AI Markets: CoreWeave soars, Microsoft takes on Nvidia, SpaceX eyes $600

CoreWeave posts a record quarter fueled by the rush for AI compute, while Microsoft pushes ahead with its own chip to loosen Nvidia’s grip. On the space front, Morgan Stanley puts a number on a $600 scenario for SpaceX, while Norway’s sovereign wealth fund discloses its first stake in Elon Musk’s rocket company.

CoreWeave: a record quarter that doesn’t reassure everyone

CoreWeave more than doubled its revenue in the second quarter of 2026, riding the corporate rush for AI compute. The stock jumped on the news, even as losses widened under the weight of the debt financing this expansion.

Key takeaways:

  • Q2 revenue: $2.58 billion, up 112% year over year, ahead of Wall Street expectations.
  • Backlog of $104 billion at the end of June (up 246% year over year), plus more than $25 billion in new contracts signed in early July that aren’t counted yet.
  • Net loss widened to $626 million (versus $290 million a year earlier), driven by interest expense that more than doubled to $640 million.
  • Shares up 13% to 16% after the close; 2026 capex budget raised to $35-39 billion, versus the $31-35 billion forecast in May.

Demand is outrunning supply

CoreWeave rents out servers packed with Nvidia chips to AI giants (Microsoft, Meta, OpenAI…) that don’t have the time or the appetite to build their own data centers. The result: the company is posting triple-digit growth quarter after quarter, and its backlog already dwarfs its annual revenue. To keep up, the company just raised its 2026 capex budget again, proof that demand for AI compute isn’t letting up, at least for now.

The price of growth on credit

The flip side is that CoreWeave is funding this expansion with massive borrowing. Interest expense more than doubled in a year and now weighs heavily on the books: the quarterly loss widened even though revenue more than doubled. That’s exactly the dynamic the most cautious investors (Michael Burry among them) are watching in the debate over a possible “bubble” in debt-financed AI data centers. For a European investor, CoreWeave (Nasdaq: CRWV) isn’t PEA-eligible and is only available through a regular brokerage account or a UCITS-format U.S. equity/tech ETF.

Sources:

Microsoft pushes ahead with its own AI chip to loosen Nvidia’s grip

Microsoft is in talks with TSMC to manufacture more than 300,000 units of its new in-house Maia 300 chip, which is expected to be unveiled as early as September. The goal: reduce its dependence on Nvidia GPUs, which weigh heavily on the costs of its Azure cloud.

Key takeaways:

  • Microsoft is discussing with TSMC the production of more than 300,000 Maia 300 chips for delivery in 2027, with an eventual goal of topping one million units.
  • Marvell Technology, the chip’s design partner (as it is for Amazon’s Trainium), jumped 8.7% in premarket trading on the news.
  • The previous generation, Maia 200, only reached mass production in January 2026 and remains confined to two U.S. data centers.
  • The bottleneck is still advanced chip assembly (“packaging”), where Nvidia controls roughly 60% of global capacity, with lead times of 52 to 78 weeks.

Why Microsoft wants its own chip

Every Nvidia GPU Microsoft buys for Azure eats into its cloud margin. By developing its own inference chip, Microsoft hopes to do what Google has done for years with its TPUs and Amazon with Trainium: regain control of its costs and become less dependent on a single supplier holding all the cards. The Maia 300 is set to be unveiled as early as September, an accelerated timeline after a Maia 200 that stayed under the radar.

The indirect winners: TSMC and Marvell

For investors, the story isn’t just Microsoft: every custom chip ordered by a cloud giant benefits the supply chain that builds it. TSMC (Taiwan, available to trade via ADR, not PEA-eligible) remains the only foundry capable of manufacturing these advanced chips. Marvell (Nasdaq, not PEA-eligible), which co-designs Maia just as it already co-designs Amazon’s Trainium, drew direct praise from analysts (Wedbush, Fubon Research). One real hurdle remains: final chip assembly, where Nvidia has reserved most of the world’s capacity with wait times that can exceed a year — Microsoft will have to negotiate a spot in an already full queue before it can produce its chips at volume.

Sources:

SpaceX: Morgan Stanley sees a $600 scenario, powered by AI

SpaceX stock is back at its IPO price after a turbulent summer. Morgan Stanley now puts a number on a $600 bull scenario, while Norway’s sovereign wealth fund discloses its very first stake in Elon Musk’s company.

Key takeaways:

  • Morgan Stanley keeps its base-case target at $300 but lays out a bull case at $600, driven by the growth of SpaceX’s AI business, notably through its coding subsidiary Cursor.
  • Norway’s sovereign wealth fund ($2.3 trillion in assets) reveals it holds 0.05% of SpaceX, a stake worth about $1.22 billion as of June 30 — its very first disclosure on the stock.
  • The stock reclaimed its IPO price ($135) on August 10, for the first time since July 15, after falling as low as $104.83 on August 3 — still about 34% below its June peak.
  • Reminder: in June, SpaceX struck a deal to buy AI coding startup Cursor for $60 billion, a transaction expected to close in the third quarter of 2026.

Where the $600 scenario comes from

Morgan Stanley’s math isn’t built on rockets but on AI. The bank is counting on future revenue from Cursor, the AI-assisted programming tool SpaceX is about to absorb for roughly $60 billion — an acquisition announced in June and slated to close by the end of the third quarter. Morgan Stanley projects Cursor’s annualized revenue at around $8 billion by the end of 2026, then roughly $33 billion in 2030. On top of that, the $600 scenario assumes SpaceX manages to deploy orbital data centers at half today’s cost and to connect Starlink to hundreds of millions of AI robots by 2040 — long-term assumptions, not a sure forecast.

An unexpected new shareholder

Norway’s sovereign wealth fund, the world’s largest institutional investor, disclosed a stake in SpaceX for the first time: about $1.22 billion, or 0.05% of the company. That’s a modest sum next to its other tech bets — the fund holds, for example, $62 billion in Nvidia and $34 billion in TSMC — but it confirms big institutional investors’ interest in the stock, a year after its IPO. For a European saver, SpaceX (Nasdaq: SPCX) remains outside the PEA, like nearly all U.S.-listed stocks.

Sources:

Nvidia’s $500 billion financing plan: bubble or the new normal?

A week after announcing deals with six Wall Street giants to raise more than $500 billion, Nvidia’s AI infrastructure financing plan has analysts split between cautious optimism and uneasy comparisons with the telecom bubble of the 2000s.

Key takeaways:

  • The Motley Fool calculates that Nvidia’s $500 billion target is roughly 20 times the $25.6 billion in vendor financing that telecom equipment makers deployed at the tail end of the dot-com bubble (2000).
  • A structural difference flagged by several analysts: this time, the money comes from Wall Street (Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR), not from Nvidia’s own balance sheet — unlike Lucent and the telecom equipment makers, which had to write off up to 80% of their loans.
  • BofA, for its part, believes the structure reduces Nvidia’s direct financial risk while confirming the strength of chip demand.
  • Nvidia is set to report quarterly earnings later in August, a closely watched event that should bring some hard data to this debate.

The parallel that worries people

The skeptics’ argument: financing your customers so they can buy your own products is never a good sign, even if the mechanics differ from the 2000s. Michael Burry had already called an earlier $250 billion Nvidia guarantee (on an OpenAI data center in Ohio) “circular financing.” The underlying risk is the same in both cases: funding compute capacity on the promise of future revenue that doesn’t exist yet.

What’s different from the telecoms of 2000

The optimists’ argument, BofA included: this time, it’s not Nvidia lending directly off its own balance sheet, but institutional investors putting up their capital in exchange for a return. If a data center project goes south, they absorb the loss, not Nvidia. That protects the chipmaker’s balance sheet without eliminating the fundamental risk: demand for AI compute has to keep growing at the expected pace for these investments to pay off. The answer should come with Nvidia’s quarterly results, due later this month — until then, the stock remains an unavoidable pillar of nearly every global tech equity ETF available in a PEA.

Sources:

OpenAI loses another key executive as the exodus continues ahead of the IPO

Brad Lightcap, OpenAI’s chief operating officer for eight years and a close ally of Sam Altman, has announced his departure. It’s the umpteenth executive exit of 2026, at a time when several outlets are reporting the IPO could slip to 2027.

Key takeaways:

  • Brad Lightcap, OpenAI’s COO since 2018 and one of Sam Altman’s closest lieutenants, is leaving the company to “start something new.”
  • He joins a long list of 2026 departures: Bill Peebles, Kevin Weil and Srinivas Narayanan in April, chief product officer Fidji Simo in July, AI ethics lead Chloé Bakalar, safety lead Johannes Heidecke and “chief futurist” Joshua Achiam, also in July.
  • According to Semafor, this wave of departures coincides with a possible delay of the IPO to 2027, reported by several outlets in recent days, amid what’s seen as cooler market enthusiasm and stiffer competition from Anthropic.
  • The $7 billion employee share buyback announced on August 10 takes on particular significance in light of these repeated departures.

A leadership team thinning out ahead of the IPO deadline

Losing a COO after eight years and half a dozen senior executives in a matter of months is never a neutral signal for investors preparing for an IPO. The timeline for the public filing (S-1), previously floated for late 2026, looks increasingly uncertain: after the August 10 share buyback, funded internally rather than by new investors, TechCrunch had already noted that the move “suggests an IPO may not be imminent.”

Anthropic, the big winner of the timing?

If OpenAI does push back to 2027, Anthropic could become the first major generative AI company to go public, with a tighter timeline of September or early October 2026 reported on its end. For individual investors, neither company is directly accessible yet: only an actual public listing will open the door to investing through a brokerage account, with the PEA remaining reserved for European companies.

Sources:

Edition prepared by our monitoring agent (data and sources as of Wednesday, August 12, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.