Google Burns Cash, Oracle's AI Lease Risk, CXMT's $8.6B IPO

Alphabet reported negative free cash flow for the first time in its public history, while Oracle faces a structural risk in its AI infrastructure bet and China’s CXMT pulled off Asia’s biggest IPO of the year. Three distinct stories, one common thread: the cost of building AI at scale is coming due.

Key points

• Google parent Alphabet posted negative free cash flow of $5.9 billion in Q2, the first such quarter since the company went public, driven by surging AI infrastructure spend.
• Alphabet’s 2026 capital expenditure forecast now reaches as high as $205 billion, even as Q2 revenue of $112.11 billion beat Wall Street estimates.
• A Seeking Alpha analysis argues Oracle is a sell because its long-term lease obligations could become a liability if AI demand softens after sub-leases expire in roughly five years.
• China’s ChangXin Memory Technologies (CXMT) raised $8.6 billion in its Shanghai Stock Exchange debut, making it Asia’s largest IPO so far this year.
• Semiconductor ETFs have pulled back from recent highs, though investor inflows have continued as buyers bet on a recovery.
• Nvidia donated its latest AI supercomputer to the U.S. Navy’s postgraduate school, extending its government and defense relationships.

What does Alphabet’s negative free cash flow actually mean?

The headline is striking because it is genuinely unprecedented for Alphabet. The company has historically been a cash generation machine. Posting a $5.9 billion free cash flow deficit in a single quarter signals how aggressively it is front-loading AI infrastructure investment.

The important context: revenue still grew, and Q2 earnings per share came in at $9.11, beating expectations. The cash burn is a capital allocation choice, not a sign of a struggling business. But a $205 billion full-year capex forecast is a number that demands scrutiny. If AI revenue ramps fast enough to justify it, this quarter will look like a blip. If the monetization timeline slips, investors will revisit this data point.

For now, the market appears to be treating the beat on revenue as the more important signal. But the free cash flow turn is a structural shift worth watching across multiple quarters, not just one.

Why does Oracle’s AI infrastructure play carry asymmetric risk?

Oracle has been one of the more aggressive players in leasing data center capacity to AI customers, positioning itself as an alternative cloud provider riding the hyperscaler buildout wave. The Seeking Alpha analysis published today lays out a specific structural problem: Oracle’s leases on the underlying infrastructure run longer than the sub-leases it signs with AI customers.

The argument is that Oracle could correctly forecast strong AI demand over the next few years, customers pay up, everything looks fine, and then in five years those sub-leases expire. If AI demand has softened by then, Oracle is left holding long-term lease obligations with no one to pass the costs to. The analysis rates ORCL a sell on that basis.

This is a bear case, not a consensus view, and Oracle has been growing cloud revenue at a rapid clip. But the structural mismatch between asset duration and revenue duration is a legitimate risk that investors in infrastructure-heavy AI plays should understand. It applies beyond Oracle: any company that is leasing rather than owning capacity, and then re-leasing at shorter terms, carries some version of this exposure.

CXMT’s IPO and what it signals for AI chip competition

ChangXin Memory Technologies raised $8.6 billion on the Shanghai Stock Exchange, making it the largest IPO in Asia this year. CXMT is China’s largest memory chipmaker, and its listing is the direct result of more than a decade of state-backed investment by the city of Hefei in domestic semiconductor development.

Memory chips are foundational to AI infrastructure. High-bandwidth memory (HBM) is a critical bottleneck in AI accelerator design, currently dominated by SK Hynix, Samsung, and Micron. CXMT’s ability to raise this kind of capital signals that China’s push to close the gap in memory is serious and well-funded, even if the company remains behind on leading-edge process nodes.

For investors in U.S. memory and AI chip stocks, CXMT’s IPO is a medium-term competitive signal rather than an immediate threat. But it underscores that the semiconductor supply chain for AI is being actively contested at a national level, and that dynamic shapes the long-term pricing and market share picture for Western chipmakers.

Semiconductor ETFs: pullback with persistent inflows

Chip ETFs have retreated from recent peaks after the extraordinary run that included names like Intel recovering sharply and TSMC’s sustained gains. The notable data point is that inflows have continued even as prices have dropped, suggesting institutional and retail buyers are treating the pullback as an entry point rather than a warning sign.

That divergence between price action and flows is worth watching. It either means the dip buyers are right and the AI infrastructure cycle has more runway, or it means positioning is getting crowded at exactly the wrong moment. The CXMT IPO and Alphabet’s capex numbers both support the view that spending is not slowing. Whether that spending translates to chip company earnings at the rate the market expects is the open question.

This update is for informational purposes only and does not constitute investment advice.