Mag 7 Cracks, SK Hynix $64B Bet, SpaceX Gets Ives

This update is a roundup of same-day reporting from the linked sources below, with editorial context from the CPJ Stock Desk.

The AI spending cycle is facing its most direct investor scrutiny yet, with Magnificent Seven stocks losing $2.3 trillion in market value during June alone as markets shift focus from infrastructure buildout to actual returns.

Key points

June erased $2.3 trillion from the combined market cap of the Magnificent Seven, with investors increasingly demanding monetisation evidence ahead of Q2 earnings.
South Korea’s KOSPI dropped more than 6% as reports that Meta plans to sell spare AI computing capacity raised fresh questions about the scale of real chip demand.
SK Hynix announced $64 billion in chip plant spending in Cheongju, part of a broader $2.1 trillion Korean semiconductor investment plan alongside Samsung.
Wedbush’s Dan Ives initiated SpaceX with an Outperform rating and a $190 price target, framing the company as a participant in the hyperscaler wars.
Asia chip stocks broadly declined on reports of OpenAI efficiency gains alongside the Meta cloud capacity news.

Is the Magnificent Seven thesis breaking down?

The group that powered global equity returns for the past three years is now its own headwind. According to Economic Times, markets are rotating away from AI hype and toward free cash flow, profitability, and concrete monetisation timelines. The $2.3 trillion June drawdown is not a blip. It reflects a maturing investor base that has watched capex at Microsoft, Google, Amazon, and Meta compound for seven quarters without a proportionate revenue inflection.

Earnings season will be the real test. Every major hyperscaler is expected to face pointed questions about AI revenue per dollar of GPU spend. The companies that can show clear unit economics will likely decouple from those still in pure build mode.

What is the Korea selloff actually telling us?

The KOSPI’s 6%-plus decline hit semiconductor names hardest. The trigger was a combination of factors: reports that Meta may monetise excess AI compute by reselling capacity to third parties, and separate signals that OpenAI has been making meaningful efficiency gains in its model training. Both stories carry the same implication. If frontier labs need fewer chips per unit of output, and hyperscalers are sitting on spare capacity, the addressable market for new silicon orders narrows.

Asian share markets broadly retreated on Thursday, with South Korea and Japan among the hardest hit. SK Hynix, one of the primary suppliers of high-bandwidth memory for AI chips, sits at the centre of this tension. The company’s $64 billion domestic investment commitment signals long-term conviction in AI-driven demand. But that bet is being made into a market that is, at least for now, questioning the pace of that demand.

SpaceX: from IPO volatility to Wall Street coverage

Three weeks after its record-setting IPO, SpaceX received its first formal analyst initiation. Wedbush’s Dan Ives assigned an Outperform rating and a $190 price target, and notably positioned SpaceX as a competitor in the hyperscaler infrastructure race rather than a pure aerospace story. Ives has a track record of early, bullish calls on AI-adjacent names, and his framing suggests he sees Starlink’s compute and connectivity infrastructure as a direct play on enterprise AI adoption.

Separately, a Wall Street Journal report claiming SpaceX was developing a smartphone-like AI hardware device was flatly denied by Elon Musk, who called the report “utterly false.” The denial did not stop the launch of SpaceX leveraged ETFs from proceeding, with ETF provider Themes ETFs moving forward amid what VettaFi describes as record mid-year ETF inflows overall.

OXMIQ: licensing as an alternative to Nvidia

Covered briefly yesterday, but worth the added context: OXMIQ’s $35 million Series A is structurally different from most chip startups. Rather than manufacturing its own silicon, the company founded by ex-Intel executive Raja Koduri licenses its OxCore GPU architecture as compute IP, letting governments and enterprises design custom processors. The target markets are India and Southeast Asia, where sovereign AI ambitions are real but the cost of proprietary Nvidia-based infrastructure is prohibitive. The model more closely resembles Arm Holdings than it does a traditional fabless chipmaker.

Nothing on this site constitutes investment advice. This update is for informational purposes only.