The second phase

The release of Moonshot AI’s open weight Kimi K3 model has reignited a familiar debate.

Is China catching up? (And are they doing so simply distilling US models?) Should the United States react by tightening export controls? Changing the rules governing distillation (as argued by Ben Thompson)? Or regulating open source AI usage itself (as suggested by OpenAI’s new head of strategic futures)?

Some commentators have gone further, describing Kimi K3 as a form of technological dumping: an attempt to destroy the economics of frontier model development by releasing a highly capable model at little or no cost (much as China has been accused of dumping steel or solar panels onto global markets).

Maybe they’re right. But treating this solely as a trade threat misses a historical reality: technological revolutions always pass through two phases…and we’re on the cusp of the second.

As economic historian Joel Mokyr argues in The Lever of Riches, the societies that ultimately win aren't always the ones that produce the initial breakthrough. They are the ones that become unusually good at accumulating and applying what he calls useful knowledge. Britain's advantage wasn't simply that it produced great inventors, but that it connected a global culture of inquiry (the so-called Republic of Letters) to practical engineering and manufacturing.

The story of James Watt helps bring this argument to life.

You may remember Watt as the inventor of the modern steam engine. He was also the lucky recipient of an extended patent on his steam engine design, giving him and his partner Matthew Boulton an unusually long monopoly over one of the defining technologies of the Industrial Revolution.

Inventors deserve to capture value from their inventions. Without that incentive, many breakthroughs would never happen. But when Watt’s patent expired, the pace of innovation skyrocketed as thousands of engineers got to work improving on the design. Indeed, in the Cornish mining districts, engineers published detailed operating data and design changes through Lean’s Engine Reporter, each building on each other’s improvements in an old-school version of open-source.

These are the two phases of every technology revolution. The first is the race to invent. And that phase rewards the individuals that produce genuine breakthroughs. The second is the race to improve. And that phase rewards the societies that become exceptionally good at compounding small improvements over time.

Most of today's AI debate remains focused on the first phase, as if this breakthrough can still be controlled. But whether policymakers like it or not, frontier intelligence has already begun to spread. Weeks ago the U.S. government sought to keep Fable contained. Today, Fable-class intelligence is available to anyone in the world. We live in a global economy, and if China has decided that making frontier intelligence abundant serves its national interest, there may be no practical way to reverse that decision.

None of this argues against protecting intellectual property or tightening export controls, but those are debates about the first phase of competition. They are not a strategy for the second.

Here’s where I think we can all learn a thing or two from Mokyr. If China is engaged in technological dumping, the West’s best response isn’t to build tariff walls or regulatory barriers. It is to do what the Cornish miners did: take the widely available technology and improve it faster than anyone else.

In this way, I find myself wholly agreeing with Thompson. Due to US frontier lab restrictions, US open weight model makers are forced to distill from Chinese models. The result is worse models (that are distilled from a distillation) and a dependence on China. What the West needs is a thriving ecosystem that encourages the diffusion of useful knowledge, as Mokyr calls it, and the compounding of small improvements over time.

That was Britain's comparative advantage during the Industrial Revolution, and it could become the West's comparative advantage now.

David Peterson

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

Another Deepseek moment? Chinese AI startup Moonshot released a new open source model, Kimi K3, which rivals the frontier models and even surpasses them by some accounts. “Moonshot said that the model, Kimi K3, was the world’s largest open-source A.I. system, allowing anyone to use, modify and build on it freely. The company said that Kimi K3 performed as well as leading models from OpenAI and Anthropic at some key tasks. The release coincided with an address by Xi Jinping, China’s leader, in which he outlined an ambitious vision for global A.I. development that cast China as the champion of an open approach to the technology.”

Sam Lessin is depressed about AI, and the reason why is interesting. Over the past two weeks, AI headlines have been dominated by three closely related narratives: growing alarm at the cost of AI utilization, the meteoric rise of open source AI capabilities, and the growing sense that leading foundational models labs are engaging in some sort of regulatory capture attempt by proposing regulatory frameworks that benefit them. In the most recent More or Less Podcast, Sam Lessin made a very compelling set of arguments (the section starts about 19 minutes in) about why he’s “depressed” by this moment in technology: “Well, look, we're back at a story where everyone's like, well, it's the application layer. I don't think that's true. So I think what it is is there are companies that can use this [stuff] to be way more profitable. They will use that [stuff] right?...and those companies will be better. And that's the story of this AI stuff. The actual ‘one model to rule them’ all? we're clearly past that… The fact that who's in the lead keeps leapfrogging and open source is right there [means] we're just multiplying big numbers and then. . .because the fundamental business structure is not playing out in a way where any of these trillion dollar companies really win. They're just in an expensive war forever war [so they will] take a shot at regulatory capture because that actually is the smart business move… I think it's very cynical for these companies to even try to switch the field of play from business to politics. I think that's a very cynical anti-American thing to do to be [say] ‘Okay, we're not going to play and compete in capitalism in a fair and open system. Instead, we're going to switch our venue to regulatory capture because it's the only play we have.’ And I get why they do it. And to some degree as businesses, I don't begrudge them the play they need to make. It's the classic ‘you can't win the game so change the game.’”

Cyber M&A. U.S. data infrastructure firm Cribl has acquired Israeli cybersecurity startup CardinalOps for approximately $100 million. CardinalOps, which had raised $40 million, specializes in AI-powered threat detection. “The acquisition will extend Cribl’s platform into security operations by adding detection engineering capabilities designed to help customers improve threat coverage, reduce data costs, and strengthen their security operations centers (SOCs). The move will allow Cribl to offer customers a more flexible alternative to traditional SIEM (Security Information and Event Management) architectures.”

HARD MARKETS

Europe’s defense tech boom continues. German defense tech giant Helsing raised $1.8B at an $18B valuation, sparking some concern among skeptics. According to the FT (paywall), “the headline valuation has raised concerns among some competitors and investors at a time when the sector was gripped by a “raise race”, according to one sceptical executive from an established defence group.“Quantum, Stark, Helsing . . . all are raising money like crazy — because it is there at the moment,” he said. “This feels somehow close to the dotcom bubble.”” At this point, however, it’s quite possible that Helsing is already too big (and too well funded) to fail, “ Plural’s Helioui said the fund had done its own “extensive independent due diligence on the company” before investing this time, speaking to the military and competitors to cross-check the facts. “On autonomy and drone warfare, we believe that Helsing has such a scale and velocity advantage versus all of the competitors that they are uniquely positioned. So that’s our bet. That they are going to be the defining player.” Von Borries said that, if Helsing really could widen its focus and grow to become a European equivalent of the American tech giants, then an $18bn valuation might end up being far too modest. “We see the vast growth of AI companies in the US. If Helsing can diversify and branch into dual-use applications in future, then maybe the valuation actually underestimates its potential.””

HOW TO STARTUP

Focus on top of funnel. This interview with Sam Blond (of Monaco, Brex, and Zenefits) by Jack Altman is great. It’s full of useful GTM insights and perspective. One of the best observations Sam shared was that a lot of GTM leaders focus on conversion rates when they should be focused on top of funnel. In many markets, you can only increase your conversation rates by small amounts with a lot of effort, but you can 10x your top of funnel with moderate effort: “You can improve all these things a little bit, but you can improve the top of funnel. if you have a company worth building, there's like a hundred times more customers that you could be talking to than you're talking to. I think this is like a very underrated thing and it's like kind of like a red pill once you see it…If you have 10% conversion rates,improving those conversion rates to 20%, which sounds like you're improving your conversion rates by 10%, but you're actually doubling conversion rates. It's really hard… it is far easier to double your leads or opportunities and so that is where I would uh put a disproportionate amount of intention.”

HOW TO VENTURE

Where’s the moat? This interview with Jenny Fielding of Everywhere Ventures was short and worth a watch. She focuses on what’s different in VC in 2026 and what she’s struggling with: “Not all your companies are going to make it, but are they not going to make it because they're not good companies or are they going to be displaced by some company coming out of YC that's like able to catch up quite quickly? And so that's the thing that is not exactly sitting with it. Yeah. I think to be in the game now, you have to have a viewpoint on what companies continue to thrive in the AI paradigm.”

Capital concentration is the story of Venture Capital right now. Trace Cohen shares the numbers (“The five largest venture managers captured 73.1% of all capital committed to the asset class in the period, and the top 15 firms captured 88.5% -- meaning roughly seven out of every eight LP dollars flowing into venture this year went to a small, well-established group of brand-name firms.”) and some potentially dangerous implications: “The bear case: extreme concentration at both the fund and company level is historically a leading indicator of a market top, not a sustainable steady state -- when capital consolidates this aggressively, it typically means the market has priced in most of the obvious winners and has little room left to broaden before either a correction or a genuine widening of opportunity. What to watch next: whether H2 2026 fundraising data for emerging and first-time managers shows further deterioration, and whether any of the mega-funded AI companies capturing outsized rounds show signs of the growth that would justify this level of concentrated conviction.”

Fixefy has been named the freight, audit & payment AI platform of the year by FreightWaves in their 2026 AI Excellence in Supply Chain Awards.

Groundcover
Backend Engineer (Tel Aviv)

Levity
Forward Deployed Engineer (Germany, Remote)

Motorica
Platform Engineer (Stockholm, Hybrid)