AI Markets: Anthropic settles its lawsuit, TSMC raises prices, SpaceX under pressure

Between a legal settlement that clears the road to Anthropic’s IPO, TSMC’s price hike and a SpaceX caught between a Starship delay and an upcoming lock-up expiration, this week confirms that the AI value chain remains under close watch by the markets. Here’s a roundup of the stories that matter for European investors, with no recommendations.

Anthropic settles a $1.5 billion lawsuit: one less obstacle before the IPO

A federal judge on Monday approved Anthropic’s record settlement in its copyright lawsuit. For investors watching the company’s IPO filing closely, a major legal risk disappears ahead of a listing it hopes to complete this fall.

Key takeaways:

  • Federal Judge Araceli Martinez-Olguin granted final approval of the settlement, the largest ever recorded in a US copyright case, rejecting arguments that it was too small.
  • The payout works out to $3,000 per work for roughly 500,000 works, split between the authors and publishers involved.
  • Anthropic says more than 91% of the authors and publishers covered by the agreement have already claimed their share of the payment.
  • Some authors and publishers chose to opt out of the class action and are pursuing separate suits against Anthropic, which are still ongoing.

What the ruling actually says

The case dates back to 2024: authors sued Anthropic, accusing the company, backed by Amazon and Alphabet, of using pirated copies of their books without permission to train Claude to respond to human prompts. Judge Alsup, who has since retired, split the difference: he ruled that Anthropic’s use of the works to train Claude was fair use, but found that the company infringed authors’ rights by keeping more than 7 million pirated books in a “central library.” That second point is what exposed Anthropic to colossal damages: a trial was originally set to open in December to determine how much Anthropic owed for the alleged piracy, with potential damages running into the hundreds of billions of dollars.

The $1.5 billion settlement therefore avoids that doomsday scenario. But note the legal nuance: final approval closes this specific case, but doesn’t settle the legal question for the whole industry, because Alsup’s decision was only a trial court ruling, and Anthropic’s choice to settle keeps the case from reaching an appeals court and becoming binding precedent. Other similar lawsuits targeting Google, Meta, Midjourney and OpenAI are moving forward in parallel.

The IPO connection

Why are traders following this case so closely? Because Anthropic is preparing a highly anticipated stock market debut. On June 1, 2026, Anthropic confidentially submitted a draft registration statement to the SEC, after a $965 billion valuation in its Series H-1 round in May. The target listing date is October 2026, though no firm date has been confirmed. Goldman Sachs, JPMorgan and Morgan Stanley are expected to lead the deal.

A pending lawsuit, with financial risk measured in the hundreds of billions, could have weighed heavily on the S-1 prospectus and spooked investors during the roadshow. Closing the case removes a major risk factor — without guaranteeing the timeline or the final IPO valuation. For European investors, remember that no direct access through a PEA will be possible: a Nasdaq listing would only be accessible through a standard brokerage account or ETFs exposed to US tech stocks.

Sources:

TSMC will raise prices by 10% in 2027: who’s going to pay the AI bill

The world’s largest chipmaker is preparing another price increase for 2027. It’s a show of strength that reassures shareholders, but it will also push up the cost of everything that contains a chip.

Key takeaways:

  • TSMC plans to raise chipmaking prices by up to 10% from 2027 to offset rising costs for materials, manufacturing equipment and overseas plant construction, with base increases ranging from 5% to 10% depending on the customer and the product.
  • Production on mature nodes (12 nm, 16 nm and 28 nm) faces increases of up to 10%.
  • Negotiations began in June and wrapped up in July, with the new price list set to take effect in early 2027.
  • TSMC doesn’t officially comment on its pricing, but a spokesperson said that “our pricing strategy is strategic, not opportunistic” and that the company will keep working with its customers to deliver its value.

Pricing power that reflects a dominant position

This isn’t the first price hike in TSMC’s history, but the context changes everything. Second-quarter revenue hit $40.2 billion, above forecasts, with a gross margin of 67.7%, and the company raised its 2026 revenue growth outlook to just over 40%. In other words: TSMC is raising prices while already posting record margins. That’s the very definition of pricing power — the ability to pass on costs without losing customers, because there’s no credible alternative.

The stated reason is familiar: TSMC and other foundries are facing soaring costs for a wide range of production inputs, including materials, equipment and energy. On top of that comes the cost of plants outside Taiwan. The company announced an additional $100 billion investment in its Arizona operations, bringing its total US commitment to $265 billion. Manufacturing on American soil is structurally more expensive than in Taiwan — and part of that bill is passed on to customers through the new price list.

A sign that analysts are taking the move seriously: a firm like Wedbush sees it as a positive indicator for the entire sector. According to Wedbush, the development could signal improving conditions in the mature-node market and support stronger pricing trends across the industry, which could benefit UMC, GlobalFoundries and Tower Semiconductor.

What it means for a European investor

TSMC is listed in Taiwan and through an ADR in New York (TSM), not on a European exchange — so the stock isn’t PEA-eligible and has to be bought through a standard brokerage account. For investors who want to stay within a PEA, exposure mostly comes through UCITS semiconductor ETFs, where TSMC is often among the top holdings.

The economic message goes beyond TSMC stock alone: a price increase at the foundry ripples through the whole chain, from high-end AI chips to the mature components found in cars, TVs and connected devices. Mature-node chips are the workhorses of routers, TVs, cars, home appliances and mid-range phones; a TSMC price hike flows through contracts and component suppliers, forcing manufacturers to raise prices, cut features or accept thinner margins. So this isn’t just a stock market story: it’s a leading indicator of tech inflation that will eventually hit everyday products.

Note: at this stage, this is a press report (Nikkei Asia, citing multiple sources), not officially confirmed by TSMC, which never comments on its pricing policy.

Sources:

SpaceX: Starship delayed, $109 billion lock-up looming, first earnings on August 4

In 48 hours, SpaceX racked up the bad news: another Starship delay, confirmation of a dizzying lock-up expiration and an earnings date set for August 4 — while the stock, now back below its IPO price, is heading all the way to the Singapore Exchange.

Key takeaways:

  • Last Friday, SpaceX said it would try again to launch on July 20, before another delay to Thursday, according to Boursorama.
  • SpaceX went public on June 12 at $135, with the stock climbing as high as $225.64 before closing Monday at $119.85 — more than 45% below its post-IPO peak.
  • Under the lock-up terms, 20% of shares held by eligible employees and investors (up to 911.5 million shares) are released two trading days after the first earnings report, an unconditional tranche representing an unprecedented flood of shares onto the market.
  • SpaceX will report quarterly results on August 4 after the close, its first since going public on Wall Street, while the Singapore Exchange announced it will introduce depositary receipts for SpaceX, Grab and Sea.

Starship: a technical hiccup that weighs heavily on the stock

On July 16, SpaceX’s Starship rocket triggered a last-minute abort before liftoff on its 13th test flight from Texas, wiping roughly $100 billion off the company’s market value. The abort came after engine ignition, when four of the Super Heavy booster’s 33 engines failed to activate. The link between tech and the stock market is direct here: every delay of this launcher, pitched as the cornerstone of SpaceX’s expansion into Starlink and future AI infrastructure, immediately translates into tens of billions in vanished market value. The slide was attributed, to some extent, to further launch delays of its Starship rocket, but mostly to speculative moves and profit-taking, according to Belga dispatches picked up by the Belgian press.

The real test: lock-up and August 4 earnings

It’s the combination of the two events that worries analysts. Those shares are currently worth about $123 billion, more than the $86 billion in shares currently available for public trading. An additional tranche of 455.8 million shares could become eligible for sale if the stock trades above $175.50 for at least five of the ten trading days before the earnings announcement — a threshold currently out of reach given where the stock is trading. Jay Hatfield, CEO of Infrastructure Capital Advisors, flagged that risk last week: “We won’t overweight the stock because they have that lock-up coming.” For a European investor, access remains indirect: no Euronext listing, but depositary receipts that now extend all the way to Singapore, alongside Grab and Sea, outside the standard PEA/UCITS framework. So mark your calendar for August 4, a date that brings together both SpaceX’s first real financial report as a public company and the trigger for the biggest share unlock in its trading history.

Sources:

AMD lands at Microsoft Azure with Helios: an alternative to Nvidia takes shape

On July 20, Microsoft made official an expanded partnership with AMD to deploy Helios, its new “rack-scale” AI system, on Azure. It’s a concrete sign that hyperscalers want to diversify their AI chip suppliers, away from Nvidia’s near-monopoly.

Key takeaways:

  • Microsoft will deploy AMD’s Helios Rackscale solution to power frontier model inference, its AI customers and Azure AI services.
  • Helios is a complete rack system combining AMD’s next-generation Instinct GPUs with 6th-generation EPYC CPUs, code-named Venice, plus Pensando networking and ROCm software.
  • AMD stock closed up about 5% on the news, while Microsoft shares added 0.6% according to Yahoo Finance (other outlets cite gains of up to +2.1% for MSFT depending on when it was measured).
  • Microsoft thus joins other early Helios customers such as Meta, OpenAI and Oracle.

What Helios really changes

This isn’t just one more chip. The key point: Microsoft isn’t buying a faster chip, it’s buying a credible second rack. The unit of competition in AI infrastructure has shifted from the component to the integrated system. Concretely, Helios packs 72 Instinct MI455X GPUs, 6th-generation EPYC processors and Pensando processing units into a liquid-cooled open rack design. In terms of capability, a Helios rack can deliver up to 1.4 exaFLOPS of FP8 compute and 2.9 exaFLOPS in FP4, with 31 terabytes of onboard HBM4 memory. The racks will be available through new Azure VMs, including one dedicated specifically to large-scale inference, another to AI data preparation and a third to high-performance computing. AMD will begin shipping Helios to customers, including Microsoft, in the second half of 2026.

One big unknown remains: software. Nvidia’s real advantage is still CUDA, a developer ecosystem more than a decade old; ROCm has improved, and Microsoft’s commitment to deploy Helios at scale is a significant signal that the software is finally mature enough to matter. But “close to good enough” and “equivalent at scale” are still two different bars.

The market bridge: diversification, not substitution

For a European investor, the message isn’t “sell Nvidia, buy AMD.” It’s more that the AI value chain is fragmenting: Google is ramping up its own TPU chips, Meta is renting AI chips from Google through a multiyear deal, Amazon is expanding use of its in-house Trainium and Inferentia chips, and Microsoft and Meta are developing their own accelerators. This diversification structurally benefits several players across the chain — foundries, networking equipment makers, HBM memory makers — rather than a single champion. That’s precisely the case for semiconductor or AI infrastructure sector ETFs, available in UCITS format for a brokerage account or certain life insurance contracts, which let you capture this reshuffle without betting on the ultimate winner of the AMD–Nvidia showdown. Note: these sector ETFs remain volatile and concentrated in tech; they don’t replace overall diversification of a PEA portfolio.

What to watch in the coming months: the pace of the production ramp (targeted for early 2027 according to several analysts), the actual scale of Azure deployments beyond the announcement, and AMD’s “Advancing AI” conference scheduled for July 23, which could bring more details.

Sources:

I now have all the information needed to write the article.

Humanoid robotics: Samsung creates a dedicated division, Hyundai deploys Atlas despite a strike

A double play in Korean robotics: Samsung is setting up a new organization reporting directly to its CEO, while Hyundai moves ahead with the industrial rollout of the Atlas robot — against the backdrop of an unprecedented strike over automation.

Key takeaways:

  • Samsung Electronics announced it is making robotics a strategic pillar of future growth and is creating the “RX Business Promotion Office,” reporting directly to the CEO.
  • Samsung appointed Executive Vice President Lee Dong-kun, who previously led robotics strategy at Hyundai Motor Group and Boston Dynamics, to head its robotics strategy team.
  • Thousands of unionized workers at Hyundai’s giant Ulsan complex have already walked off the job across several shifts, and a new round of four-hour strikes began on July 20, sparked by the plan to deploy more than 25,000 Atlas humanoid robots.
  • Right after the walkout, Hyundai announced it was buying SoftBank’s remaining stake in Boston Dynamics, making the robot maker a wholly owned subsidiary.

Samsung organizes its robotics, Hyundai beefs up its industrial alliance

Samsung said the new “Robotics eXperience” (RX) unit will consolidate all of the group’s robotics capabilities under a single organization, responsible for medium- to long-term strategy, key technology development and commercialization. A dedicated data center will also be built at the Gumi site to collect the real-world data robots need before they work on factory floors, with that data then sent back to teams in Seoul to refine intelligence and control. Samsung also plans to open research centers in the United States, China and Japan to secure talent and technologies specific to each region.

A sign the trend goes beyond Samsung alone: LG Electronics had already created a “Robotics Business Center” reporting to its CEO the previous month, up and running since the 1st of the month. The entire Korean electronics industry is gearing up for the physical AI battle, with Samsung’s hiring of an executive from Hyundai/Boston Dynamics symbolizing this competition for talent.

On Hyundai’s side, industrial progress continues despite labor tensions. Atlas deployments are already fully booked for 2026, with fleets slated for Hyundai’s Robotics Metaplant Application Center and for Google DeepMind in the coming months. Hyundai aims to deploy more than 25,000 units across its Hyundai and Kia sites, with a potential start at the Georgia Metaplant factory in 2028, and is targeting annual production capacity of 30,000 Atlas units that same year.

The strike, a real-world test for automation

This is likely the first auto industry strike explicitly triggered by the deployment of humanoid robots, coming after Hyundai announced a plan to deploy more than 25,000 robots worldwide. Notably, the robot at the heart of the dispute, Boston Dynamics’ Atlas, hasn’t yet been assigned to any task in a Korean factory — so the strike is preemptive. According to the Wall Street Journal, the economic impact is already tangible: the slowdown could disrupt production of about 5,000 vehicles and cut Hyundai’s revenue by about $135 million.

For European investors following the theme through UCITS ETFs available in a PEA or brokerage account (such as iShares Automation & Robotics, ISIN IE00BYZK4552), these two moves widen the map of players to watch: Samsung Electronics joins Hyundai, Boston Dynamics and LG Electronics in the Korean humanoid robotics race — a bet on the autonomous factory, but one whose timeline remains subject to labor trade-offs, as the Hyundai case shows.

Sources:

Oracle under pressure: $7 billion in collateral demanded for its AI data centers

After its credit rating downgrade in early July, Oracle now has to deal with a collateral requirement of more than $7 billion imposed by a US energy regulator. A new front is opening in the cloud giant’s credit story, which the markets were already watching closely.

Key takeaways:

  • Wisconsin’s energy regulator is refusing to relax its financial guarantee rules for Oracle’s AI data center in Port Washington, a $15 billion project developed with OpenAI and Vantage.
  • That decision, combined with S&P’s July 9 downgrade (BBB-, one notch above “junk” status), triggers a collateral obligation of more than $7 billion.
  • Oracle stock fell nearly 4% on July 21 on the news, hitting a 52-week low.
  • The stock has lost more than 60% since its September 2025 peak, amid doubts about debt financing of the AI infrastructure race.

What happened in Wisconsin

The crux of the problem is purely contractual. S&P Global Ratings downgraded Oracle to BBB- on July 9, putting it below the A- threshold required by We Energies’ tariff and triggering a collateral requirement of more than $7 billion. Concretely, under We Energies’ tariff rules, data center developers without at least an A- rating from S&P are legally required to secure their projects with cash or letters of credit.

The site in question is no small matter: the project, designed to draw nearly a gigawatt of electricity, is a key part of Oracle’s broader commitment to deliver $300 billion of computing capacity for OpenAI, a goal shared with two development partners. Oracle has even taken the matter to court to challenge the rules, without success so far. An attorney quoted by Wisconsin Public Radio said Oracle’s downgrade shows that utilities’ massive investments to serve data centers need to be backed by the right kinds of collateral.

The immediate market fallout: Oracle stock dropped 3.98% to close at $121.38, after the Financial Times reported that the Wisconsin regulator was sticking with its collateral requirement, which could cost the tech giant up to $7 billion.

A broader credit problem

This case is just a symptom. Oracle is burning cash faster than it generates it, in the middle of a $250 billion data center expansion, which led S&P to cut its rating to BBB-, one notch above junk status; Moody’s has a negative outlook, meaning another downgrade is possible in the medium term. The numbers are staggering: S&P now forecasts capital spending of $90 billion to $95 billion for fiscal 2027, versus a previous estimate of $60 billion, and projects negative free cash flow of $42 billion, versus an earlier projection of negative $24 billion.

For a European investor, access to Oracle remains simple through US or global ETFs eligible for a brokerage account (the stock, however, can’t be held in a standard PEA, which is reserved for European stocks). Above all, this story highlights a structural risk: debt financing of the AI infrastructure race is starting to weigh on balance sheets, even at companies considered solid. No investment decision should be made based solely on this news, which is still developing.

Sources:

Edition prepared by our monitoring agent (data and sources as of Tuesday, July 21, 2026), reviewed according to our editorial policy. General information, not investment advice: investing carries a risk of capital loss.