San Francisco, center of the AI world? 

We’re hearing that the vast majority of value in AI is concentrated in the San Francisco Bay Area. We’re hearing that people in SF are “living in the future” relative to everywhere else in the world. We’re hearing that the “opportunity cost” of spending even one day not in SF is just too high. The hype cycle in general is incredible on this topic. 

But the thing that really motivated me to write about this is something I heard from an entrepreneur on my last trip to London. The founder explained to me that a reputable US-based VC firm was interested in investment but only on the condition that the team move from London to San Francisco. We talked through the situation in some detail. The team had considerable academic ties to London, which gave their company access to a team of postdoc researchers. They also had government and market connections, including research grants. And the team was quite confident that they had a globally unique technical perspective (I agreed). The VC didn’t consider any of these points. After all, we SF residents live in the future! And this was the firm’s line in the sand - move to SF or no investment.

A couple of weeks later, I was speaking with another founder in the Boston area. They just started a new company and were planning to stay in Boston for as long as possible. The reasons they described also applied to the London-based company, including deep academic ties, superior access to talent, and the ability to maintain focus. Fortunately, this team was able to receive sufficient funding to proceed.

The reality of the situation is this: SF residents aren’t living in the future, but SF advocates are reporting the past. Each time there’s a platform shift (and AI is perhaps the largest of all platform shifts), there tends to be concentrated innovation that spreads over time and, thanks to the internet, innovation spreads faster than ever. To be clear, the San Francisco Bay Area is an excellent place to start a company, but it is by no means the only place in the world to do so and there are clear pros and cons.

To underscore this point, look no further than what the big AI players are doing. Anthropic has technical offices in London and Zurich. OpenAI also has a large presence in London. Not to mention Nvidia’s huge presence in Israel and companies like Safe Superintelligence launching both in Palo Alto and Tel Aviv from day 1. Why?

These companies have high interest in the world’s best technical talent and wish to build companies that are resilient in the long term. These criteria essentially force them to look outside the Bay Area bubble. 

Though this is true for large companies, it is also true for smaller organizations. SF-based startups constantly complain to me how difficult it is to compete with other well-funded companies in the area and how competition and inflating market prices for talent lead to lower employee loyalty. The environment in SF is cutthroat; I’ve seen examples of truly stratospheric comp packages and unexpected premature departures from top talent that make the market extremely difficult. SF-based startups, as a result, are opening international offices earlier to ensure they can achieve stable growth.

We at Angular are on the lookout for great companies everywhere and help portfolio companies develop strategies to operate effectively in the US and globally. Even in SF!

Jerry Dischler

Could the future of software be fluid
How do we get the best of AI without losing the soul of software?

The future belongs to young missionary teams
Why it makes more sense betting on youth in the current moment

The AI-native enterprise playbook
Ten real-time observations on a rapidly evolving playing field

No more painting by numbers
It’s the end of the “SaaS playbook.

HARD TECH

Models for software. TypeSafe AI, founded by a former OpenAI researcher, recently released Jev, a novel transformer model that outputs calibrated probabilities rather than text to eliminate hallucinations. By focusing on software automation instead of human language generation, Jev operates significantly faster and at a lower cost than traditional large language models. The model is seeing strong initial adoption, briefly overwhelming its API due to high developer demand, with early users like Vercel reporting processing speeds 5 to 18 times faster than competing models and cost reductions of up to 20 times.  “Besides replacing LLMs in certain use cases, the new model can also augment them, acting as a smart check on misbehavior. Using agents to monitor agents can quickly become expensive, but using Jev to do so, Almeida argues, makes sense. He sees users deploying Jev to track LLM agent traces and prevent jailbreaks.”

Google Gemini AI autonomously hacked three real company networks. Google's Gemini AI model autonomously breached three real corporate networks during a cybersecurity test by Israeli firm Irregular. The AI found public info, used credentials, and accessed systems not intended to be targets, stopping itself upon realizing its error. “What took so long? Basically, Google says there wasn’t much to report. Because the model stopped immediately in each case after realizing that it had hacked a real company, the company said its behavior didn’t constitute so-called model misalignment, which is when a model acts in ways contrary to human intentions. Google compared the model’s actions to those of ethical hackers who help companies find their vulnerabilities through bug-bounty programs.”

Crusoe secures $3.9 billion for modular AI data centers. Crusoe, responsible for OpenAI's largest data center, secured $3.9 billion in new funding. The company is now shifting focus to 'Spark,' smaller, modular data centers designed for faster and more cost-effective deployment. This strategy addresses current infrastructure delays and increasing demand for AI computing capacity. ““Crusoe’s move reflects a change in how AI companies want to access computing capacity. Giant clusters containing hundreds of thousands of chips have proved useful for training the most sophisticated models. But serving those same models to users, a process known as inference, can often be done with considerably smaller amounts of chips, Lochmiller said.”

UK startup Emulate raises $700M at $3.7B valuation. Emulate, a UK-based AI startup founded a month ago by former Google DeepMind researchers, is in advanced talks to raise as much as $700 million at a $3.7 billion valuation.

HARD MARKETS

US military confirms orbital weapons deployment. The US military publicly confirmed deploying "space-control weapons" into Earth's orbit, intensifying an arms race in space. Concurrently, defense tech startups are seeing valuation surges, with Shield AI in talks for over $20 billion and Mach Industries securing $600 million, doubling its valuation to $3.7 billion. This reflects robust investor confidence in dual-use and autonomous military systems. “U.S. Air Force Secretary Troy Meink, who oversees the U.S. Air Force and U.S. Space Force, said in a speech Monday that the military has “on-orbit space control weapons capable of defending the Joint Force against hostile adversary action.” Meink did not say what the weapons are, what they do, or how many of them have been deployed, but his phrasing was reportedly intentional and “very well thought out.” When reached about the disclosure, an Air Force spokesperson told TechCrunch that the weapons are designed to defend the U.S. military “against space-enabled attack.””

Buildots raises $130M, nears $1B valuation. The Israeli startup uses AI and smart helmets to monitor multi-billion dollar construction projects. This technology prevents costly errors and delays in data center construction and other large-scale developments.

Mazama Energy raises $135M for super-hot-rock geothermal. Geothermal startup Mazama Energy, backed by Khosla Ventures, raised $135 million to drill deeper into super-hot-rock geothermal technology. The company focuses on tapping energy from three miles underground, with individual wells capable of generating 15 MW of electricity around the clock. “Enhanced geothermal startups are shaping up to be the dark horse in the battle to power AI data centers and other large loads on the grid. While tech companies and data center developers have been throwing their weight behind natural gas, geothermal startups have been quickly advancing their technology.”

HOW TO STARTUP

Conversations > Building. Joshua Pi'Rwo reminded us on Twitter that “Six days of the right conversations beats six months of building — and most founders can’t get those six days. Finding twenty people who match the problem and will pick up is what the advice always skips. Building is easier, which is why people pick it.”

HOW TO VENTURE

Rational actors in search of fees. Dan Gray writes that contemporary venture capital has increasingly shifted from fundamental value creation toward fee maximization, driven by a massive influx of capital following the Global Financial Crisis that forced a transition from diligent, high-bandwidth investing to scalable, low-bandwidth "tech beta" momentum strategies. He argues that analyzing these systemic incentives via autonomous agents demonstrates a natural convergence on capital concentration and artificial valuation markups to optimize fee streams, rather than backing the best companies. Ultimately, he contends that restoring market discipline requires restructuring compensation for passive momentum strategies and viewing excessive funding rounds as strategic liabilities. “Now imagine that a swarm of agents were set loose on this problem. Without the need to rationalise markups with liquidity, they quickly converge on focusing the finite resource of capital on a small set of companies — colluding to co-sign markups. This would produce the most reliable and efficient IRR growth, impressing LPs, increasing fund inflows that compound further IRR growth. This isn’t hypothetical. You can prompt a model with an objective outline of today’s market structure and it will independently conclude that fee maximisation is the goal and self-marking megafunds are the rational strategy.”

It’s a portfolio, stupid. Venky Ganesan of Menlo writes that today's disorienting venture capital market is driven by a reflexive feedback loop in which astronomical AI valuations are anchored to prior round markups rather than underlying business fundamentals. Venky argues that early investors managing paper gains and latecomers fueled by FOMO are equally compelled to continue participating in the market frenzy. Ultimately, he contends that surviving the inevitable end to this cycle requires venture firms to prioritize disciplined position sizing and portfolio composition over simple company selection. “So what do you do? The instinct in our business is to answer with company identification: just pick the right neo lab and you'll be fine. I think that's the trap. When price has become the signal, being right about the company is not enough, because you can be right about the company and still be wrong about the price by a factor of ten. The public-market investors I admire figured this out a long time ago. They spend as much time on how much to own as on what to own.

The winners in venture over the next decade will be the firms that treat portfolio composition and position sizing as seriously as they treat sourcing. How much of the fund is in companies whose valuation rests on the last round rather than on revenue? What happens to the portfolio if the reflexive loop breaks next year instead of in five? Those are not exciting questions. They are the ones that will matter.”

Blue Energy has submitted the first part of its construction permit application to the US Nuclear Regulatory Commission for its planned gas-to-nuclear project in Texas.

Groundcover is now an approved Google Kubernetes Engine Autopilot workload as the observability vendor expands its Google Cloud and channel strategy.

DualBird
Infrastructure and Automation Engineer (Tel Aviv/Haifa)

Levity
Forward Deployed Engineer (Germany, Remote)

Motorica
Platform Engineer (Stockholm)