastructure business with a gross margin profile that was inconceivable three years ago.

The read for our thesis is direct. We own MU as the memory and HBM leg of the chip pillar, and this print confirms the mechanism: AI compute demand is not softening at the memory layer, it's accelerating. CEO Sanjay Mehrotra noted on the call that Micron has a "strong roadmap for future HBM products" and is working with Nvidia on the industry's first custom HBM implementation. A co-development agreement between the two dominant names in AI memory and AI compute is not a press release. It's a moat being poured in real time.

The Bond Market's Week, and Why Friday Matters

The week opened with the 10-year already at 5.24% following the September Fed hike to 3.75-4.00% and a 5-year auction that tailed badly the prior week. By Wednesday, the 10-year was pushing 5.298%, recovering from a brief pullback, with the benchmark yield trading near 2007 highs. The 30-year reached its highest level since 2002 before retreating to 5.553% Wednesday. The anxiety was real. At one point this week, traders had priced in more than an 80% chance of a quarter-point rate hike in October.

Then PCE printed. Core PCE posted a 0.2% increase in August, putting the annual core level at 3.0%, against respective forecasts of 0.3% and 3.3%. The market exhaled, briefly. October hike odds fell to around 37% after the release, with traders pushing the next expected increase to December. Then Friday delivered the body blow to the hike narrative. Nonfarm payrolls rose by just 29,000 in September, well below the 84,000 forecast, while the unemployment rate increased to 4.2%. August was revised lower to 133,000 and July switched from a gain to a loss as payrolls fell by 10,000, with the two revisions together showing 60,000 fewer jobs than previously reported.

The 10-year yield dropped 7 basis points to 5.175% on Friday, extending the retreat from this week's peak above 5.34% (intraday Thursday). That's still a positive real yield of roughly 168 basis points over CPI at 3.4%. Our financial repression thesis, which rests on the Fed being pressured to hold rates below inflation, is still in tension with the evidence. The Fed hiked on September 16, said 16 of 18 dots see another one coming, and then the labor market handed them a reason to pause. The Fed will meet again in October where it is now expected to hold rates steady. That expectation could shift again between now and October 27.

Here's the non-consensus read: a 29,000 print, in the context of a prior 12-month average near 33,000 per month, is not a labor market in crisis. The private sector added 46,000 jobs while government shed 17,000, extending a 216,000 decline over the past year. Information services lost 10,000 jobs, amid worries over the impact artificial intelligence may have on the jobs picture. Government drag and AI displacement are structural features, not recession signals. The market traded it as a hike reprieve. That's probably right for October. Whether it holds for December depends on whether the next two months look more like August (162K) or September (29K).

NVDA: Stalled on the Price, Not on the Thesis

We own NVDA as the anchor of the chip pillar, and it's been a frustrating hold from a price-action standpoint. Nvidia has risen about 21% so far this year, which looks fine in isolation but obscures the fact that the stock has gone mostly sideways even as the fundamental case keeps getting larger. The five largest hyperscalers are expected to spend nearly $800 billion on data center build-outs this year, and the guidance trajectories into 2027 are more aggressive still. Barclays analyst Tom O'Malley estimated this week that Nvidia could generate $401 billion in hyperscaler revenue in 2027.

Friday's session gave NVDA a lift. On Friday, October 2, NVDA was trending up more than 2%, closing near $234, as lower yields helped long-duration growth and AI chip names broadly caught a bid. Nvidia gained 1% Thursday on the Barclays hyperscaler revenue estimate alone. The disconnect between the capex trajectory and the share price is either an opportunity or a sign the market has correctly identified that translating $800 billion in AI spending into NVDA revenue per share is a longer path than the bulls admit. Our view is the former. CEO Jensen Huang has guided for roughly 70% revenue growth through fiscal 2028, targeting approximately $673 billion, and the mechanism, contracted supply in an HBM-constrained world, is exactly what Micron's print just confirmed.

The Hard Asset and Platform Pillars: Where the Repression Frame Sits

IBIT is our remaining hard asset position, and the week's rate volatility is a direct test of the thesis. Bitcoin as liquidity torque works cleanly in a financial repression world, where real rates stay negative and fiscal expansion is the default setting. At the end of September, the 10-year yield weekly average stood at 5.08% while inflation remained at 3.40%, meaning real yields are meaningfully positive. That's the falsifier scenario for the repression frame, and it's been in force all month. Friday's NFP-driven rate relief is a one-session reprieve, not a regime change. We hold IBIT with that tension stated plainly: until real short rates start declining durably, the hard asset case requires a patience that the tape isn't rewarding right now.

On the platform pillar, we own MSFT, AMZN, and GOOGL. The quarter-end narrative for all three is straightforward: they are the largest buyers of the AI infrastructure MU just confirmed is sold out through 2027. Their Q3 earnings, due in the second half of October, will be the next major test of whether the hyperscaler capex cycle is translating into revenue and margin expansion on the software side. The market is anxious about monetization lag. Micron's HBM sell-out actually argues the other way: if the infrastructure isn't slowing, the revenue isn't slowing. We'll know more when the hyperscalers report.

CRWD is a standing position, and one that's largely insulated from the rate debate. CrowdStrike sells because enterprises need it regardless of whether the 10-year is at 4% or 5.3%. The platform consolidation thesis doesn't depend on the repression frame. That's exactly why it's in the book.

ABNB is our deliberate non-AI diversifier. The thesis there is supply-side moat: unique host inventory that AI booking agents can't replicate. It didn't move much this week, which is the point. It won't look brilliant until the AI agent disruption narrative gets stress-tested against actual booking data.

What's on Deck for the Week of October 6

The data calendar eases up meaningfully, but the bond market doesn't take weeks off. Next week provides a breather from the recent data blitz, though several Treasury auctions could provide a sense of demand for U.S. debt with yields at multi-decade highs, and recent auction demand was far from glowing. A repeat of the September 5-year tail, which was the second-largest on record, would send the 10-year back above 5.25% and renew the October hike conversation. A clean auction would validate Friday's move and give tech room to breathe.

Fed speakers return to the circuit. Several FOMC members are scheduled to comment publicly in the week ahead, and the October 27-28 meeting is close enough that any hawkish comment will be parsed aggressively. The soft September jobs report gives the doves a clean argument, but the dot plot from September already signaled another hike, and the committee doesn't usually pivot that fast on a single soft print.

Hyperscaler earnings kick off in the second half of October. The Q3 reports from MSFT, AMZN, GOOGL, and META will be the most consequential data points of the month for the AI infrastructure thesis. The question isn't whether capex is growing (it is), it's whether it's growing fast enough to close the monetization lag the market has been worried about since the spring. Micron's print suggests the demand side of that equation is intact. The supply side (cloud revenue growth, AI product attach rates) is what Q3 earnings will answer.

Also worth watching: TSLA reports Q3 earnings October 23. Our small, thesis-oriented position in frontier autonomy and physical AI is a long-horizon hold, but the delivery numbers (already reported, came in above reduced expectations) and margin trajectory will matter for near-term sentiment.

Thesis by Pillar: Conviction and What Would Change It

AI Chips / Hardware (NVDA, AMD, MU, TSM). Conviction is high and MU's print raised it further. The HBM shortage is structural, not cyclical. 2027 supply is contracted at margins that would have looked implausible three years ago. The thesis break here is a hyperscaler capex pause, and that isn't showing up in any of the data. NVDA's guide of 70% revenue growth through fiscal 2028 and Micron's $61.5 billion Q1 guide are pointing in the same direction. We watch the hyperscaler Q3 earnings as the next confirmation or challenge.

Power and Grid (VST). The pillar is thinner since CEG's exit from our book, and higher discount rates compress the multiples on long-duration power assets twice: once on valuation and again on project financing. VST, as firm 24/7 dispatchable generation with datacenter PPA exposure, is the expression we've chosen to maintain. The rate relief from Friday matters here, but one soft jobs print doesn't change the regime. If the 10-year retraces back above 5.25%, the pressure on this pillar returns.

Platforms and AI Software (MSFT, AMZN, GOOGL). The thesis is intact: these are the capex owners on the other side of the chip trade, with cloud and AI product revenue that should eventually close the monetization gap the market is pricing. The honest bear case, especially for GOOGL, remains Search cannibalization by AI assistants. We carry that risk explicitly. The October earnings will either validate the AI attach rate story or widen the debate.

Security Software (CRWD). The platform consolidation thesis is rate-agnostic, and this week's volatility is a reminder of why we want that in the book. No change to conviction. Category spend continues regardless of where the 10-year settles.

Hard Assets (IBIT). Bitcoin as liquidity torque sits in genuine tension with the current rate regime. The 10-year weekly average near 5.08% against CPI at 3.40% means real yields are positive and meaningful. That's the condition under which the repression frame underperforms. Friday's NFP miss is a one-session improvement, not a regime shift. We hold the position as a debasement hedge and watch the rate trajectory into the October FOMC. A genuine policy pivot, not just a one-meeting pause, would materially strengthen the hard asset case.

Financials. No position. The rate factor sits in bills, which at nearly 4% against 3.4% inflation are earning a real return for the first time in years. That's the honest observation: the reserve is working. We watch Interactive Brokers and Robinhood as proxies for the brokerage cycle and consider how to re-express the rate theme when the moment calls for it.

Off-AI Diversifier (ABNB). Airbnb sold off earlier in the quarter on the thesis that AI booking agents disintermediate the platforms. Our conviction is that unique supply, the hosts and properties that aren't on every other platform, is the moat, and it doesn't evaporate because an AI agent can now write a search query. The September quarter is the next test: if nights booked and take rate hold, the fear was priced.

Frontier (TSLA). Small position, long horizon. The autonomy and physical AI thesis remains intact. October 23 earnings are the near-term marker.

This document is for informational purposes only and does not constitute investment advice, an offer to sell, or a solicitation to buy any security. Nova One Capital LLC and the Nova X Convergence Fund hold positions in securities discussed in this letter, including NVDA, MU, AMD, TSM, VST, MSFT, AMZN, GOOGL, CRWD, IBIT, ABNB, and TSLA. Past performance is not indicative of future results. This letter reflects the views of the author as of October 3, 2026, and those views may change without notice. Nothing herein should be construed as a statement of current trading activity, allocation, or position sizing.