AI Stocks & Markets

Today's Stories

1. 1

   Burry shorts Caterpillar after 172% surge; analyst says bet will miss the mark

   Michael Burry, the investor famous for predicting the 2008 financial crisis, disclosed a short position in Caterpillar at $1,060.98 per share on Tuesday. Caterpillar shares fell nearly 7% by Wednesday and as much as 4% by Thursday, hitting their lowest point since mid-June at around $949 per share. Burry said the stock is overvalued after surging more than 100% over the past year and reaching its highest price-to-sales ratio in three decades. Burry believes the AI infrastructure boom has inflated Caterpillar's valuation, viewing it as part of a broader AI market bubble (he has compared the market to the 1999–2000 bubble). However, analyst Sergey Glinyanov argues Burry is missing the real driver: Caterpillar is benefiting from a structural shift in spending on power infrastructure. AI data centers are increasingly buying Caterpillar's diesel and natural-gas generator systems as alternatives to an aging electrical grid that cannot keep up with soaring energy demands. Glinyanov points to the company's strong Q1 results—sales jumped 22% year-over-year to $17.4 billion（約2.8兆円）—as evidence the strength is fundamental, not hype.

   Glinyanov's firm has set a price target of $910 for Caterpillar, indicating a "potential near-term pullback." The key risk is whether large cloud providers (hyperscalers) continue to spend aggressively on data centers; if they slow their investments or face deteriorating cash flow or rising debt burdens, the optimism around Caterpillar could fade quickly.

2. 2

   Memory chip ETF hits record $10B in assets, nears stock split

   The Roundhill Memory ETF, which tracks companies making high-bandwidth memory chips used in AI accelerators, launched on April 2 and gathered nearly $10 billion（約1.6兆円） in net assets in under 45 days—a record pace for ETF asset growth. The share price rose from around $26 at launch to more than $81 at its peak, a roughly 78% gain in under two months. Memory is a crucial but overlooked part of AI infrastructure. As AI models grow larger and inference workloads become more complex, the amount of memory required per chip is increasing—a structural trend rather than a short-term phenomenon. The fund gives retail investors exposure to major memory producers like SK Hynix and Samsung Electronics, which have been difficult to access directly in the U.S. market.

   The fund's rapid price rise may trigger a stock split later this year to make share prices more accessible to everyday investors, similar to what Roundhill has done with other thematic funds. However, the concentration risk is real: Micron Technology accounts for roughly 26% of the portfolio (including 10% in leveraged shares), SK Hynix and Samsung each around 24.5%, with nearly 80% of the fund in four companies.

3. 3

   Palantir, SanDisk post blowout AI quarters on opposite ends of the stack

   Palantir's U.S. commercial revenue jumped 133% to $595 million（約950億円） in Q1 FY2026, driven by AIP (its AI platform) adoption at customers including AIG, GE Aerospace, and Freedom Mortgage. SanDisk's Datacenter segment surged 645% year over year to $1.47 billion（約2400億円） on AI memory demand, with gross margin expanding to 78.4% from 22.5% as NAND flash supply could not keep pace with demand. The two companies illustrate different paths to AI profitability—Palantir through software depth and customer lock-in (its Rule of 40 score reached 145%), SanDisk through hardware scarcity and pricing power. SanDisk's management expects supply to remain tight through the end of calendar year '26 and beyond, suggesting sustained margin gains. Palantir raised full-year 2026 revenue guidance to $7.65 billion（約1.2兆円） to $7.662 billion（約1.2兆円）, roughly 71% growth, signaling strong enterprise AI demand.

   Palantir trades at a P/E near 165 with stock-based compensation of $201.6 million（約320億円） in the quarter, reflecting high valuation and equity costs. SanDisk holds zero debt and has begun buybacks, offering a different risk profile. The contrast highlights the trade-off between software scaling and cyclical hardware economics.

4. 4

   NVIDIA scores Blackwell server order, Palantir partnership

   Bit Origin announced a purchase of approximately $11 million（約18億円） worth of NVIDIA Blackwell B300 AI servers (16 units), with delivery expected in Q3 2026 and deployment planned for a Malaysian data center. Separately, Palantir Technologies announced a strategic collaboration with NVIDIA to build an AI platform for deploying NVIDIA's AI and Nemotron open models in secure sovereign environments, targeting U.S. government and critical infrastructure sectors. The Blackwell server deal represents concrete near-term infrastructure demand for NVIDIA's latest AI hardware, with the customer expected to generate about $360,000 in recurring monthly revenue from the deployment. The Palantir partnership signals NVIDIA's expansion into government and defense AI systems, where data sovereignty and control are critical—a growing market segment for the chip maker.

   The Blackwell servers are expected to arrive in the third quarter of 2026. The transaction structure included $1 million（約1.6億円） in cash and $10 million（約16億円） in equity through pre-funded warrants.

5. 5

   Broadcom vs. ON Semiconductor: Growth vs. Recovery Play in 2026

   Broadcom reported FY 2025 revenue of nearly $63.9 billion（約10兆円）, up approximately 23.9% year-over-year, with net income of roughly $23.1 billion（約3.7兆円） and free cash flow of nearly $26.9 billion（約4.3兆円）. ON Semiconductor reported FY 2025 revenue of nearly $6.0 billion（約9600億円）, down approximately 15.3% versus the prior year, with net income of roughly $121.0 million（約190億円） and free cash flow of nearly $1.4 billion（約2200億円）. Broadcom's strength in AI and data-center demand has driven substantial earnings power and a net margin of close to 36.2%, but its shares already trade at a premium (Forward P/E of 31.5x versus a sector benchmark of 36.4x). ON Semiconductor faces cyclical pressure from automotive and industrial markets—which account for about 51% of its revenue—but trades at a lower Forward P/E of 29.4x and may benefit from a market recovery.

   ON Semiconductor agreed in June 2026 to acquire Synaptics in an all-stock deal valued at about $7 billion（約1.1兆円）, aimed at expanding exposure to physical AI and AI-enabled devices. Broadcom's customer concentration—40% of revenue from its top five end customers—poses a risk if capital spending slows.

6. 6

   Nvidia Partners With 'Neoclouds' Sharon AI, Firmus Technologies

   Nvidia Partners With 'Neoclouds' Sharon AI, Firmus Technologies