AI Markets: Nvidia as OpenAI’s guarantor, Starship succeeds, Korean memory records

Nvidia is reportedly ready to guarantee $250 billion so OpenAI can lease a giant data center in Ohio, while SpaceX scores a technical win and a first lukewarm analyst rating. On the chip side, Samsung and SK Hynix lock in $950 billion in memory contracts through 2030, without their stocks getting any boost.

Nvidia ready to guarantee $250 billion for OpenAI: AI’s circular financing hits a new level

According to the Wall Street Journal, Nvidia is negotiating a financial guarantee of roughly $250 billion to let OpenAI lease a giant data center in Ohio. The deal reignites the debate over the circular financing behind the AI bubble.

Key points:

  • On July 26, the WSJ reported that Nvidia is discussing a guarantee of about $250 billion covering the lease and debt of a 10 GW data center being developed in southern Ohio by SB Energy, SoftBank’s energy subsidiary.
  • The full project, chips included, would top $500 billion; Nvidia is also reportedly discussing financing for OpenAI’s chip purchases, up to an additional $350 billion.
  • Investor Michael Burry reacted publicly on X: “Nvidia is going to guarantee $200 billion of ChatGPT spending on Nvidia chips. Around and around we go.”
  • Nvidia stock barely moved on Friday (-0.92%, at $206.84), a sign the market hasn’t made up its mind on a deal that is still only under negotiation.

A deal that loops back on itself

Here’s how it works: Nvidia guarantees SB Energy repayment of the lease and debt taken on to build the data center leased by OpenAI. If OpenAI’s revenue doesn’t grow fast enough to cover the lease, Nvidia absorbs the shortfall. For OpenAI, the appeal is breaking out of a model where it rents compute from Microsoft, Amazon or Oracle and becoming the owner of part of its infrastructure. For Nvidia, the deal locks in chip demand for several years, at the cost of direct financial exposure to OpenAI’s ability to generate revenue. That loop is what worries people: one way or another, Nvidia would be financing its biggest customer’s purchases of its own chips. Talks are still at an early stage and could fall through or see their terms change; meanwhile, Anthropic, Microsoft and Google have reportedly also floated similar arrangements with the U.S. administration in recent weeks, a sign that this kind of guarantee could become a standard tool for financing AI data centers.

What it means for a European investor

Neither Nvidia nor OpenAI (which is private) is eligible for the PEA: Nvidia goes in a regular brokerage account, or into a PEA via UCITS ETFs that track the Nasdaq-100 through synthetic swaps (Amundi, for example), which replicate the exposure without holding the stock directly. For these holders, what matters isn’t this particular contract but what it reveals: a growing share of Nvidia’s projected growth relies on financial commitments to its own customers rather than on traditional cash sales. Worth keeping a close eye on how this plays out, especially if a signed agreement ends up putting firm numbers on amounts that are still only under negotiation.

Sources:

SpaceX: Starship nails Flight 13, but analysts are split on the stock price

After two failed attempts, Starship completed its mission on July 24. On the market side, two completely opposite analyst views have just surfaced, days before a key deadline for SpaceX stock (SPCX).

Key points:

  • Starship lifted off on July 24, deployed operational Starlink V3 satellites into orbit for the first time, and pulled off a controlled splashdown of the upper stage in the Indian Ocean.
  • On July 26, HSBC initiated coverage of SPCX with a “Hold” rating and a $115 price target, almost exactly Friday’s closing price ($115.07).
  • Morgan Stanley, by contrast, reiterated its “Overweight” rating with a $300 target, arguing the market undervalues SpaceX’s space and connectivity businesses.
  • The post-IPO lock-up expires on August 6, the next deadline that could shake up the stock’s still very limited float.

Two opposite readings of the same story

HSBC justifies its caution with doubts about xAI, SpaceX’s AI subsidiary, which it sees as lagging the industry leaders in enterprise adoption and computing power. Morgan Stanley flips the argument: by its own analysis, if the stock fell back to $100, that would amount to assigning almost no value to the AI business, a scenario it considers excessive given fundamentals that are “largely unchanged” since the IPO. The gap of more than 150% between the two price targets shows how much SpaceX’s valuation still hinges on assumptions about a young AI business rather than on the group’s historical core (launches, Starlink). As a reminder, SPCX joined the Nasdaq-100 on July 7, just 15 trading sessions after its IPO, a fast-tracked entry thanks to loosened eligibility rules, but its weight in the index remains below 1% because of the limited float (a modified market-cap methodology for low-float stocks). The QQQ and QQQM ETFs had to buy SPCX to track the index, with passive flows estimated at between $4.3 billion and $27 billion depending on the analyst; that weight could climb to as much as 5% after the lock-up expirations scheduled for 2027.

Access for a European investor

SpaceX is still a U.S. stock, held in a regular brokerage account (not PEA-eligible). Holders of Nasdaq-100 ETFs or broad ETFs tracking U.S. indexes, including some Europe-domiciled UCITS funds that are PEA-eligible through synthetic replication, have already had indirect exposure to SpaceX since early July, whether they chose it or not. It’s a useful reminder: when an IPO enters a major index, its risk spreads well beyond the people who buy the stock directly.

Sources:

Samsung and SK Hynix lock in $950 billion in AI memory chips, and their stocks fall anyway

The two Korean memory giants have signed record supply contracts with Nvidia and Broadcom. Yet their shares fell on the day of the announcement, a paradox reminiscent of what we already saw with U.S. chipmaker Micron.

Key points:

  • SK Hynix and Samsung Electronics have signed AI chip supply agreements totaling about $950 billion through 2030, unveiled around July 25-26.
  • SK Hynix would supply about $750 billion in chips, mainly to Nvidia (a 5-year contract including next-generation HBM4 memory); Samsung signed a 5-year memorandum with Broadcom worth about $200 billion, covering HBM memory, advanced foundry and packaging.
  • That figure represents the projected value of firm purchase commitments over five years, not a one-time payment: memory contracts, historically annual, are turning into long-term agreements (LTAs).
  • Despite the announcement, SK Hynix and Samsung fell about 8.3% and 7.6% respectively during the week of July 24-27, amid a broad sell-off in Korean tech.

Record deals, profit-taking on the market

The pullback in both stocks doesn’t call the agreement itself into question: Korean investors mostly seem to be taking profits after a spectacular year-to-date rally in both names, fueled by demand for HBM memory. The correction was also amplified by tighter regulation: starting July 31, South Korea’s Financial Services Commission is raising deposit requirements on leveraged ETFs, pushing some retail investors toward highly leveraged CFDs on these two stocks, which adds to the volatility. The pattern echoes the Micron case covered in a previous edition: record results or contracts in AI memory don’t prevent a sell-off, as investors stay nervous about the sustainability of demand three to five years out rather than about the immediate order book.

Access for a European investor

Samsung Electronics and SK Hynix are listed in Seoul and can only be held in a regular brokerage account, with no direct PEA equivalent. Indirect exposure is available through global semiconductor UCITS ETFs (iShares, VanEck) that include a Korean weighting, but it is generally a minority position compared with the weight of Nvidia, TSMC or ASML in those same funds, so check the holdings before treating them as a bet on Korean memory.

Sources:

Humanoid robotics: BYD joins the race while Tesla warns of a very slow start for Optimus

An accidental leak has revealed the humanoid robot project of China’s top EV maker, just as Tesla admits that production of its own Optimus robot will take longer than expected.

Key points:

  • A post accidentally published and then deleted by a BYD center in Zhengzhou revealed a humanoid robot called “Xiao Di,” with a public unveiling expected in early August.
  • BYD has been working on the project since late 2024 through its 15th division, with plans to use it first in dealerships (2-3 robots per location to greet customers) and then in factories.
  • At Tesla, the Optimus production line in Fremont (former Model S/X lines, now decommissioned) was still being installed as of July 23, with no units produced; Elon Musk warned in early July that the ramp would be “extremely slow,” sending the stock down 7.5% on the day he said it.
  • In Japan, engineers who tore down a Chinese Unitree G1 publicly admitted they can’t catch up with China’s lead in cost and engineering in the short term.

China speeds up while Western timelines slip

BYD joins Xpeng (and its Iron robot) in China’s embodied robotics race, with a structural advantage: the industrial know-how and cost base of its auto production chain, which can be reused directly to mass-produce robots. The pace of Chinese announcements (Unitree, LimX, Xiaomi, and now BYD) contrasts with the more cautious timeline on the Western side: Tesla is still targeting 50,000 to 100,000 Optimus units in 2026, but Musk himself calls that number “impossible to predict,” since robotics lacks the manufacturing repeatability of cars (around 10,000 unique parts to assemble for a single robot). This caution isn’t unique to Tesla: it confirms the view already voiced by Unitree’s founder, who put robotics’ “ChatGPT moment” still one to three years away.

Access for a European investor

None of the companies mentioned (BYD, Tesla, Unitree, Xpeng) is directly PEA-eligible for a French retail investor outside a regular brokerage account. The KOID thematic ETF, eligible in its UCITS version on the Paris exchange, remains the simplest route to diversified exposure to the sector without betting on a single manufacturer or on any one of them hitting its production schedule.

Sources:

AI data centers: the bond market starts charging the sector’s giants more for their debt

After months of massive borrowing to fund AI data centers, bond investors are starting to demand a higher risk premium. Two concrete cases this week, at Meta and CoreWeave, show it.

Key points:

  • Meta is looking to raise $12 billion through a special-purpose vehicle backed by BlackRock to fund a data center of nearly 1 GW in El Paso, Texas; investors are demanding a yield above 7%, about 0.4 percentage points more than on its previous Hyperion deal ($27 billion raised in October 2025).
  • CoreWeave has lost about 30% of its market cap in a month, with interest expense ($536 million in the last quarter alone) eating up nearly half of its adjusted EBITDA, against $50.8 billion in liabilities and $4.8 billion in equity.
  • According to CNBC (July 24), credit spreads for Google, Amazon and Meta are widening as the hyperscalers get ready, starting next year, to spend more on capex than they generate in free cash flow.
  • Oracle’s 5-year CDS, already hit by an S&P downgrade in early July, now serves as an informal barometer of market nervousness about AI-related debt, and sits at a multi-year high.

A signal to watch ahead of quarterly earnings

This credit tightening says nothing about demand for AI compute itself, which remains strong: it’s about operators’ ability to turn that demand into enough cash to repay ballooning debt. CoreWeave’s case is more acute than Meta’s: the company was shaken by a report that Meta, its customer, is exploring renting out its own excess capacity to other companies rather than continuing to rely on providers like CoreWeave. For Meta, the higher yield demanded is more modest but reflects the same caution among lenders, at a time when Apollo’s chief economist has warned that this wave of AI-related borrowing could, on its own, help push interest rates higher across the board.

What it means for a European investor

This bond market signal usually comes before moves in stocks: it’s worth tracking whether or not you hold Meta, CoreWeave or Oracle shares directly, since it indirectly affects anyone holding tech or AI equity ETFs with heavy exposure to the U.S. hyperscalers. A lasting rise in the cost of AI debt automatically feeds through to the future margins of the whole sector, well beyond the companies mentioned here.

Sources:

Edition prepared by our monitoring agent (data and sources as of Monday, July 27, 2026), reviewed under our editorial policy. General information, not investment advice: investing carries a risk of capital loss.