Wall Street banks gush over SpaceX but stock won’t fly
Plus: Chip darlings crack as AI doubts bite; DeepSeek seen developing its own AI chip: Reuters; Anthropic finds hidden reasoning space inside Claude.
Good morning. Here’s what happened overnight and what you need to know today.
1.
Surprise, surprise: SpaceX drew bullish coverage from more than a dozen Wall Street brokerages as it entered the Nasdaq-100 index less than a month after its debut. The first attempts to value the rocket, satellite and AI company using conventional metrics rather than faith in Musk’s long-term vision follows the end of the quiet period for banks that underwrote the company’s IPO last month. Morgan Stanley set a USD300 ($432.6) price target with an overweight rating, while Deutsche Bank, Bank of America, Goldman Sachs, JPMorgan, UBS, Citi and others issued buy or equivalent ratings. Retail-focused brokerage firm Raymond James set the Wall Street-high target of USD800, which Bloomberg noted would value SpaceX at roughly USD10.5 trillion, more than double Nvidia. Most pointed to Starship and the AI business as key growth drivers. Independent research house CFRA was the only one with a sell rating and a Street-low USD115 target, Reuters reported. Index entry could draw USD4.3 billion in passive inflows, JPMorgan estimated. Yet that forced buying has not lifted the stock, which is slightly down since its 12 June debut (down 6.8% overnight). Harvard researcher Marco Sammon suggests hedge funds and large investors who got in early on SpaceX to quickly flip their shares may be supplying stock into the index funds. (Reuters)(FT)(Bloomberg)(Yahoo Finance)
2.
Priced for perfection: Wall Street’s chipmakers sold off sharply overnight as investors questioned whether this year’s vast AI spending will justify high valuations, with a record profit from Samsung failing to satisfy the market. The Philadelphia semiconductor index fell 4.65%, trimming its 2026 gain to around 74%, and the Nasdaq 100 dropped 1.16%. Samsung projected a near 19-fold surge in quarterly earnings but its results topped estimates without clearing elevated buy-side expectations. That seemed to spark a global chip selloff that began in Seoul, where the Kospi lost nearly 5%. Micron fell 4.7% and Sandisk 7.3%. Adding to the pressure, Reuters reported Chinese startup DeepSeek is developing its own AI chip. Most S&P 500 companies still rose, with advancers outnumbering decliners, signalling a rotation into other sectors. Separately, oil jumped as much as 5% to above USD72 after the US revoked a waiver allowing Iranian oil sales, following attacks on ships near the Strait of Hormuz. That also lifted the 10-year Treasury yield 7 basis points to 4.54%. (WSJ)(Reuters)(Bloomberg)
3.
Chip off: Reuters reported citing three unnamed sources that Chinese startup DeepSeek is developing its own AI chip, in a push that could reduce its reliance on Nvidia and Huawei chips. The chip is designed for inference rather than for training new models, according to the report. If successful, the move would mark a major strategic shift for a company widely hailed in China as the country’s AI champion, and could add to challenges facing Huawei. The effort is at an early stage, with DeepSeek in discussions with chip-design, foundry and memory companies and increasing its hiring of chip-design engineers. Designing a competitive chip would take years and a lot of capital, and US curbs cut Chinese designers off from the most advanced foundries and from high-bandwidth memory critical to inference chips, Reuters noted. The scale of the players is also notable. DeepSeek’s most recent funding round would value it at up to USD59 billion, against Nvidia’s market value of about USD4.7 trillion. Separately, AI startup Perplexity said it plans to use Nvidia’s new “Vera” CPU, a more general-purpose chip the company is pushing to diversify beyond its AI processors as customers like OpenAI and DeepSeek build their own. (Reuters)
4.
Inner life: Anthropic said it found a small set of previously undetected internal neural patterns inside its Claude AI models that play a special role in how the model reasons. In a paper published earlier this week, the company calls the mechanism the “J-space” after the mathematical technique used to find it, and says each pattern is linked to a word that can “light up” when Claude is thinking “about a concept without writing it down.” The company said it emerged on its own during training rather than being designed. Anthropic also reported that Claude can report on what is in this space, deliberately control it, and use it for multi-step reasoning. When Anthropic blocked it, Claude still spoke fluently and recalled facts but lost higher-order functions like multi-step reasoning. Anthropic framed the work as a potential safety tool, saying it could read hidden thoughts like catching a model noticing it is being tested, fabricating data. In one case, a model deliberately trained to sabotage code showed the words “fake” and “fraud” internally even while producing ordinary-looking output. The company stressed the results do not show Claude is conscious, distinguishing the functional sense of the term from the capacity to have experiences. (Anthropic)(Washington Examiner)(Bloomberg)
5.
Power hungry: Australia’s data centre sector faces a defining test over whether it can absorb Anthropic’s reported demand for 1.4 gigawatts of capacity, which is close to the size of the entire existing industry. The AI lab is reportedly weighing making the country its second home for training its models. The maker of the Claude chatbot confirmed it is in early talks with local operators but was cautious on scale, after The Australian Financial Review reported it was seeking to sign first contracts this month. The industry lobby called the demand a potential turning point but warned the investment may hinge on firmed energy supply, grid connection speeds and planning approvals. And a copyright fight is looming over any local training plans. The government has ruled out weakening protections, with assistant treasurer Andrew Charlton rejecting a text and data-mining exemption and attorney-general Michelle Rowland urging the tech and creative sectors to find a workable deal. The push has split Labor and drawn Greens calls for a moratorium on new hyperscale sites, against community resistance that survey data put at 83% of Australians unwilling to live near a data centre. Anthropic is also planning to expand in New York, with plans to double its workforce in the city to 1,000 people, the NYT reported. (Capital Brief)(NYT)
6.
Second act: French far-right leader Marine Le Pen said she will run for president in 2027, hours after a Paris appeals court upheld her embezzlement conviction but shortened a ban on her holding public office, clearing her path to the ballot. “This evening, I am a candidate in the presidential election,” the National Rally leader told broadcaster TF1. Le Pen, who has lost the presidency twice to Emmanuel Macron but now leads early polling to succeed him, was found to have embezzled European Parliament funds meant for parliamentary assistants to pay her party’s staff. The court sentenced her to three years in prison, two suspended and one under house arrest with an electronic tag, and cut her electoral ban to 45 months, most of it suspended and backdated. That meant the disqualifying portion is already served. Le Pen said she would appeal to France’s highest court, a step she argued suspends the ruling and lets her campaign freely. She would run alongside her 30-year-old protégé Jordan Bardella, who would serve as prime minister. (NYT)(Le Monde)(BBC)
7.
Fake move?: British politician Nigel Farage said he will resign as an MP and stand in the resulting by-election, a move the Reform UK leader cast as “people versus the establishment” but which rival parties refused to contest. Farage, the anti-immigration, Brexit-championing figure whose party leads UK opinion polls, made the announcement on Tuesday amid a scandal over undeclared gifts. He said “the people of Clacton”, his seaside constituency in England’s southeast, “should be the judges of my actions” and insisted he had done “nothing wrong”. The governing Labour party, the opposition Conservatives, the centrist Liberal Democrats and a smaller right-wing rival all said they would not field a candidate, variously branding it a “circus”, a “fake by-election” and a “vanity project”, according to media reports. Farage has been under investigation by parliament’s standards watchdog since May over an undeclared GBP5 million ($9.6 million) gift from a cryptocurrency investor, and faces fresh scrutiny after The Sunday Times reported he took undeclared benefits from a convicted fraudster. (BBC)(FT)
8.
Crowding out effect: In the wake of SpaceX’s mega IPO and ahead of the potential trillion-dollar floats of Anthropic and OpenAI over the coming year or so, fund managers and strategists are urging investors not to just chase headlines but to hunt for the hidden winners across the AI stack. Growing talk of a “crowding out” effect is seen by some as a risk, and by others as an opportunity. ECP Asset Management’s Annabelle Miller told Capital Brief companies directly involved in AI are “acting as a liquidity pull on everything else”, contracting the valuations of everything else and opening a window for “the patient investor” to acquire businesses that might benefit from deploying AI internally. Even within AI, Pengana Capital’s Adam Myers argues there could be multiple winners across the stack, because “we don’t know who is ultimately going to win”. Betashares and BlackRock strategists likewise urge diversification across the value chain, where returns are already diverging. (Capital Brief)
9.
Mail order: Prince Harry and six other high-profile figures lost their privacy case against the Daily Mail’s publisher, after London’s High Court ruled they failed to prove their information was gathered unlawfully. Harry, Elton John, David Furnish, Liz Hurley, Sadie Frost, Doreen Lawrence and former British Liberal Democrat MP Simon Hughes had accused Associated Newspapers of phone hacking, blagging and other unlawful methods across dozens of articles published between the 1990s and 2015. Mr Justice Nicklin dismissed every claim, writing that “suspicion, even understandable suspicion, is not proof” and rejecting attempts to prove the case by broad inference where a lawful source pathway remained realistic. He declined to rule on whether unlawful information gathering was widespread at the Mail, deciding each claim individually. The company said the ruling was “an overwhelming victory for the Daily Mail and its journalists” and that it would seek to recover costs, estimated at more than GBP50 million. (Capital Brief)(Reuters)(Guardian)(BBC)(NYT)
10.
Lacking a stick: The head of one of Australia’s largest mid-tier networks warned the government’s crackdown will fail without multi-million-dollar penalties in line with US standards. After KPMG and EY were hit by scandals in recent weeks, Moore Australia chair David Tomasi told Capital Brief that current measures were not enough of a deterrent. “When it’s to do with professional conduct… that’s where we lack a little bit of a stick here in Australia,” he said, pointing to PwC’s $50,000 penalty over its 2023 tax scandal against the US oversight board’s USD25 million fine on KPMG Netherlands last year. Mid-tier accounting firms like Moore are poised to take market share from the Big Four following the scandals and a government push to direct work away from them, which have also helped firms attract talent and clients. It comes after the government, less than 24 hours after the EY scandal surfaced, released an options paper proposing reforms including mandatory audit firm rotation and separating audit from consulting. (Capital Brief)