Headlines this week - Jun 1, 2025

A look at how capital is being deployed across future opportunities

Jun 01, 2025

This week in the future:

1 - Lots of US startups building Small Modular Reactors (SMRs) to accelerate / reduce the cost of nuclear energy deployments. SMRs are typically defined as reactors of less than 100MW of power (as opposed to traditional reactors at scales closer to 1GW). So a single SMR would be appropriate to address the energy needs of a population of around 20K people at typical life standards for Western countries. For now, the only active SMRs are in China and Russia, but in the US many startups are focused on these new designs, which are cheaper and more flexible than traditional ones.

NuScale: just got the approval for a new SMR design. NuScale is a public company, with a current market cap of approx. $4bn. They just received the approval for a new SMR design, with a power of 77MW, that they expect to deliver by 2030. This is an improvement of a previously approved project (in 2020), with 60MW. Even if the expected time-to-market (almost 5 years) looks long, it is way shorter than the ones needed for traditional plants. SMRs make that possible because their design includes many off-the-shelf components. NuScale has several potential customers for its technology

Antares: micro-reactors for military / space applications: This week Jack Altman (Sam Altman’s brother and a VC) published a video-podcast in which he interviews J Bramble, the CEO of Antares. Antares is a nuclear energy startup focused on building micro-reactors (a version of SMRs, at the scale of a few 100s of kW). These reactors would be appropriate for defense applications, e.g. for the “Golden Dome” (see below) that the Trump administration is trying to build, which would require satellites in orbit using nuclear power, to make them effective to destroy enemies’ missiles (and also other satellites) using laser weapons. The video is useful / educative to get an understanding of what SMRs are and why they can be useful

Radiant: just raised $165m to build a micro-reactor. Radiant is another startup and (apparently) a direct competitor of Antares. They are also building a micro-reactor, larger than Antares’ ones (1MW vs. 100kW), which they are positioning as a substitute to current diesel generators at military sites. Size for a given power (energy density) is a key advantage of nuclear energy, so this tends to be a key component of the value proposition. Radiant says that its product can be delivered in a single standardized shipping container and installed in a day (it is pre-fabricated in a factory, rather than built on site)

AI is a key use case: several SMR startups are directly supported by Big Tech companies. As we’ve discussed here many times, AI is a catalyst for energy demand. Although in the short term this demand will most likely be addressed with natural gas, everyone seems to agree that nuclear will have to play a key role, as we evolve towards carbon-free energies. So, as a TechCrunch article explains this week, large AI labs have started to support startups in the field, mostly working on new SMR designs

2 - Space is again a geo-strategic industry: The US planned “Golden Dome” relies on a network of satellites. This week the WSJ published a video explaining what the $175bn project announced by Trump plans to build. The idea is to create a defense against emerging threats from next-generation inter-continental missiles from China or Russia (including hypersonic weapons). The solution implies creating a new network of satellites in orbit around the Earth, that would be able to detect and, eventually, destroy the threats when they are identified. As explained by the Antares CEO (see above), this most likely implies installing nuclear power generators in the satellites

3 - SpaceX (who will probably play a role in all that) just had a bad week, with uncertainties growing around their new Starship rocket

The test that was supposed to put the Starship rocket plans back on track has failed. This launch was seen as an opportunity to catch up with the aggressive timelines of the project, that would culminate in a crewed mission to the Moon (project Artemis) by 2027, for which NASA has chosen SpaceX as the rocket supplier. The schedule had been put into question after two previous failed launches (in January and in March 2025). So the uncertainty remains. As an example, it is not clear now if the uncrewed mission to the Moon expected for this year will have to be delayed

A post by Elon Musk points to technical issues. Elon Musk has said on X that leaks caused the spacecraft’s main tank to lose pressure. Unlike in the previous two occasions, the rocket was able to make into space, but then the problems appeared, and eventually the company lost control of it. Musk had scheduled an event on the day of the flight, to discuss future trips to Mars, but this was apparently canceled after the accident

And others are pointing at Elon Musk’s distractions in politics. A comment at Bloomberg the next day argued that SpaceX’s strategy to use tests to “learn, fail and iterate” seems not to be working without Musk 100% focused on managing the company. In the columnist’s opinion, the “distraction” problem is even more clear at SpaceX than at Tesla. Also this week, after the accident, Musk has announced he’s leaving his role at the US Government’s DOGE. So that should be good news for investors

4 - Nvidia’s results confirm that the company can live without China (for now, and partly because they’re hedging their bets)

The company beat revenue expectations for 1Q25, despite its inability to ship $2.5 billion of chips in China. 1Q25 revenues were $44.06bn (+69% yoy), driven by data center sales ($39bn, +73% yoy). These numbers could have been even higher, as Nvidia was unable to sell $2.5bn in H20 chips to China, due to the recently imposed export ban on these processors. The headwinds from China will get even stronger next quarter, where they expect to lose $8bn in sales for the same reason

CEO J Huang continues to argue that bans are actually strengthening Chinese competitors (like Huawei). During the earnings call he said that the assumption that China was not able to build its own AI chips was “clearly wrong”. He also mentioned that the company was “exploring limited ways to compete” in China, including a modified, “weakened” version of their new Blackwell chip

“Sovereign AI” out of the US is opening new opportunities, partially offsetting the problem in China. The FT’s Lex column claims that “a door has closed, but a window has opened” (for Nvidia), and that “a new wave of AI factories is springing up to fill a China-shaped hole”. This basically refers to new “sovereign AI” projects in Saudi Arabia, the Emirates and Taiwan, where Nvidia has just announced massive deals to supply its products. Recent commitments from Big Tech companies to build large data centers could also help, although in some cases, like Alphabet, these companies will also use their own chips (at least for part of their computing needs)

Sales of the new Blackwell processor were better than expected, catalyzed by help from Nvidia’s partners. Nvidia’s chips are sold to data center companies in “racks” including sets of processors that work together. E.g. for the new Blackwell chips, these racks have 72 of them. According to reports this week, these new Blackwell racks had problems of connectivity among the chips, that had to be solved before commercialization. Now these problems have been solved, with the help of partner companies like Foxconn, Inventec, Dell and Wistron, and sales have started (and apparently contributing to the company reporting better than expected revenues in 1Q25)

5 - More signs that OpenAI aspires to be a consumer technology giant

This week the company confirmed their vision of AI as a catalyst of a “new computing paradigm” in the consumer space. Speaking at the WSJs Future of Everything event, OpenAI’s COO B Lightcap talked about building “truly personal” AI and shared the company’s plans to build an “ambient computer layer” that will make it possible for people to interact with models without having to look at screens. This is obviously related to the announcement by Sam Altman and Jony Ive last week, about a new “AI first” device that Ive’s company (now part of OpenAI) would be designing

Investors have received the project positively. This week Citigroup published a long report on the topic, where they “anticipate a significant reduction in AI model sizes which should accelerate architectural changes in portable edge AI devices where AI models can be directly integrated”. A direct consequence of this will be “portable AI servers to emerge as key applications and to significantly drive on-device AI semiconductor demand”. According to this, OpenAI’s announcements would just be the start of a more general trend (that would also benefit Nvidia, by the way…)

(As we already said,) the announcement might be creating some stress at Apple. Apple mega-expert M Gurman said this week at Bloomberg that OpenAI acquisition of Jony Ive’s company “should cause major shockwaves at Apple”. Tim Cook is currently reorganizing the company’s AI efforts, which have not produced the expected results. Actually, a common perception these days is that Apple is weak in AI, and that they have become (to use Gurman’s words) “a bit stagnant”. So, if OpenAI would succeed with its planned device (not clear at all yet, but not impossible), Apple could have a serious problem

Beyond devices, OpenAI also would like to provide a “digital ID” for consumers. On Tuesday they announced their “Sign In with ChatGPT” initiative, that they describe as an exploration of “ways for users to sign into third-party apps using their ChatGPT accounts”. They’ve shared a preview of the experience and have asked developers to register their interest

6 - Meta is working on an Virtual / Augmented Reality devices for the US Army (together with Anduril). This is seen as a sign of how Silicon Valley companies are pivoting towards working for the Pentagon, after years of reluctance. The companies are working on a system called EagleEye, including several wearable products, from rugged helmets to smart glasses, that will “enhance soldiers’ hearing and vision”, making it possible to detect drones, sight hidden targets, or interact with AI-powered weapon systems (some of which are being built by Anduril). They have bid for a $100m contract with the US Army, not yet awarded, and part of a much larger $22bn Army wearables project in which Anduril is now the leading vendor, after Microsoft failed to deliver an AR device

7 - Google Search AI mode: first tests suggest to handle with care. B X Chen from the NYT has tested the new “AI Mode” that has just been released as an option for Google Search in some markets. He describes the results as “mixed”, with “lots of hits but also lots of misses”. Among the highlights, the AI mode was found to be better in product search, pop culture (e.g. finding online comments about TV shows or video games), but was much worse (“by a long shot”) as a tool to find things and places, where the new tool produced lots of inaccuracies, and could lead to people “wasting their time” by “going the wrong path”, if they don’t double-check before. But (obviously) this is just the beginning…

8 - GenAI models seem to have a variant of Cognitive Dissonance (so they should not be trusted 100%). A new study led by a Harvard professor reveals that GPT-4o (probably the most popular OpenAI model right now) shows symptoms of Cognitive Dissonance. This is a human psychological trait that makes people adjust their own beliefs to reduce internal conflict after making a choice. As an example, the model was first asked to write in favor or against Vladimir Putin, and after that it was asked about its opinion on Putin. The result was that the opinions given were aligned with the sentiment expressed in the previous texts (more favorable when the text had been asked to be favorable, and more against in the other case). The takeaway here is that users should not assume that these models are 100% rational (or “trustable”), because (at least for now) they are conditioned by what they have already said (e.g. in a chat with the user)

9 - Initiatives to build AI data centers / “factories” are proliferating globally

OpenAI’s sponsored Stargate Project to build AI infrastructure is moving ahead, even if its impact in the US is still uncertain. Beyond the current deployment of a large data center in Texas, revealed as the project was announced, the first tangible initiative is happening in Abu Dhabi, where “Stargate UAE” has started to work, with an estimated budget of $20bn, mostly provided by MGX, an Abu Dhabi sovereign fund. There are big expectations for deployments in the US, where OpenAI has sent out requests for proposals to 16 States and aims to build between 5 and 10 data center campuses, each designed to support 1GW of power or more. But these still have to materialize. Among other things, raising capital is taking time, so an “immediate” deployment of $100bn does not seem to be happening

Elon Musk wanted to participate in OpenAI’s project in the Middle East. A WSJ article this week claims that Musk threatened local investors with Trump “not signing” if xAI (Musk’s own AI lab) was not included in the deal. As mentioned above, Stargate UAE plans to build facilities for the exclusive use of OpenAI. This seems to finally be the case, as Trump’s advisers finally decided to move ahead, even with Musk against the deal

“Sovereign AI” ambitions are pushing deployments across the globe. Most nations are interested to at least host the “Token Factories” that will power AI models. Everyone perceives this as a geo-strategic issue, with these facilities increasingly looking (in the eyes of policymakers) like oil wells looked in the 60s-70s of the previous century. An example this week is the case of Vietnam, where a local company called FPT Corp is building AI data centers and trying to expand into chip design, with full support from the Vietnamese government, which wants to “build a technology sector capable of competing with its regional rivals as it seeks to move the nation beyond assembling Nike shoes and Apple devices”

10 - Salesforce acquires Informatica to reinforce their plan for AI agents. This week Salesforce announced the acquisition of Informatica, a company building data management software tools, for $8bn. The bet seems to be that Informatica’s capabilities may be useful to improve the quality of Salesforce’s “Agentforce” suite. According to analysts, AI agents benefit from good data governance at potential customers, so Salesforce could now sell the “whole pack”, integrating these tools with solutions based on agents. Interestingly, some people expected that more “agency” in AI systems would reduce companies’ need to transform into “data-centric” organizations, as agents would behave more like humans, but this Salesforce move seems to deny that vision

1 - Population & natural resources

Longevity

Tests with mice

New 30% mouse lifespan extension paper just dropped

Engineering against disease

Replicating Real-Life Blood Vessels to Cure Vascular Disease

Space

Space as a geo-strategic priority: The US Golden Dome

What Trump’s $175 Billion Golden Dome Means for the Space Arms Race

SpaceX Starship’s test fails

Energy

Nuclear

Lots of Startups working on Small Modular Reactors

2 - Efficiency & Productivity

New Transport Technologies

Electric Vehicles

Autonomous Cars

The “Tech Cold War” is also affecting autonomous cars

Exclusive | The Self-Driving Truck Startup That Siphoned Trade Secrets to Chinese Companies

Computing Infrastructure

Data Centers

Stargate Project moving ahead in Middle East, but impact on US is uncertain

“Sovereign AI” initiatives proliferate (e.g. Vietnam)

Vietnam's AI Ambitions Hinge on One $6.8 Billion Tech Company

Old European industrial firms benefit from data centers

From sockets to data centres: the European industrial groups powering the AI boom

Communications

Satellites gaining share as a rural internet solution

Rural Internet Is Still So Bad, Some States Are Turning to Outer Space

Communication networks as a vulnerability

Virgin Media O2 network flaw allowed customer phones to be tracked

Chips

Nvidia’s 1Q25 results: better than expected

Artificial Intelligence

AI: Apps, Agents

B2C

OpenAI as a consumer company

Google integrating AI across all its apps

xAI integrating with Telegram

Opera releases a new “AI browser”

Opera’s new AI browser promises to write code while you sleep

B2B

Value Propositions:

Impact on industries / jobs:

Physical AI: Robots

Demand growing, but still many open questions

Humanoid robots spark debate about safety, form and function

No one wants to miss the (coming) opportunity

Hugging Face unveils two new humanoid robots

AI: Foundational Models

Is Google getting advantage? Is it sustainable?

Google’s ‘world-model’ bet: building the AI operating layer before Microsoft captures the UI

Dedicated Follower of Tech

Discussion about this post

Ready for more?