Bradford Cross, CEO of Alpha City, on why frontier AI is evolving so rapidly with foundation models but less rapidly in other STEM.

There’s too much FUD - "fear, uncertainty and doubt" around AGI and what it means. For complex goals like hiring a new person, the ability to do the whole thing with an agent isn’t really there. It’s going to be a while. The people who are going to be really valuable, have jobs and get paid more are going to be orchestrating everything with agents and foundation models. They’re not going to train people like they were training an AI system. They’re just going to train an AI system. - @bradfordcross on AI agents and the future of work.

"There just wasn’t enough good stuff to really deploy capital into. So we decided, okay, now is the time we should try to build something. I had accumulated enough wisdom and confidence over those first few years from watching everything to have a pretty decent thesis about what I thought we should do. Patri pointed out that one of our LPs was super connected to people in Portuguese-speaking Africa. We started in São Tomé and Príncipe, then Guinea-Bissau, Angola and Cape Verde. We made great progress, got three MOUs done in the first year, and proceeded quickly into land concessions and joint venture agreements." - @bradfordcross, CEO of Alpha City, on how Alpha started.

"The person who’s able to ask questions and set everything up is just going to automate the rest of it themselves. They’re not going to delegate all these smaller tasks to you. If you’re performing like a recruiting coordinator not a recruiter then the real recruiter who knows how to ask the questions and set everything up is going to use AI and not use you. There will be mass job displacement, but we’re a long way from magically telling AI, “Get me a financial controller by the end of June,” and having three meetings with the finalists appear on your calendar." @bradfordcross on how AI will change the nature of work.

The smartest thing I did when I came into the space to bridge myself from the tech world was I didn’t try to start anything right away. I came in humbly. I spent a ton of time following all the projects, talking to the founders, and learning different things about different people’s models, what’s working and what’s not working. We ended up taking a zone law that looks a lot like the historical work on charter cities and the ZEDE stuff, but we evolved it a little bit so that it doesn’t require any constitutional change. Unless the zone law is going, there’s nothing to really innovate on. I spent a huge amount of time with all these different players when I came into the space, so I really knew what I was doing before I pulled the trigger. - @bradfordcross, CEO of Alpha City Inc. on his strategic approach to entering the startup-city space.

Replacing electricians, plumbers, and people who have to be mobile and go around to different places with humanoid robots or other special task-based robots is a long, long way out, in my opinion. But if you have to be told specifically what to do as a task, and your manager has to keep repeating that to you until you get it right, they’re basically training you like you train an AI. Those jobs are going to evaporate quickly. There is going to be mass AI job displacement, but it’s not like the whole field will be out of a job. The nature of the job will change dramatically. In the new world, you’ve got to ask questions. That’s going to be the key. - Bradford Cross, CEO of Alpha City Inc.

“One of the things I learned is that there's a super high activation cost for these projects. The lead time to do all this political work and deal-making and stuff is a couple years. It's a lot of time, energy, and money. And then you have the infrastructure, you have power, you have water, you have roads, you have all this stuff. And so if you're going to invest all this, the more land you have, the better the justification, the better the ROI of all these infrastructure investments. If you size it too small, you end up really kind of hurting yourself a lot because you're putting so much money into activating the project.” @bradfordcross on the economics of large-scale land development projects

“The biggest problem in Pronomos, we’ve had a dearth of deal flow in the space. There’s just been a lot of people that come into it with a very highfalutin type of approach and not really very practical. What we decided was, okay, now is the time. We should try to build something because there’s just not enough good stuff to really deploy capital into. I had accumulated enough wisdom and confidence over those first few years from watching everything to have a pretty decent thesis about what I thought we should do. We started in São Tomé and Príncipe, then Guinea-Bissau, Angola, Cape Verde, all of Portuguese West Africa initially. We made great progress, got on it very quickly, got three MoUs done in the first year and proceeded quickly into land concessions and the joint venture agreements and everything else.” @bradfordcross on how Alpha started.

''For the serious AI training campuses going forwards, we should be looking at 2 gigawatts and up. The Stargate project had more than 5 gigawatts of AI data center capacity under development. Two and up is a reasonable standard for now. That’s a big energy megaproject that serves up that much baseload. In order for that to be stable, you need a mix of sources or a connection to the grid. When you look at our massive-scale power needs, I don’t really think it’s a good idea to be drawing from the grid.'' @bradfordcross on the power requirements of serious AI training campuses.

''You can’t bring that move fast and break things type vibe into a developing world government that’s trying to figure out how to accelerate its economic development strategy and bring in more FDI. You also don't have the fail fast rapid experimentation thing where you can just stand up an app and start shipping features and see how people use it and start iterating in place on top of that. The iteration cycles are longer in these other domains.'' @bradfordcross on why city building and economic development can’t run on Silicon Valley software logic

Google DeepMind’s GNoME identified 381,000 newly discovered stable materials from 2.2 million predicted crystal structures. AI can now produce candidate structures far faster than physical labs can test them. Berkeley Lab’s A-Lab realized 36 compounds from 57 targets in 17 days using robotics, machine learning, active learning, literature-mined synthesis data and automated characterization. That creates a new bottleneck in materials discovery: synthesis, characterization, experimental data quality and the speed at which failed experiments inform the next round. CAMEO reduced the number of experiments required for materials discovery by 10x by using AI to select more useful experiments. Coscientist, a GPT-4-based system, autonomously designed, planned and executed chemistry experiments using cloud-lab commands and liquid-handling instruments. So the companies that win in materials, chemistry and biology will likely be the ones that control the full experimental loop: AI-generated hypotheses, automated testing, proprietary experimental data and verified physical results.

Rhodium Group estimates geothermal could economically meet up to 64% of expected data center electricity demand growth in the early 2030s under baseline assumptions. For serious AI training campuses, we should be thinking in terms of 2 GW and above. At that scale, stable power becomes a core infrastructure requirement. Training clusters run continuously, drawing enormous amounts of electricity around the clock. Geothermal power plants typically operate at roughly 90% capacity factor, producing electricity continuously throughout the year. Utility-scale solar generally operates around 25-35% capacity factor, while wind typically ranges between 30-45%. The U.S. Geological Survey estimates approximately 135 GW of enhanced geothermal potential in the Great Basin alone, enough to supply roughly 10% of current U.S. electricity demand. A 1 GW geothermal-powered AI campus can generate approximately 8 TWh of electricity annually, while a 20 GW platform would generate approximately 160 TWh annually.

"A lot of the world's resources go into politics. Much less goes into execution." @bradfordcross on why Africa has an opportunity to rethink how cities are built, governed and operated.

DOE and Lawrence Berkeley National Laboratory estimate that U.S. data center electricity use rose from 58 TWh in 2014 to 176 TWh in 2023, and could reach 325 to 580 TWh by 2028. That would push data centers from about 4.4% of U.S. electricity consumption in 2023 to as much as 12% by 2028. LBNL’s latest interconnection data shows more than 2,000 GW of generation and storage capacity still seeking connection to the U.S. transmission grid, while projects that do reach operation are taking longer to move from interconnection request to commercial operation. @OpenAI, @Oracle and SoftBank’s Stargate program is now described as a $500 billion, 10-gigawatt AI infrastructure commitment. xAI’s Colossus buildout in Memphis used dozens of gas turbines to support rapid deployment when grid capacity could not move fast enough. Together, these cases show that power has become a core site-selection constraint for AI infrastructure and not a secondary utility question. Large AI campuses need dedicated power strategies from day one, including on-site generation, co-located generation, storage and firm power procurement, instead of assuming the existing grid can absorb gigawatt-scale demand on AI timelines.

"I think we've built a lot of things wrong. We have sustainability challenges, cultural challenges, health epidemics, so many problems in the West. We can learn from those and try to develop Africa in a new way that avoids a lot of those types of problems." — @bradfordcross , speaking about Africa's development opportunities

In 50 years, over half of the world's largest cities will be African. Maybe 13 or 14 of the top 20 cities in the world will be here on the continent. What's going to happen is this leapfrogging effect where a lot of the cities will be just super new, state-of-the-art, just amazing. Like some of the best cities on earth will be here in Africa. They'll be very futuristic.

As an entrepreneur, you accept uncertainty. You know you're never going to be the best at everything, but you want to have at least enough expertise that you can do some damage. You don't want to come in and be like pretty helpless and learning on the fly.