just now
Last week answered the single most important question of the quarter: can the hyperscalers back their capex commitments with financial performance? Microsoft, Amazon, and Apple all beat revenue and earnings estimates. Capex guidance was reaffirmed across the board. The underlying economics remains compelling.
This week tests the next layer of the thesis. AMD reports and tells us whether the data center buildout is translating into market share gains beyond NVIDIA. SpaceX reports as a trillion-dollar public company for the first time and either justifies or undermines the most aggressive IPO valuation of the decade. Western Digital reports into the storage demand cycle. Palantir reports with secular AI tailwinds at its back. Eli Lilly and Novo Nordisk report into the GLP-1 demand wave. Disney is reporting amid a recovery in both advertising and streaming.
The setup is structurally bullish. The bearish catalysts that defined the last several weeks (FOMC uncertainty, Iran escalation, GDP concerns) are fading. Earnings are taking over as the primary driver. And the earnings are delivering.
Last Week: The Hyperscalers Delivered
Microsoft beat on revenue and earnings, with Azure continuing to accelerate. The capex commitment was reaffirmed. The most interesting disclosure was the change in estimated useful life for data center assets to over 20 years. This is worth understanding because it has direct implications for how ROIC is calculated.
When a company extends the useful life of an asset, depreciation expense per quarter declines. Lower depreciation means higher reported net income, which mechanically improves ROIC. The bears argue that this is accounting engineering rather than genuine improvement. The reality is more nuanced.
A data center is not a single asset with a single useful life. It is a collection of components with very different lifespans. The physical structure (concrete, steel, cooling infrastructure, electrical systems) can genuinely last 25-30 years. The networking equipment might last 7-10 years. The GPUs and accelerators might last 4-6 years before they need to be replaced by the next generation. Extending the blended useful life to 20+ years is reasonable for the structural components and aggressive for the compute components.
The real answer to the depreciation debate is that the accounting treatment matters less than the economic value the assets generate. If a data center built in 2026 is generating cloud revenue in 2046, the 20-year useful life is conservative. If the compute inside it needs to be replaced every 5 years, the structural shell persists, but the contents turn over multiple times. The capex story is not about one investment paying off over 20 years. It is about a platform that generates recurring revenue while the components inside it are refreshed on shorter cycles. Both the bulls and the bears are oversimplifying.
Amazon beat revenue ($181.5B, +17% YoY) and operating income ($23.9B, 13.1% margin, a company record). AWS grew 28%, its fastest pace in 15 quarters. The $200 billion capex commitment was maintained. Free cash flow on a trailing twelve-month basis is $1.2 billion, down 95% year-over-year. As we detailed in our hyperscaler FCF analysis published yesterday, this compression is the result of choosing to reinvest virtually all operating cash flow, not of operating cash flow deteriorating. OCF grew 30% year-over-year to $148.5 billion TTM.
Apple delivered its first earnings under CEO John Ternus. Revenue and earnings beat estimates. Services revenue continued its high-margin growth trajectory. Greater China performance and iPhone ASP trends were the near-term focal points. The strategic direction under Ternus will take 18-36 months to manifest in new product cycles, but the early signals from the earnings call regarding AI integration across the product line are worth tracking.
This Week: The Next Layer
AMD (Tuesday). This is the earnings report that tests the breadth of the data center buildout. Microsoft committed $120 billion. Amazon committed $200 billion. If AMD's data center revenue shows meaningful growth and market share gains against NVIDIA, it confirms that the capex is flowing to multiple suppliers, not just one. The Sanmina partnership for rack-level integration is the differentiation story: AMD is offering a lower total cost of deployment by bundling silicon with enclosure and cooling optimization. If data center revenue accelerates, the capex-to-revenue thesis we modeled for MSFT applies across the supply chain. If it disappoints, the market will question whether the capex benefits are narrower than assumed.
SpaceX (likely Thursday). The most consequential earnings report of the week from a market sentiment perspective. SpaceX is trading near $110 after collapsing from a post-IPO high of $225. The company is now worth over a trillion dollars. It needs to demonstrate that Starlink revenue is growing fast enough and Starship economics are improving enough to justify the valuation. A miss risks a further selloff that, given the stock's index weighting and retail ownership, could create negative sentiment spillover into the broader market. A beat stabilizes the stock and removes the overhang.
Western Digital (Wednesday). Reports into the storage demand cycle at a time when data generation from AI workloads is driving enterprise HDD and flash demand simultaneously. WDC operates in both markets (HDDs and NAND flash after the SanDisk spin-in). The earnings will provide a direct read on whether cold storage demand (HDDs for data center archival) is scaling alongside the hot storage demand (flash for active inference) that SanDisk's 736% YTD gain already reflects.
Palantir (Monday). Secular tailwinds from government AI adoption, defense modernization, and enterprise AIP deployment. The company has beaten estimates for 11 consecutive quarters. Revenue growth of 85% in Q1 2026 with US commercial revenue up 133% sets a high bar. The stock's valuation (127x trailing P/E, 43x P/S) prices in continued hyper-growth. Any deceleration will be punished. But if the company delivers, it puts further distance between mission-critical AI software (Palantir, CrowdStrike, ServiceNow) and the commoditized SaaS names that are still de-rating.
Eli Lilly and Novo Nordisk. Both report into the GLP-1 demand wave. The total addressable market for weight management drugs continues to expand as clinical data supports use cases beyond obesity (cardiovascular, NASH, Type 2 diabetes). The competitive dynamics between these two companies will define pricing power and margin trajectory for the category. Supply constraints have been the primary revenue limiter. Any signal of supply normalization is a catalyst.
GDP: Cooler Than Expected, But Read the Composition
Q2 GDP came in at 1.5%, below the 2.1% consensus. The headline looks weak. The composition is more instructive.
Consumer spending grew 2% year-over-year and drove the bulk of growth. This is a modest improvement from the 1.4% consumption growth in Q1 and suggests the consumer has not fallen off a cliff, though the pace remains well below the 3%+ levels that characterized 2023-2024.
Investment growth came from industrial equipment and information processing equipment, consistent with the AI infrastructure buildout appearing in the national accounts. This is the same pattern we identified in Q1: corporate investment carrying the economy while consumption treads water.
The inventory drawdown subtracted from the headline number. This is mechanical and likely temporary. Data center construction involves large inventory buildups (servers, networking equipment, cooling systems being staged for deployment) followed by drawdowns as the equipment is installed. The ebb and flow of inventories in a buildout cycle creates quarterly noise that should not be mistaken for a demand signal.
Imports rose, driven primarily by semiconductor and telecom equipment purchases. Exports rose on fuel shipments. The import composition tells you that the AI infrastructure buildout is pulling in foreign-manufactured components. The export composition tells you that the US energy sector is benefiting from global demand, particularly from allies diversifying away from Middle Eastern supply during the Iran conflict.
We expect Q3 to be stronger as the inventory cycle reverses and the lagged effects of capex-driven equipment purchases flow through to GDP.
Iran: The Pause*
The US-Iran conflict entered a pause last week. Oil dropped meaningfully on the cessation of bombing. The structural demand picture remains bearish (China coal pivot, ASEAN demand destruction), and each successive escalation-de-escalation cycle has produced a smaller oil price response.
The pause is fragile. Neither side has demonstrated the ability to sustain diplomatic engagement for more than a few weeks. But for this week's market action, the absence of active military operations removes the geopolitical volatility premium that has weighed on risk assets since February. If the pause holds through earnings season, it clears the runway for equities to trade on fundamentals rather than headlines.
The Situational Awareness Implosion
The most discussed market event of last week was not an earnings report or an economic release. It was the implosion of Situational Awareness LP, a fund that had raised capital on the thesis that AI infrastructure spending would produce outsized returns across the value chain.
The fund's collapse is worth examining because it illustrates a failure mode that is relevant to any investor in the AI trade: thesis conviction without risk management.
The original investment thesis was not wrong in direction. AI infrastructure spending is real, accelerating, and producing measurable returns for the companies involved. The problem was concentration, leverage, and a refusal to stress-test the thesis against adverse scenarios. The fund held positions in neocloud companies with weak balance sheets, leveraged exposure to semiconductor names, and insufficient hedging against the possibility that the thesis could be right on a 3-year horizon but wrong on a 3-month horizon.
Several specific failures are worth noting for their broader lessons:
The thesis assumed that all AI infrastructure beneficiaries would appreciate uniformly. They did not. The hyperscalers, with diversified revenue bases and fortress balance sheets, weathered the Q2 volatility. The neoclouds, with undifferentiated GPU leasing models and aggressive leverage, did not. Treating a commodity GPU lessor and Microsoft as equivalent expressions of the same thesis was a fundamental analytical error.
The thesis dismissed the resilience of vertical software companies with genuine switching costs. CrowdStrike, Zscaler, Veeva, and ServiceNow have all demonstrated that mission-critical enterprise software with deep integration is not displaced by AI agents. It is enhanced by them. The blanket short thesis on software that animated part of the fund's positioning did not account for the dispersion between vulnerable and defensible software names.
The fire sale from the fund's liquidation contributed to downward pressure on the names it held, compounding losses for other investors in similar positions. Leveraged Korean retail positioning in many of the same names amplified by the effect. The lesson is not that the AI thesis is wrong. It is that expressing a correct thesis through leveraged, concentrated, unhedged positions in the weakest names in the value chain is a risk management failure, not an investment strategy.
Labor Market: Watching for the Jevons Paradox
The labor market presents a contradictory picture. Factor productivity is trending upward. Labor force participation is declining. Job openings are consolidating in the 6-8 million range. This is not a labor market that is either collapsing or booming. It is a labor market in transition.
The productivity gains from AI adoption are real but unevenly distributed. Knowledge workers in roles where AI automates a significant share of their output (content production, data analysis, basic coding, customer support) are experiencing wage compression and reduced hiring. Blue collar workers in roles that AI cannot automate (construction, electrical work, plumbing, precision machining, data center construction) are experiencing wage growth and sustained demand.
The Jevons Paradox is the dynamic to watch. In the 19th century, when steam engines became more efficient, the total consumption of coal increased rather than decreased, because efficiency made new applications economically viable that were previously too expensive. The same logic may apply to knowledge work: as AI makes cognitive tasks cheaper, the total demand for cognitive output may increase because new applications become feasible. An analyst who can produce 3x the research output with AI tools is not replaced. They are given 3x the coverage universe.
If this dynamic plays out, the labor market stabilizes without significant job destruction, and the case for rate cuts strengthens because wage-driven inflation moderates without a recession. This is not a near-term catalyst. It is a medium-term structural argument for why rates can stay stable or move lower without the economy needing to weaken first.
Yield Curve and Rates
The 30-year Treasury yield has broken past 5.29%, surpassing the prior cycle high. This is notable but potentially misleading. Very little issuance is happening at the long end of the curve right now. The bills-over-bonds strategy under Secretary Bessent continues to concentrate on new issuance at the short end, which means the 30-year yield is being set by a thin market rather than by heavy new supply.
Warsh's continued lack of explicit forward guidance on the policy rate is deliberate, not confused. He is removing the Fed's scope creep that accumulated under Powell, where the Fed felt obligated to manage expectations across the entire curve.
Warsh's approach is to set the overnight rate and let the market determine the rest. Bessent and Treasury handle the long end through issuance composition. The division of labor is intentional: monetary policy manages inflation expectations through the short rate; fiscal policy manages the term structure through issuance mix.
If rates stay where they are, the long-duration debt issued during the 2019-2021 period begins rolling off over the next 3-5 years. As it does, it gets replaced by shorter-duration bills at lower all-in cost. The weighted average cost of government debt declines gradually, reducing the interest expense burden without requiring rate cuts. This is the structural reform that Warsh and Bessent are engineering. It is slow, unglamorous, and entirely invisible to anyone watching the policy rate alone.
The Bottom Line
The bearish catalysts that dominated the market narrative for the past several weeks are fading. The FOMC delivered its decision without a hike. The Iran conflict is paused. GDP was soft but not alarming. The hyperscaler earnings were delivered.
What remains is the earnings cycle. AMD, SpaceX, Western Digital, Palantir, Eli Lilly, and Novo Nordisk all report this week. Each one tests a different branch of the current market thesis: AI infrastructure breadth (AMD), IPO valuation sustainability (SpaceX), storage demand (WDC), AI software resilience (PLTR), and healthcare innovation (LLY, NVO).
The problem children in the market are not the companies reporting earnings. They are the leveraged, concentrated, unhedged positions that treated a directionally correct thesis as a license to ignore risk management. The Situational Awareness implosion is the cautionary tale. The hyperscaler earnings are the counter-example. The difference between the two is not thesis quality. It is balance sheet quality, diversification, and the discipline to survive being right on a different timeline than expected.
We expect a constructive week. The fundamental backdrop supports it.
Appendix
US pause on strikes (July 25-27):
- Al Jazeera (July 27): https://www.aljazeera.com/news/2026/7/27/why-has-the-us-halted-its-bombing-of-iran
Ceasefire breakdown timeline and context:
