Editor’s Note
This week, the market’s AI story became more demanding. Investors still rewarded infrastructure, compute, and operational leverage, but the easy phase of AI enthusiasm kept giving way to a harder question: where are the earnings, margins, and real customer budgets? Softer inflation gave equities breathing room, yet momentum fractures, oil risk, and earnings quality concerns made the rally feel more selective than euphoric.
AI Infrastructure Still Sets the Market’s Pulse
AI remained the center of gravity across the week, but the market increasingly separated infrastructure winners from product noise. Meta Platforms started the period at $585 and surged 14%, with earnings still 19 days away, as investors appeared to look beyond Muse Spark 1.1 and a new image generation model toward a larger monetization thesis. The key question is whether Meta’s AI infrastructure can become a strategic asset rather than a pure expense, especially if excess compute supports partnerships with companies like Anthropic or Claude in exchange for model access or tokens instead of paying “hundreds of millions of dollars, if not a billion dollars a month” externally.
That thesis gained force as Meta’s broader buildout came into focus: a Meta-made AI chip expected to enter production in September, Broadcom’s stronger-back-half signal, work on a one gigawatt data center in Canada, and a Richland Parish, Louisiana expansion to 5GW of compute capacity. Disclosed investment there rose to over $50B from an original $10B plan, while Bloomberg reported total site cost could exceed $250B including chips.
Nvidia also remained a bellwether. The stock bounced nearly 8%, from $195 to $210, after holding its 200-day moving average, while trading around 22 times forward earnings, described as its cheapest valuation since before the AI boom. A potential reopening of China demand added another layer, with China possibly allowing major tech firms to buy H200 GPUs, worth $6 billion to $8 billion.
The infrastructure boom was visible elsewhere. TSMC reported June revenue of about $13.8B, up 6.2% MoM and 67.9% YoY, with Q2 revenue of about $39.6B, up 36% YoY and above the roughly $39.4B estimate. ASML reported Q2 net sales of €9.33B vs €8.85B expected, EPS of €7.59 vs €6.90, and gross margin of 54.0% vs 52% expected, while raising FY26 guidance. Nebius agreed to sell $1B+ of AI compute to Reflection AI through 2029, including access to Nvidia GB300 chips.
But investors also started examining the terms of the buildout. CoreWeave is reportedly exploring hedges against future memory and storage price declines because long-term supply deals with price floors could leave AI cloud buyers paying above-market rates. New York is reportedly set to enact the first statewide U.S. data center moratorium, pausing approvals for new hyperscale data centers using 50MW+ of power for one year. AI demand is real, but power, pricing risk, and permitting are becoming part of the thesis.
The Next AI Winners May Be Less Obvious
One of the week’s most important ideas was that AI’s biggest profit impact may not be in high-margin software, but in low-margin industrial businesses. Chamath Palihapitiya said his CTO told him token costs were “doubling every 45 days,” while productivity gains were “maybe 5% max.” As Chamath summarized: “My costs are doubling every 45 days. My upside is essentially flat.”
That same 5% max productivity improvement looks very different in a business with 3% margins. A company keeping $3 of profit on every $100 of sales can see profit rise from about $3 to nearly $4 if costs fall just 1%, a roughly 30% jump in profit.
GE Aerospace became the cleanest case study. Since 2024, GE has worked with Palantir to improve supplier efficiency and strengthen its supply chain. The effort reportedly helped GE pull more than 40% more parts from its most critical suppliers in a single year, while operating profits increased 25%. The broader lesson is that AI may be most financially powerful where operations are messy, supply chains are constrained, and small throughput gains unlock high-value aftermarket revenue.
This theme also showed up in Nvidia’s expanding Toyota partnership, which now stretches beyond autonomous driving into smart cities, traffic systems, factories, Woven City, Omniverse digital twins, Isaac robotics, and Nemotron LLMs. The next phase of AI investing may be less about chasing model launches and more about finding where AI improves real-world bottlenecks.
Inflation Helped, but Momentum Cracked
Macro data gave the market a reason to breathe. June headline CPI was 3.5% YoY versus 3.8% expected, down from 4.2% in May, while core inflation eased from 2.9% to 2.6%. Another recap framed the report as prices falling 0.4% MoM versus expectations for 0.0%, with core CPI flat MoM versus 0.2% expected. Tom Lee highlighted June core CPI at -0.02% versus consensus of 0.26%, with 68% of core CPI items deflating and shelter at 3.28% year over year, slightly below the 40-year average.
PPI reinforced the cooling trend. Headline PPI was 5.5% YoY vs 6.2% expected and fell 0.3% MoM vs expectations for 0.0%. Core PPI was 4.7% YoY vs 5.1% expected and rose 0.2% MoM vs 0.3% expected. Lee called the rate-of-change shift “a dovish development.”
Yet internals deteriorated. Goldman Sachs’ High-Beta Momentum Index fell 24% month-to-date through the first half of July, its worst stretch since April 2009. Morgan Stanley’s Tech Momentum Index posted a 35% decline in its 17-day rate of change, the worst move in its 27-year history. U.S.-listed leveraged ETFs reached a record 700, more than double the count at the end of 2024, with 400+ tied to individual stocks.
Oil and geopolitics added another complication. Crude moved from below $70 to almost $78 per barrel, with risk of a move above $90 if escalation continues. Later in the week, WTI was back above $80 and Brent crude above $85. United Airlines showed why that matters, expecting nearly $6 billion in added fuel expense for full-year 2026, with Q2 fuel expense up $2.3 billion, or 84% year over year.
Earnings Are Becoming a Proof Test
Earnings season started drawing a sharper line between durable execution and vulnerable narratives. Banks helped stabilize the tape: Goldman Sachs, Bank of America, JPMorgan, and Wells Fargo all beat Q2 expectations, led by strong trading and resilient credit. J.B. Hunt also beat, with EPS of $1.91 versus $1.74 expected and revenue of $3.5 billion versus $3.25 billion expected.
Software looked more fragile. IBM fell 24% after preliminary Q2 results missed expectations, with revenue of $17.2B versus $17.86B expected, up just 1% YoY, consulting flat, and infrastructure down 7% YoY. CEO Arvind Krishna cited late-June customer capex shifts toward servers, storage, and memory. IBM lost roughly $65B in market cap. Palantir was the counterpoint, rising from $122 premarket to $135 at the open and closing up 3%, as investors treated it more like an AI software winner than a budget casualty.
Netflix showed the cost of lower visibility. Revenue was $12.56 billion, just below $12.58 billion expected. EPS was $0.80, beating by one penny. Free cash flow was $1.53 billion, below the $2.72 billion estimate and down 33% year over year. Q3 guidance called for $12.86 billion in revenue versus $13 billion expected and EPS of $0.82 versus $0.84 expected. Management still expects about $3 billion in ads revenue in 2026, and audiences watched more than 97,000,000,000 hours in the first half of 2026, but reduced disclosure means revenue, profit, and free cash flow now have to carry more weight.
Looking Ahead
Next week’s market needs confirmation. Watch whether Meta can justify its AI infrastructure re-rating, whether Nvidia and the semiconductor complex stabilize after the momentum break, and whether enterprise software weakness is limited to IBM or signals broader budget crowding-out. Inflation is helping, but oil, leverage, and earnings quality remain the swing factors. The market still wants AI exposure, but this week made one thing clear: proof now matters more than promises.