In this week’s newsletter, we will cover the following highlights:

While markets have been absorbed by the semiconductor selloff and IBM's earnings warning, a sector with no relation to AI has been building momentum without much notice. Agriculture and fertilizer stocks are showing some of the strongest momentum score improvements in our model over the past three weeks, and the drivers behind the move are straightforward.

The energy connection is the first one to understand. Nitrogen fertilizers — the most widely used category — are produced through the Haber-Bosch process, which is heavily dependent on natural gas. When energy prices rise, nitrogen fertilizer production costs rise with them, and those costs pass through to buyers. The sustained elevation of crude oil and natural gas prices following the Iran ceasefire collapse in early July has re-established a cost floor for nitrogen fertilizer producers that was briefly eroded during the six-week ceasefire period. CF Industries, the largest nitrogen fertilizer producer in North America, is one of the clearest beneficiaries: higher natural gas prices hurt its input costs modestly but tend to be more than offset by the pricing power it gains over buyers who face the same energy-linked cost environment.

The supply chain disruption earlier in the year amplified this. When the Strait of Hormuz was effectively closed in the spring, fertilizer logistics — particularly for producers relying on maritime trade routes — were thrown into disarray. Prices spiked sharply in March and April, and while they have partially retraced since the ceasefire was first signed in June, the collapse of that ceasefire on July 8 has re-introduced the supply risk premium. The United Nations Food and Agriculture Organization reported that global food prices rose for the first time in five months as recently as February, with vegetable oils leading the move — up 3.3% to their highest level since mid-2022. That trend has not reversed.

The names drawing the most consistent volume in the screener are CF Industries (CF), Mosaic (MOS), and Nutrien (NTR), with smaller names like Intrepid Potash (IPI) and ICL Group also appearing regularly in high-volume lists. The VanEck Agribusiness ETF (MOO), which holds exposure across the full agricultural supply chain including fertilizer producers, equipment manufacturers, and food processors, is trading around $80 and is up approximately 16% year-to-date. The Invesco DB Agriculture Fund (DBA), which tracks agricultural commodity futures, has recently consolidated near all-time highs and carries a 3.25% yield.

The risk in this trade is also clear: a durable Iran peace agreement, if one materializes, would likely take energy prices and fertilizer costs lower, and the pricing power currently flowing to producers would compress. For now that scenario remains theoretical; the July 8 collapse of the ceasefire makes a fresh agreement in the near term less likely, not more. A secondary risk is that higher fertilizer prices eventually become a headwind for food producers and agricultural commodity prices if farmers reduce planting due to input cost pressures — a dynamic that takes multiple crop seasons to fully develop, but is worth monitoring.

Alphabet reported Q2 2026 earnings on July 22. The headline numbers were strong: revenue of $119.8 billion against a consensus estimate of $116.9 billion, and adjusted EPS of $9.11 against an expectation closer to $2.90. Google Cloud revenue grew 82% year-over-year to $24.8 billion, ahead of the $24.6 billion expected. Remaining performance obligations — the contractual backlog — reached $514 billion, well above the $488 billion Wall Street had projected.

The stock still fell more than 6% the following day. The reason was the capital expenditure disclosure.

Q2 2026 CAPEX came in at $44.9 billion, representing 101% growth year-over-year. That is an extraordinary absolute number. But the user's point is the relevant one: in Q1 2026, Alphabet's CAPEX grew 130% year-over-year. The sequential deceleration in the growth rate — from 130% to 101% — is the first such deceleration since the AI infrastructure build-out began accelerating in earnest. Alphabet simultaneously raised its full-year 2026 CAPEX guidance to $195–205 billion from the prior range of $180–190 billion, and its CFO stated that 2027 CAPEX would "significantly increase." Approximately 60% of the Q2 spend went to servers; 40% to data centers and networking.

The free cash flow consequence is visible in the right panel of the chart. Q2 2026 was the first quarter in recent memory in which Alphabet generated negative free cash flow, coming in at minus $5.9 billion. This is directly attributable to the pace of capital spending. The company ended the quarter with $242.5 billion in cash and marketable securities, so the negative FCF is not a liquidity concern — but it is a signal that the current rate of investment is not self-funding, and that the company has had to lean on its cash reserves and a planned $80 billion equity raise to sustain the build-out.

The market's reaction, and the reason this matters for semiconductors, is that Alphabet is the first major hyperscaler to report Q2 results. Its CAPEX deceleration — modest as it may be in absolute terms — gives investors something to work with when asking whether the $750 billion in collective hyperscaler AI infrastructure spending projected for 2026 will hold, accelerate, or begin to moderate. The remaining hyperscalers — Microsoft, Meta, Amazon, and Apple — report in the days ahead. If any of them shows a similar pattern, the "demand is infinite" narrative that has supported semiconductor valuations for the past eighteen months comes under genuine pressure.

The question is not whether Google is still spending at a staggering rate. It is. The question is whether the second derivative — the rate of change in the growth rate — has peaked. One quarter is not a trend, but it is the first data point pointing in that direction, and in a market that has been running on the assumption that AI capex can only go up, the first exception gets priced quickly.

The past two weeks have produced enough ammunition for both sides of the chip debate that the honest answer is: neither side has clearly won.

The bearish case got stronger with the release of Kimi K3 — Moonshot AI's 2.8 trillion parameter open-weight model, which benchmarked near the top of the global leaderboard and is scheduled for full public weight release on July 27. Combined with DeepSeek's continued work and the general proliferation of capable open-source models, the argument is that you no longer need to spend at the scale of a major U.S. hyperscaler to access frontier-level AI capability. If that is true, the $750 billion in annual AI infrastructure capex starts to look like it is being driven partly by competitive fear rather than pure demand — a more fragile foundation.

Then Liang Wenfeng's investor meeting notes were leaked.

DeepSeek's founder, at a recent investor meeting, said plainly that the biggest gap between his company and U.S. AI leaders lies not in talent or model design but in computing resources — and that all the differences visible in model capabilities and applications trace back to that single disparity. This from the man who runs the lab that supposedly proved frontier AI does not require massive compute.

The specifics are more revealing than the headline. Liang said he needed 200,000 Huawei 950 chips to train a frontier model, but Huawei could only supply 16,000. Huawei's total expected chip capacity for 2026 is approximately 750,000 units — split across all Chinese AI companies — and Liang said he expected the constraint to persist for at least three years. DeepSeek has since released its V4 model adapted for Huawei's Ascend chips, and Huawei confirmed its processors handled part of the training workload, a direct consequence of being locked out of Nvidia's most advanced hardware.

DeepSeek estimates it remains 12 to 18 months behind leading U.S. AI companies and attributes that gap to computing resources rather than algorithmic capability. Liang has noted separately that U.S. export restrictions force Chinese companies to use two to four times the computing power to achieve equivalent results — the penalty of running on constrained hardware rather than Nvidia's latest generation.

The picture that emerges is not one where open-source AI has made compute irrelevant. It is one where the most capable Chinese lab — the one whose efficiency advances genuinely rattled Silicon Valley in January 2025 — is telling its investors that chips remain its binding constraint, and that U.S. export controls are working in exactly the way they were intended to work.

That does not mean the chip trade is unchanged. The market is correct to ask whether hyperscaler CAPEX will grow at the same rate indefinitely, particularly as Google's results show FCF turning negative and the growth rate beginning to moderate. The specific question of how many Nvidia GPUs the hyperscalers need to buy next year is genuinely uncertain. But the argument that open-source AI has made chip investment unnecessary — the argument that pushed the SOX into bear market territory last week — runs directly into the statement of DeepSeek's own founder, who says he cannot buy enough chips and expects that situation to last for years.

The balance point is somewhere between those two realities. The chip trade is being repriced, not ended. The valuation correction in semiconductor stocks reflects a reasonable recalibration of the rate of CAPEX growth, not a collapse in the underlying demand thesis. What the market is currently working out is how much of the demand story was structural and how much was FOMO — and Google's Q2 report, for all its strength, gave the market its first concrete reason to ask that question.

Microsoft, Meta, and Amazon all report this week. Their CAPEX guidance will either confirm or challenge the pattern suggested by Alphabet's Q2. DeepSeek's Kimi K3 full model weights release on July 27 will allow independent developers to evaluate the model directly — expect a fresh round of benchmark comparisons and media coverage. In agriculture, the next catalyst is the USDA's monthly World Agricultural Supply and Demand Estimates (WASDE) report, due August 12, which will give the first real read on how elevated input costs are affecting planted acreage. Consumer Defensive and cybersecurity continue to be our model's top momentum scores. The SOX is attempting a technical stabilization after last week's 12.5% drawdown; whether it holds depends on whether the remaining hyperscaler CAPEX numbers provide a floor or a further leg down.