Editor’s Note
This week, the AI trade became less abstract and more financial. The market is no longer just debating whether AI is transformative; it is asking who earns durable free cash flow, who is stuck funding the buildout, and what happens if expectations outrun returns. At the same time, a softer labor market lowered Fed hike odds and gave equities a more supportive macro setup, even as valuation and concentration risks stayed front and center.
AI Infrastructure Becomes the Market’s Center of Gravity
The clearest expression of the AI infrastructure boom is South Korea. The KOSPI index is up by more than 100% so far this year and has climbed 200% over the last 12 months, meaning it has more than tripled. That is not a normal broad-market move; it is a national equity market turning into a concentrated proxy for global AI capital spending.
The reason is straightforward. In 2025, Amazon, Google, Meta, and Microsoft spent $376 billion on capital expenditures, and the industry is on track to spend $725 billion in 2026. That money is pouring into data centers, chips, networking, power, cooling, and memory. South Korea sits directly in that flow because Samsung Electronics and SK Hynix dominate key parts of the memory supply chain. At points this year, Samsung stock has risen 500%, while SK Hynix has surged more than 1,000%.
But this is also where the risk begins. Samsung and SK Hynix make up more than 50% of the KOSPI, turning the broader market into a concentrated bet on two AI-linked companies. One warning captured the setup bluntly:
“The Korean stock market trades like a penny stock or trades like an individual volatile stock. Because in many ways it is just two stocks.”
That concentration is amplified by retail leverage. South Korea has more than 14 million retail investors in a country of 51 million, and some are using debt and leveraged ETFs designed to deliver two times or three times daily returns. One investor’s portfolio has risen to around $655,000, with a large portion in leveraged ETFs. Another, who had worked in AI since 2017, invested five years ago because she believed in the technology’s future; her position is now up 1,300%.
The bullish case is real: AI demand is driving earnings. The bearish case is also real: if hyperscaler spending slows, Korea may be one of the first places where the pressure shows up.
The Memory Shortage Turns AI Into an Inflation Story
AI is often described as deflationary over the long run, but this week’s data made clear that the buildout is inflationary right now. Apple’s Tim Cook called the memory shortage a “100-year flood,” saying he had never seen anything like it in more than 40 years.
The pricing pressure is visible in consumer hardware: MacBook Pro from $1,699 to $1,999, Microsoft Surface Pro from $999 to $1,599, and an estimated iPhone 18 from $1,099 to $1,299. DRAM prices are up 600% over the past few years, with most of the increase coming in recent quarters.
The winners are the memory suppliers. Micron, Samsung, and SK Hynix are capturing extraordinary economics. A Samsung memory worker with a $52,000 base salary is expected to receive a $410,000 bonus, while SK Hynix employees are expected to receive $454,000. In South Korea, luxury jewelry sales are up 146% year-over-year, and luxury watch sales are up 85% year-over-year.
Markets are reflecting the same split. Semiconductors posted a 246% rolling 14-month return, above the 234% dot-com peak from February 2000, while the Magnificent Seven are down 7% this year. Micron is up 10x over the last year, with net income up 15x; last quarter it reported $28 billion in net income, nearly matching Apple’s $29 billion.
This is the AI trade’s tension in one chart: the infrastructure suppliers are minting profits, while the buyers may eventually face pressure if free cash flow weakens and debt or equity issuance becomes necessary.
Compute Becomes a Toll Road, But Not All Toll Roads Are Equal
Meta’s potential move to monetize “excess compute” by creating its own cloud business may be one of the week’s most important strategic signals. The company could pursue an Amazon AWS-style model, hosting large language models and collecting usage fees, or a NeoClouds-style GPU rental business with hourly, monthly, or yearly contracts.
The bigger implication is that Meta may be reassessing its position in the frontier model race against OpenAI, Anthropic, and Google’s Gemini. Excess compute suggests the company may have overbuilt around an unclear strategy. That does not mean Meta loses AI entirely; it could still dominate low-cost content generation across Instagram and other platforms, with the sharp forecast that “AI slop is going to become a gigantic business.”
The likely winners are frontier labs and hyperscalers: OpenAI, Anthropic, Amazon, Google, Microsoft, and potentially Meta. The losers may be the NeoClouds, including CoreWeave, Nebulus, and Iron, especially if three- or five-year hyperscaler contracts worth 10-15 billion roll off and customers migrate to lower-cost platforms.
That distinction fits Palantir’s CEO’s comment that the two places that actually make money are “our application layer called ontology and compute.” The application and data layer includes businesses like Snowflake, Datadog, Databricks, and database assets inside Microsoft, Oracle, IBM, and AWS. But compute is more complicated. Nvidia is expected to report gross margins of 75% in roughly 45 to 50 days, while Micron recently reported gross margins above 80%. Those economics are very different from running expensive data centers where pricing may eventually fall to just above the cost of electricity.
Valuation, Debt, and Market Timing
Aswath Damodaran’s warning gave the week its valuation backbone: “This has been the biggest infrastructure run-up I’ve ever seen in business.” His point is that this cycle is not the dot-com bubble repeated. The late-1990s internet boom was mostly equity-funded software speculation. Today’s AI boom is heavier: real assets, debt, private capital, and infrastructure.
A cited IBM CEO comment put the scale in perspective: $80 billion of capex per gigawatt for hyperscaler AI infrastructure. At 100 gigawatts, that implies $8 trillion of capital, with $800 billion of profit required to cover interest alone.
That does not prove AI is a bubble. It means investors must separate technological importance from financial return.
The same discipline applies to broader markets. New highs are not automatically bearish. From 1970 through May 2026, Ben Felix reviewed 10 developed stock markets and the world stock market and found that all-time highs are common: 20% of months across 10 countries, 30% in the US, 23% in Canada, and 31% for the world index. The real issue is valuation, especially with the Shiller CAPE ratio close to dot-com-era levels.
Macro Gives the Bulls Some Help
The June jobs report strengthened the case for a more dovish Fed. The economy added 57,000 jobs in June, below expectations for 115,000. The CME FedWatch Tool showed the odds of a July 29 rate hike falling sharply: 1 week ago, a 32.1% chance of a hike; yesterday, a 71.1% chance of no change; after the report, an 82.4% chance of no change; and now only a 17.6% chance of a hike.
That helps the broadening-rally case for banks, housing stocks, gold, materials, and cyclicals. Banks passed stress tests and moved toward dividends and buybacks, which could support confidence and potentially revive M&A activity.
Still, the labor market is cooling, not collapsing. Unemployment declined from 4.3% in May to 4.2% in June, while JOLTS openings rose to a 2-year high. But Technology has seen 139,156 job cuts so far this year, up 83%, and AI was cited for 101,743 job cuts, or 23% of all job cuts.
Looking Ahead
Next period, the key question is whether macro support can offset AI valuation risk. Watch Fed expectations into July 29, memory pricing, hyperscaler capex commentary, and whether semiconductor margins remain exceptional. Also watch for breadth: if banks, housing, materials, and cyclicals start participating, the rally becomes healthier. If AI leaders keep carrying the tape alone, the market remains powerful but fragile.