AI Markets: SpaceX, the OpenAI and Anthropic IPO race, chips under pressure

SpaceX faces its first financial checkpoint as a public company under a hail of bearish bets, while OpenAI beefs up its board and Anthropic kicks off its pre-IPO roadshow. Between a European humanoid robotics unicorn, a new AI chip system at Microsoft and an unexpected bill for Oracle, the week confirms just how fast AI is redrawing the balance of power on the stock market.

SpaceX: August 4 is the date, with short sellers circling

SpaceX has set August 4 as the date for its very first quarterly results as a public company, a milestone that coincides with the release of a first tranche of shares locked up since its IPO. The stock, which has slipped back below its IPO price, is also facing a record level of bearish bets.

Key takeaways:

  • SpaceX (Nasdaq: SPCX) will report its second-quarter 2026 results on August 4, on an audio call at 4:30 p.m. (New York time) streamed on X.
  • That date automatically triggers the first post-IPO “lock-up” expiration, which could put additional shares back on the market.
  • According to S3 Partners data, short positions now represent about 32% of the listed float (about 206 million shares, ~$25 billion), up from 29% a week earlier and barely 5 to 7% a month before.
  • Elon Musk warned that the “probability of survival” for short sellers who stay positioned against SpaceX for the long haul is “very low”; Starship test flight No. 13 is targeted for July 23, after a last-second scrub.

A thin float that amplifies every move

Since its IPO on June 12 at $135 per share — a record raise of about $86 billion — only a fraction of SpaceX’s capital trades freely: the float is estimated at about 4% of shares. That scarcity explains why the stock was able to climb as high as $225.64 before falling back: it now trades around $120 to $126, roughly 40% below its post-IPO peak and slightly below its IPO price. Against that backdrop, bearish bets climbing to a third of the listed float is a strong signal: it reflects sharp disagreement among investors over the company’s valuation, torn between Starlink’s recurring revenue and the more cyclical launch and Starship business.

Two catalysts to watch in early August

August 4 will be the first time the market gets official financial statements from SpaceX: margins, the revenue split between Starlink and launches, cash trajectory. It’s also the day the first lock-up tranche expires on shares held by employees and early investors, which could increase the supply of available stock. Starship flight No. 13 adds a third wildcard: a technical mishap would feed the short sellers’ narrative, while a success could trigger a short squeeze, given the small float. For a European investor, SPCX remains a Nasdaq stock available only through a standard brokerage account, not eligible for the PEA, in a name whose volatility will stay high as long as the float remains this thin.

Sources:

AI lab IPOs: OpenAI bulks up its board, Anthropic launches its roadshow

OpenAI and Anthropic, the two generative AI heavyweights eyeing a stock market listing, hit new milestones this week. The former strengthened its board with two top-tier finance executives, while the latter began meeting its future investors.

Key takeaways:

  • OpenAI appointed David Vélez (founder and CEO of Nubank, Latin America’s largest digital bank) and Robin Vince (CEO of BNY) to the boards of the OpenAI Foundation and OpenAI Group PBC.
  • These additions strengthen the financial governance of OpenAI, which confidentially filed for an IPO in early June 2026.
  • At Anthropic, Goldman Sachs, Morgan Stanley and JPMorgan have started scheduling meetings between management and prospective investors, the classic opening phase of the “roadshow” that precedes an IPO.
  • If the usual timeline holds, Anthropic could file a public registration statement (S-1) in August or September, targeting a Nasdaq listing around October-November, based on its latest private valuation of $965 billion.

OpenAI strengthens its governance ahead of the IPO

Bringing in two executives from banking and fintech is no accident: Robin Vince spent 26 years at Goldman Sachs before running BNY, one of the world’s oldest financial groups (more than 240 years old); David Vélez built Nubank to more than 135 million customers in Latin America. Their profiles — compliance, risk, capital markets — are exactly what a future public issuer needs to bring to its board before facing institutional investors. The announcement comes on top of the confidential filing made in early June, with no official date or valuation range disclosed yet.

Anthropic moves to the investor phase

Anthropic’s timeline is better documented: the confidential filing was submitted on June 1, and the lead banks are now organizing meetings with potential investors — a step that typically comes four to eight weeks before the public filing of the prospectus. If that pace holds, Anthropic could make its S-1 public in August or September, with an actual listing in the fall, likely on the Nasdaq. The reference valuation remains the one from its May 2026 funding round ($65 billion raised, $965 billion valuation), but no official revenue figure has been disclosed since.

For a European investor, both names remain off-limits until they list, probably only on a US exchange, meaning outside the PEA and through a standard brokerage account. SpaceX’s recent experience — thin float, extreme volatility from the very first weeks — offers a preview of the valuation debate that could accompany these upcoming IPOs.

Sources:

Humanoid robotics: London unicorn Humanoid raises $152 million

A young British startup has just become Europe’s first unicorn entirely dedicated to humanoid robotics, just as the boss of China’s Unitree is tempering expectations about the technology’s timeline.

Key takeaways:

  • Humanoid, a London company founded two years ago, raised $152 million in a Series A led by Prime Movers Lab, with participation from Schaeffler, Bosch, Fubon Financial Holding Venture Capital and Aglaé Ventures, at a post-money valuation of $1.35 billion.
  • The company claims the title of Europe’s first “pure-play” humanoid robotics unicorn, bringing its total funding to $270 million.
  • It signed a binding agreement with German auto supplier Schaeffler to deploy 1,000 to 2,000 robots in its factories by 2032, and is relying on Bosch to build up to 100,000 units of its wheeled HMND 01 robot over five years.
  • Unitree founder Wang Xingxing, for his part, estimated that robotics’ “ChatGPT moment” — the breakthrough that would make humanoids truly autonomous in an unfamiliar environment — was still one to three years away.

A bet on certification rather than two legs

Unlike Figure, Tesla Optimus or Unitree, Humanoid chose a wheeled base rather than articulated legs for its HMND 01 robot. The choice, less flashy, serves a specific industrial goal: speeding up certification in European production environments, where safety standards for mobile machines on factory floors are long-established and well documented, unlike bipedal robots, which are still in a regulatory gray zone. That’s what allows Humanoid to announce commercial pilots and a quantified rollout (1,000 to 2,000 units at Schaeffler) while most competitors are still at the demo stage.

Schaeffler, a listed entry point for European investors

Humanoid remains a private company, so it can’t be bought directly on the stock market. But one of its shareholders and customers, Schaeffler, is listed on the Frankfurt Stock Exchange and available through a standard European brokerage account: it offers indirect, minority exposure to this bet on industrial robotics. The deal fits into a broader trend we’ve already covered in previous editions (Unitree’s IPO in Shanghai, the acquisition of Boston Dynamics, thematic ETFs such as KOID): it confirms that Europe, long absent from the humanoid race, now has a credible player in it, with a local industrial base (Bosch, Schaeffler) acting as backers rather than mere subcontractors.

Sources:

AI chips: AMD launches Helios, Microsoft rolls it out at scale in Azure

AMD unveiled its first rack-scale AI system, designed to take on Nvidia head-on, and immediately convinced Microsoft to put it in its Azure data centers. The market cheered the news: AMD shares jumped more than 7% in its wake.

Key takeaways:

  • The “Helios” rack packs 72 Instinct MI455X accelerators and 31 TB of HBM4 memory, paired with sixth-generation EPYC “Venice” processors and Pensando networking chips, for up to 2.9 exaflops of FP4 compute.
  • Unlike Nvidia, AMD is betting on open standards — UALink between GPUs within a single rack, Ultra Ethernet Consortium specifications between racks — rather than a proprietary interconnect.
  • Microsoft announced it will deploy Helios in Azure starting in the second half of 2026, through a new family of virtual machines dedicated to production AI inference: AMD’s highest-profile customer for this product to date.
  • Meta, OpenAI, Oracle and Tata Consultancy Services are also among the early adopters; AMD says eight of the world’s ten largest AI companies already use its Instinct chips.

A credible alternative to Nvidia’s closed ecosystem

AMD’s bet isn’t just about raw power, but about openness: by relying on industry-shared interconnect standards rather than proprietary technology like Nvidia’s NVLink, Helios makes it easier for hyperscalers to mix suppliers and negotiate prices. Microsoft’s immediate buy-in — going as far as creating a dedicated line of virtual machines — is the most concrete signal yet that a major cloud customer is ready to deploy AMD chips “at scale,” and not just in testing.

What it means for investors

The market’s reaction — AMD up more than 7% in the announcement session alone — reflects a shift in perception: AMD is no longer seen as merely a challenger to Nvidia on paper, but as an AI infrastructure supplier deployed in production at a major hyperscaler. For a European investor, AMD (Nasdaq) remains available only through a standard brokerage account, outside the PEA. The story does, however, indirectly benefit European players in the supply chain, starting with ASML (Euronext Amsterdam, PEA-eligible), whose lithography machines are needed to make these chips, whichever designer wins the Nvidia-AMD battle.

Sources:

Oracle: its credit downgrade turns into a $7 billion bill

Wisconsin’s energy regulation has just caught up with Oracle: because its credit rating fell below the required threshold, the company now has to post a $7 billion guarantee for its giant data center, a cornerstone of its contract with OpenAI.

Key takeaways:

  • Wisconsin’s regulator (the Public Service Commission) refused to reverse a rule imposed on utility We Energies that requires Oracle to provide a guarantee (“letter of credit”) of more than $7 billion, at an annual cost of over $100 million.
  • This so-called “very large customer” clause applies to any company whose credit rating falls below A-, and S&P Global Ratings downgraded Oracle to BBB- on July 9, 2026.
  • The data center in question, in Port Washington (Wisconsin), is expected to approach 1 gigawatt of capacity and is a cornerstone of the $300 billion contract between Oracle and OpenAI for computing capacity.
  • Oracle is challenging the decision in court (Ozaukee County Circuit Court); the regulator defends the principle that other electricity customers shouldn’t fund infrastructure dedicated to a single private customer.

When a credit rating becomes an operational problem

Our previous editions had already flagged S&P’s downgrade of Oracle (BBB→BBB-) and Moody’s negative assessment, against the broader backdrop of the debate over a possible bubble in AI data center spending. This Wisconsin episode shows, for the first time, a concrete consequence of that downgrade with a dollar figure attached: a regional utility mechanically applies a contractual clause that turns a simple rating agency move into a multibillion-dollar bill, with a direct impact on the timeline of a strategic project.

A test for the entire AI data center sector

Beyond Oracle, this episode illustrates a structural risk for every operator financing power plants and power lines dedicated to AI: energy regulators, in the US and potentially in Europe too, are increasingly looking to shield other consumers from the cost of these giant infrastructure projects. For a European investor, Oracle (NYSE) remains available through a standard brokerage account, outside the PEA; this kind of regulatory risk also deserves a close look from anyone holding thematic funds or ETFs exposed to data centers and AI infrastructure.

Sources:

Chip supply chain: TSMC raises prices, ASML validates its most advanced technology

Two separate but complementary stories confirm the current balance of power in semiconductors: Taiwanese foundry TSMC is preparing an across-the-board price increase for 2027, while Dutch equipment maker ASML clears a major technical milestone at Intel.

Key takeaways:

  • TSMC has finalized base price increases of 5 to 10% for 2027 with its customers, on advanced chips as well as mature technologies (12, 16 and 28 nanometer nodes), according to Nikkei Asia.
  • Additional high-performance computing orders placed beyond initial forecasts will carry an extra 10 to 15% premium, which could push the total surcharge above 10% on some advanced chips.
  • On the ASML side, Intel Foundry has begun volume production on its 18A node using the Dutch company’s “High-NA” EUV lithography for the first time, on part of its Core Ultra Series 3 (“Panther Lake”) processors.
  • ASML used the announcement to confirm a 30% annual increase in EUV and DUV immersion production capacity in 2027 and 2028, to meet demand for AI chips.

TSMC passes on costs from a position of dominance

With a near-monopoly on manufacturing the most advanced chips, TSMC wields pricing power that is rare in the industry: the announced increase, meant to cover rising costs for materials, equipment and building new fabs abroad, will automatically flow through to its customers’ margins — Nvidia, AMD and Apple first among them — or to the final prices of the products that use these chips. TSMC has not commented publicly beyond a cautious line: its pricing policy is said to be “strategic, not opportunistic.”

ASML validates ten years of R&D investment

For ASML, High-NA lithography entering volume production at Intel is an industrial milestone years in the making: it’s the first time this technology, the most advanced the company has ever brought to market, is being used to make chips in high volume. This technical validation, combined with a capacity increase announced for two consecutive years, supports the so-called “picks and shovels” investment thesis for AI: whoever wins the battle among chip designers, ASML remains an essential supplier. For a European investor, ASML (Euronext Amsterdam) remains PEA-eligible, while TSMC (listed in Taiwan and via ADRs in New York) is available only through a standard brokerage account.

Sources:

Edition prepared by our monitoring agent (data and sources as of Wednesday, July 22, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.