AI Markets: SpaceX disappoints, Musk drops AMD, Palantir soars

Earnings night in AI markets: SpaceX releases its first results as a public company while Elon Musk picks between Nvidia and AMD, and Palantir electrifies Wall Street with an off-the-charts quarter. Meanwhile, Unitree prices its Shanghai IPO and an earthquake is a reminder of how fragile the chip supply chain is.

SpaceX unveils its first public-company numbers… and sends investors running

Two months after a record IPO, SpaceX released its very first quarterly report as a public company overnight from August 4 to 5. Revenue beat expectations, but the stock fell as much as 8% in after-hours trading.

Key points:

  • Revenue: $7.8 billion in Q2, up 92% year over year, above analyst expectations ($6.8 billion)
  • Net loss: $541 million, half what was expected, but capex nearly doubled to $18.4 billion
  • Elon Musk confirms SpaceX will build its AI exclusively on Nvidia chips, shutting out AMD (whose stock fell the same day)
  • On August 6, the expiration of a lock-up clause is set to make roughly 4 times as many shares tradable as today

Results that beat consensus, but capex that worries

SpaceX has been listed since June 12 (IPO at $135, peaking at $225.64 six days later, for a record market cap of about $2,640 billion). The stock has since lost roughly half its value, and this first quarterly report was supposed to reassure.

On paper, the numbers look good: $7.8 billion in revenue, up 92% year over year, above the expected $6.8 billion. The net loss also narrowed, to $541 million from $1 billion a year earlier.

But Wall Street is looking elsewhere: capex jumped to $18.4 billion for the quarter (versus $10.1 billion the previous quarter), mostly to build AI data centers in Memphis and fund Starship. Only Starlink generates a real operating profit ($1.66 billion, 12 million subscribers, twice as many as a year ago). AI is still weighing on margins: $2.56 billion in revenue (up 247%) but still operating at a loss. Hence the stock’s drop, as much as 8% after hours.

Musk bets on space AI, drops AMD, and takes aim at telecom

On the earnings call, Elon Musk laid out his ambitions: revenue “of $1 trillion” reached as early as 2030, or even 2029 by his account — a target he admits is optimistic, far above the $207 billion analysts project for 2029. He is also targeting 20 gigawatts of computing capacity (“space compute”) by the end of 2027, notably in Memphis, where Anthropic and Google have already reserved capacity.

Another announcement: SpaceX will now build its AI exclusively on Nvidia chips (the “Vera Rubin” architecture), including for onboard computing in orbit (“Starmind”). Nvidia rose on the news; AMD, left out, fell about 8% the same day — a topic covered in detail in another article in this section.

SpaceX also wants to launch a Starlink mobile service competing head-on with AT&T, Verizon and T-Mobile (not just through partnerships, as with T-Mobile today). President Gwynne Shotwell says she’s aiming for “quite a few” of their customers.

What to watch: the lock-up expiration on August 6

Two days after these results, a lock-up clause expires. About 911 million shares held by insiders become tradable, nearly 4 times the current float. According to NBC News, insiders are already preparing to sell — nothing is settled, but this potential wave of supply could weigh on the share price.

For a European investor: SpaceX (SPCX) trades on the Nasdaq and is not eligible for the PEA — it is only available through a standard securities account.

Sources:

AMD breaks its own records… and tanks on the market: the Q2 double surprise

Overnight from August 4 to 5, 2026, AMD reported record quarterly revenue. The immediate result: the stock fell about 8% after hours, a paradox worth unpacking.

Key points:

  • Q2 2026 revenue: $11.5 billion, up 50% year over year, an all-time record for AMD.
  • Data center (EPYC and Instinct chips for AI) is booming: $6.7 billion, up 107% year over year, or 58% of total revenue.
  • Adjusted earnings per share of $1.66 (up 246% year over year); Q3 guidance set at about $13 billion, above expectations.
  • Despite this double “beat,” the stock fell about 7 to 9% depending on the source, dragged down by investments seen as too high and by confirmation that Elon Musk is choosing Nvidia over AMD for SpaceX.

A quarter that should have reassured

The numbers speak for themselves. AMD is selling more EPYC chips (traditional servers) and, above all, more Instinct chips, built for AI. The data center segment more than doubled in a year, with operating income swinging from -$155 million to +$2.1 billion.

This quarter confirms the momentum that kicked off in late July 2026: the launch of the Helios system (72 Instinct MI455X GPUs per rack), with Microsoft Azure as its first customer, followed by a deal with Anthropic for up to 2 GW of MI450 chips (with AMD investing up to $5 billion in the AI startup). At the time, AMD claimed 8 of the world’s 10 largest AI companies as customers.

For Q3 2026, AMD is targeting about $13 billion in revenue, up 41% year over year, guidance above analyst consensus.

So why is the stock falling?

Two reasons are at play, and they need to be kept apart.

First, the investment bill. AMD spent $808 million on equipment during the quarter, about three times what analysts expected. The direct consequence: gross margin came in at 54%, below the 56% consensus. The market fears that ramping up Helios will cost more than expected before it pays off.

Second, a symbolic blow. On his own earnings call the same evening, Elon Musk confirmed that SpaceX will build its AI infrastructure “exclusively” with Nvidia chips (the Vera Rubin platform), including for onboard computing in space via future Starmind satellites. SpaceX wasn’t among AMD’s biggest customers, but the message to investors is clear: even a partner close to the AI ecosystem picks Nvidia by default. It doesn’t call into question the contracts AMD has already signed (Microsoft, Anthropic), but it’s a reminder of the current balance of power between the two AI chip rivals.

For a European investor: AMD (Nasdaq: AMD) is not eligible for the PEA. The stock can be bought through a standard securities account, or indirectly through a PEA-eligible UCITS US equities/tech ETF. This quarter illustrates a classic phenomenon in the AI sector: a record “beat” is no longer enough if guidance or spending disappoints already sky-high expectations.

Sources:

Palantir: an “otherworldly” quarter sends the stock up 30%

On August 3, 2026, Palantir reported quarterly results well above expectations, driven by sales of its AI to businesses and the US military. The stock first climbed after hours, dipped slightly overnight, then exploded 30% during the next day’s session.

Key points:

  • Q2 2026 revenue: $1.94 billion (up 93% year over year), well above consensus (~$1.81 billion).
  • Earnings per share of $0.41, versus the $0.35 analysts expected.
  • US commercial revenue up 149% to $764 million; US government revenue up 90% to $809 million.
  • 2026 guidance raised to $8.15-8.16 billion in revenue (up 82% year over year), with an adjusted operating margin of 62%.

AIP: how Palantir turns AI into revenue

AIP, Palantir’s artificial intelligence platform, connects a company’s data to AI models that can automate concrete decisions (logistics, finance, operations) without rebuilding the entire existing IT system. This quarter confirms its adoption is accelerating: US commercial revenue jumped 149% in a year, proof that businesses are already paying for applied AI, not just experiments. On the defense side, the Maven program with the Pentagon keeps expanding, with more than 25,000 users now building applications on the platform. CEO Alex Karp called the quarter “otherworldly,” pointing to “AI sovereignty” as a driver of demand.

Market timeline: spike, pullback, then a 30% rally

The results came out Monday, August 3, after Wall Street closed. The stock first rose nearly 15% after hours, before slipping about 2% later in the night — hence headlines describing a stock “retreating after hours.” But from the opening bell on Tuesday, August 4, the move reversed: the stock closed up nearly 30% ($162.66), its best session since February 2024. The rally cost short sellers dearly, who had built up about $2.7 billion in paper gains before the announcement: they took about $3 billion in losses in a single session, according to Bloomberg. Several analysts (UBS, Goldman Sachs, Citi) raised their price targets, citing the absence of any visible “competitive pressure” from generic AI.

A clear winner from AI “monetization,” without settling the debate

Palantir stands out as one of the few AI companies that are already highly profitable (adjusted operating margin of 62%), in a market that is more broadly questioning the real return on investment of enterprise AI. For a European investor, Palantir remains outside the PEA: it’s a US stock listed on the Nasdaq, available only through a standard securities account or a UCITS-eligible US equities/technology ETF.

Sources:

Unitree prices its Shanghai IPO: two valuations, one story to untangle

On Wednesday, August 5, 2026, Unitree Robotics entered the pricing phase of its IPO on Shanghai’s STAR Market. Two valuation figures have been circulating in the press in recent days — we checked them, and they don’t tell the same story.

Key points:

  • Timeline: pricing on August 5, subscriptions August 8-10, listing expected around August 19, 2026.
  • Target raise: 4.2 billion yuan (about $610-620 million), through the issuance of 40.45 million shares, or 10% of post-IPO capital.
  • Implied IPO valuation: about 42 billion yuan (~$5.9 to $6.2 billion) — this is the figure reflected in Global Times’ “$5.9 billion,” expressed in dollars.
  • Different figure, different nature: CITIC Securities (the lead underwriter) expects a valuation of 50.6 to 55.9 billion yuan (~$7.4 to $7.9 billion) within 6 to 12 months of listing — this is what comes closest to the “56 billion” cited by Zonebourse.

Why two different figures aren’t a mistake

The IPO price (42 billion yuan) sets what the shares are worth at the moment they are sold to investors. CITIC’s range (50.6-55.9 billion) is a secondary-market forecast, an estimate of what the stock could be worth once listed, put forward by the same bank running the deal — so it should be read with caution, like any projection from a player with a stake in the IPO’s success. Some analysts even float, more speculatively, a market cap topping 100 billion yuan (~$14.8 billion) from the first trades. These are bets, not facts.

Slowing growth, intensifying competition

Revenue for the first half of 2026 is up between 35.6% and 45.4% year over year — a sharp slowdown from last year’s 332.6% — and adjusted net income is reportedly even down, weighed down by rising R&D spending. If the deal goes through, Unitree would become the first pure-play humanoid/quadruped robot company listed on a major global exchange: neither Boston Dynamics (a SoftBank subsidiary) nor Figure AI (still private, valued at $39 billion with no revenue) has crossed that threshold. In China, the competition is heating up in parallel: BYD is unveiling its “Xiao Di” robot, Xpeng is preparing 50,000 units a year of its “Iron,” and Zhiyuan Robotics (AgiBot) forms a still-unstable duopoly with Unitree.

For a European investor: Unitree shares, listed in Shanghai as A-shares, are not available through a PEA, nor easily through a standard securities account. Indirect exposure goes through thematic robotics/AI UCITS ETFs (such as KOID), available from Paris.

Sources:

Kumamoto earthquake: TSMC turns the page, but not completely

On July 28, 2026, a magnitude 7.1 earthquake struck southern Japan near Kumamoto, disrupting TSMC’s JASM plant as well as nearby Tokyo Electron and Sony sites. A week later, TSMC announced a return to normal — a real-world stress test of the global chip supply chain’s resilience.

Key points:

  • On August 3, 2026, TSMC announced that production had returned to normal at its JASM plant in Kumamoto (Plant 1), after a structural inspection and recalibration of its machines; a “full recovery” is still to come, however.
  • This plant makes mature chips (12/16/22/28nm) for Sony (image sensors) and the auto industry — not the most advanced AI chips, which are still made in Taiwan.
  • The recovery was much faster than in 2016 (months of disruption back then), thanks to reinforced buildings and better business continuity plans — but Tokyo Electron, a key equipment supplier, worried markets more than TSMC itself.

A plant for “ordinary” chips, not the brains of AI

TSMC’s JASM plant in Kumamoto doesn’t make the company’s most advanced AI chips: those are still produced in Taiwan, at the most modern sites. It produces mature nodes (12 to 28 nanometers), used for image sensors (main customer: Sony) and automotive components (Denso). It accounts for less than 3% of TSMC’s total capacity. So the earthquake didn’t threaten production of the GPUs or AI accelerators that drive the group’s market valuation — even if the episode revives the debate over the world’s dependence on a handful of Asian sites.

The real signal to watch isn’t necessarily TSMC

The market reacted: TSMC’s ADR lost as much as 3.5% in premarket trading, and the US semiconductor index (Philadelphia SOX) fell 3.7%. But Tokyo Electron took the biggest hit, dropping 6.5% — the stock dragged the Nikkei down nearly 560 points. The reason: its Kumamoto site supplies about 91% of the world’s “coater/developer” equipment, a critical building block for virtually every foundry on the planet. A more concentrated choke point, and therefore a riskier one, than TSMC’s plant itself. A second JASM plant, dedicated to 3 nanometers and under construction on the same site, was not affected — TSMC is pressing ahead with geographic diversification, but this earthquake is a reminder that it doesn’t eliminate every risk: on top of Taiwan’s geopolitical risk, there is now seismic risk, including in Japan.

For a European investor: TSMC, listed in Taiwan and as an ADR in New York, is not eligible for the PEA. It remains available through a standard securities account or technology/semiconductor equity ETFs in UCITS format.

Sources:

Edition prepared by our monitoring agent (data and sources as of Wednesday, August 5, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing involves a risk of capital loss.