Headlines this week - Jun 29, 2025

A look at how capital is being deployed across future opportunities

C Gavilanes

Jun 29, 2025

This week in the future:

1 - Nvidia is on the way to a $4trn valuation, driven by AI labs’ aggressive spend on computing infrastructure

They just surpassed Microsoft as the World’s most valuable company. Nvidia’s market cap has added nearly $1.5trn since April, with a current market cap of about $3.8trn. The company’s growth is driven by robust earnings and strong demand for its AI chips, specially at the company’s biggest customers (large AI labs).

The company is now close to be the first one reaching a $4trn valuation. Some bullish analysts are predicting a $6trn market cap, justified by Nvidia’s unique position and high pricing power in a product with explosive demand (AI chips)

At the base of this success is “Huang’s Law”, a performance improvement trajectory for GPUs, which has enabled the acceleration of AI computing. This is a new scaling law that specifically applies to AI workloads. To some extent, it has replaced “Moore’s Law”, linked to general purpose processors (CPUs), and that seems to have slowed down recently. This week a SemiAnalysis post discussed how Nvidia has made all this happen

Is the cloud the next step for the company? This week the WSJ presents Nvidia as a potential threat to cloud hyper-scalers. The company , which already has a high bargaining power as a supplier, is now also expanding fast into its buyers’ business, with an AI-cloud computing service (DGX Cloud) that they launched 2 years ago.
- This resonates with the current emergence of AI-cloud specialists (like Coreweave and others)
- It can also be seen as a response to big AI labs’ designing their own chips (in detriment to Nvidia’s business). Indeed, this week we had news about Alphabet selling its chips (TPUs) to OpenAI, in what would be a much more direct threat on Nvidia. So the market could have started to move towards vertical integration

2 - Meta signals that talent is now a key resource for AI labs to build a competitive advantage

Zuckerberg views AI talent acquisition as an “existential” question. As we already said last week, he’s personally leading the “recruitment blitz”, offering packages of up to $100m to leading AI researchers. Money would be a way to offset their reluctance to join Meta, given what they perceive as a lack of clarity in the company’s AI strategy

He’s been compiling a list of top AI engineers to target. According to the WSJ, Zuckerberg has spent months building a list of the most talented engineers and researchers in AI (now known as “The List” across Silicon Valley). Now he would be targeting the people in this list with aggressive hiring offers

They have just hired 3 top researchers from OpenAI. The plan is producing its first fruits, as shown by the news this week that Meta has hired 3 top researchers from OpenAI’s Zurich office. They will now start working in the new “Superintelligence” unit that Meta is setting up (apparently under the direction of A Wang from Scale AI)

They’re also using M&A to capture AI talent. According to Bloomberg, Meta is in advanced talks to acquire PlayAI, a startup that uses AI to replicate voices, as part of its push to acquire top talent and catch up in the AI race. The deal would include the acquisition of the company’s technology and some of its employees

3 - Tesla has launched its first robotaxi operation. Should Alphabet’s shareholders be happy?

Last Sunday they launched a robotaxi service in Austin, TX. They are starting small, with only 10 cars and with an explicit focus on safety, apparently targeting the growing regulatory scrutiny on this technology. Related to this, all the taxis have a human “safety driver” on board (but sitting in the passenger side)

The cars have the same setup as the Model Y vehicles currently available for purchase. So if this goes well, it could be a big positive for Tesla’s sales of cars to consumers. Maybe because of this, E Musk has forecasted that autonomous vehicles could add $5-10trn to Tesla’s market cap, which is currently around $1trn

Waymo is the leading robotaxi player in the US, but this is poorly reflected in Alphabet’s valuation. Although Waymo is clearly ahead of Tesla in terms of commercial launches (they are already operating in 5 US cities and their cars have driven more than 71m miles without a human at the wheel), the company’s valuation (around $45bn, so just a small part of Alphabet’s $2trn market cap for now) is way less exuberant than Tesla’s

So some analysts view this as an opportunity for Alphabet. According to these analysis, autonomous technology would already be driving most of Tesla’s $1trn valuation, so there would be a massive value creation opportunity for Alphabet if Tesla’s multiples would be applied to their Waymo subsidiary

4 - Xiaomi (the smartphone company) heats up the Chinese electric car market.

200,000 pre-orders in 3min, taking the company’s stock to an all-time high. Last Thursday, the company’s new YU7 sport utility vehicle, seen as a direct competitor to Tesla’s Model Y, received almost 300,000 pre-orders in the first hour, but two thirds of them happened in the first 3 minutes, according to the FT. The company is using similar marketing tactics to the very successful ones they used when they entered the smartphone market, and this partially explains these results. But analysts expected just around 100,000 pre-orders in the first 24h. The YU7 is sold at a relatively low price ($35,370) so this is seen as increased pressure for Tesla in China

5 - Nuclear energy deployments accelerating in the US

The State of New York wants to build a new 1GW nuclear plant, to meet demand for clean energy. The Governor of the state has directed the NY Power Authority to build the facility, which would have enough capacity to address the needs of 1m homes. The initiative is justified by the growing need for electricity to power data centers, factories, and increasingly electrified homes

Meanwhile, the project to restart the Three Mile Island plant (Pennsylvania) is being fast-tracked. The delivery date has been moved to 1 year earlier than initially scheduled (2027 vs. 2028). The project is led by Constellation Energy, the leading company for nuclear energy production in the US, and the plan is to deliver the power from the plant to Microsoft, to run AI data centers. Constellation has also received the support of the NY State Government to start a new project at one of the company’s 3 existing plants in the State

6 - Court rulings in the US about use of copyrighted material to train AI models favor AI labs, but tension continues

A judge ruled Anthropic’s AI training using purchased books is legal. The court’s rationale was to compare the use of books for training models to the use of these same books by humans, to learn. So the conclusion was that Android’s activity was legal as long as they had previously purchased the books they were using

Meta’s use of millions of books to train its AI models has been judged “fair” by a federal court. In this case, the judge warned that the ruling did not stand that Meta’s use of books is “lawful”, and that it instead reflected the fact that authors and their lawyers had “made the wrong arguments” and failed to properly make their case

However, controversy continues. Particularly in the case of Meta, which is accused of making the models “memorize” the data in the books, rather than learn from them (this could possibly be even against the arguments in the Anthropic ruling). This is shown in a paper by a team of researchers from Stanford, Cornell and West Virginia

Martin Wolf at the FT compares this with a similar debate in the 19th Century. For much of the century, the US did not recognize international copyright in its domestic law, apparently to stimulate economic activity around books. But this was eventually solved with an approach that balanced this with incentives for creators

7 - Concern growing at local communities about AI data centers potentially affect their energy supply

A large Meta data center in Louisiana might raise energy costs for local homes. Meta is building a 4m square feet data center (the size of Manhattan…) in Louisiana, with a power consumption of more than 2GW. At the same time, they are subsidizing 3 gas power plants and a transmission line to feed their energy to the computing facility. Entergy, the local utility, will need additional cash to fund these deployments, and some analysts see a big risk that this could lead to a raise in local homes’ energy bills

Could we better manage demand? And in an article this week, Joanna Stern from the WSJ suggests that transparency about how much energy do different end user prompts consume could lead to an optimization of demand. But identifying the needs of each activity is difficult.

Understanding the energy needs of each online activity could help. In a separate video, the author makes some funny comparison between different data center tasks and times dedicated to grilling a steak (e.g. 2 video generations are equivalent to 1 medium steak; and the short video she produced with Google Veo in May -using 1,000 clips- would equal grilling 478 steaks, or powering an average US home for 3.5 days)

8 - Can an AI agent run a business? Anthropic is testing that, and they’re optimistic

After using Claude to run a vending machine, they think AI agents will be able to run things in the real economy. A team at Anthropic used their Claude Sonnet 3.7 model to operate a small, automated vending machine at their office. The test let the model take control of tasks linked to running a profitable shop, e.g. maintaining inventory, setting prices, or avoiding bankruptcy. Even if the agent failed at some things, the team is optimistic about “additional scaffolding” being able to improve performance. They are already planning the next phase, looking to “anticipate the features and challenges of an economy increasingly suffused with AI.”

9 - The UK just launched a public initiative to develop quantum computing (mostly linked to option value)

They have committed a £500m investment during the next 4 years. Most of this will be dedicated to funding the National Quantum Computing Centre, launched last year, which aims to “push the boundaries of what is possible with the technology”. Local experts point to an “advantage” that the UK would have today in this field, that would be at risk of being lost without the government’s support. As expected, the consensus is that there is an “option value” in these investments, rather than an actual value to be delivered before (most likely) the mid 2030s

10 - Blue Origin would be trying to capitalize the Musk-Trump split, to catch up with SpaceX

Blue Origin is increasing the efforts to get government contracts for space missions. According to the WSJ, this would a consequence of the recent divergences between E Musk and D Trump, that would have weakened the position of SpaceX for these deals. Jeff Bezos has been working to improve his relationship with Trump, including inviting him to his wedding