- The Angle
- Posts
- The full-stack delusion
The full-stack delusion
David Peterson
In 2017, VanMoof started shipping some of the most beautiful e-bikes ever designed. By 2023, the company had raised nearly $200 million. And by mid-2023, it was bankrupt. The cause of death wasn’t lack of demand. The problem was that VanMoof had engineered every component in-house, from the motor to the bolts, ignoring the highly efficient global supply chain for bicycle parts. When something broke, no local shop could fix it. And that beautiful bike became a beautiful brick.
VanMoof is an example of a larger pattern. Somewhere between the success of SpaceX and the flood of capital into deep tech, venture capital developed a vertical integration problem. We decided that if a founder wasn't building the software, the hardware, and the factory from scratch, they weren't thinking big enough. Capital expenditure became a proxy for ambition.
Northvolt raised over $13 billion to build a fully-integrated European battery supply chain, trying to out-manufacture China in standard lithium-ion cells. But you can't buy China's 20 years of accumulated process knowledge with venture dollars. (For a deep dive on how the Chinese accumulated all that process knowledge, read Breakneck by Dan Wang.) Northvolt filed for bankruptcy in early 2025. Companies like Infarm, Plenty Unlimited, and Bowery Farming raised billions combined to grow lettuce indoors. Their AI and robotics were legitimately impressive. But that tech could have been licensed to existing growers or deployed through partnerships. Instead, they vertically integrated everything. Not just custom software, but massive real estate leases and facilities built from scratch. All to sell a commodity whose unit economics could never justify the capital structure. By mid-2025, funding for the category had dried up and the biggest players were bankrupt.
When does vertical integration make sense?
Vertical integration does make sense in certain cases. For example, sometimes the manufacturing process is the product. My good friend Julia is one of the founders of Antares, which designs and builds next-gen nuclear microreactors. They need to be vertically integrated, because the physics of the reactor and the manufacturing of the reactor are inseparable.
As Packy McCormick argues in his Vertical Integrators series, sometimes the integration itself is the innovation. One of our portfolio companies, Blue Energy, is a perfect case study of this approach. Blue Energy is a nuclear power plant developer that uses mature light water reactors. No new reactor technology. The insight is that the reactor accounts for only 7% of a nuclear power plant's cost. The other 93% is construction overhead, interest during construction, and materials. Blue Energy redesigned the plant to be manufactured at existing shipyards, compressing build times from 10+ years to less than 3 and cutting costs from $13K/kW to $5K/kW. The integration is the innovation.
So, yes, sometimes vertical integration makes sense. But everything is being vertically integrated these days, and not always for the right reasons. The most common pushback I hear from founders is that legacy incumbents are terrible software buyers, or that legacy markets are too concentrated. So they have to vertically integrate. They're right about the diagnosis. But you can't fix a go-to-market problem by adopting a worse business model. If VanMoof and Northvolt have taught us anything, the cure will be fatal.
Stop catching up, start leapfrogging
The default in deep tech right now is vertical integration. I think the default should be leapfrogging.
Leapfrogging is when a country, market, or company bypasses an expected stage of development entirely. The canonical example is mobile phones in sub-Saharan Africa. In the early 2000s, countries like Kenya and Ghana had almost no fixed-line telephone infrastructure. Rather than lay copper wire across vast geographies (the path every Western country had followed), they skipped landlines altogether and went straight to mobile.
Then it happened again. M-Pesa, launched in Kenya in 2007 by Vodafone and Safaricom, leapfrogged not just landlines but the entire Western banking infrastructure. In a country where fewer than 30% of citizens had a bank account, M-Pesa enabled money transfers, deposits, and payments via text message. The technology ran on basic SMS. But the architectural insight was powerful: don't build a branch network to compete with Western banks. Build something native to the infrastructure that already exists and skip the intermediate step entirely. By 2023, the value of transactions flowing through M-Pesa was multiples of Kenya's entire GDP.
And note what Safaricom didn't do. They didn't build handsets or construct cell towers from scratch. They built the software layer and let existing infrastructure handle deployment. The capital to expand came from government grants, telecom operating budgets, and eventually the profitability of the service itself.
For a more recent example, look at how China hollowed out Western automotive leadership. Chinese policymakers realized they would never catch Western manufacturers in the internal combustion engine. So they refused to play the game, poured investment into battery-powered EVs, and shifted competition from mechanical engineering to electrochemistry. They bypassed a century of dominance by changing the rules.
We need to run that playbook back at them. Not by building slightly cheaper legacy factories, but by commercializing solid-state batteries, synthetic biomanufacturing, next-gen nuclear, and post-silicon compute.
The economic gravity test
The litmus test for a true leapfrog: does the technology exert enough economic gravity that incumbents will fund the physical deployment just to access the IP?
ARM designed the low-power chip architecture that came to dominate mobile computing. They could have tried to build multi-billion dollar fabs. Instead, they licensed the IP. Apple, Qualcomm, and Samsung paid for architecture access and did their own chip design on top. TSMC built the factories. ARM kept gross margins above 95%. (The Acquired podcast episode on ARM is a must-listen.)
LanzaTech engineered microbes that turn carbon emissions into valuable chemicals. They could have raised venture capital to buy steel mills. Instead, they partnered with ArcelorMittal, which, with EU grants and a European Investment Bank loan, built a €200 million facility in Ghent. LanzaTech provided the biocatalyst and the process technology. ArcelorMittal provided the steel plant, the gas feedstock, and the capital. The facility began producing ethanol in 2023. LanzaTech's technology was valuable enough to ArcelorMittal's decarbonization strategy that the steel giant spent hundreds of millions to build around it.
M-Pesa fits the same pattern. Safaricom's platform was valuable enough that Vodafone matched a UK government grant to fund the pilot, existing mobile agents became the distribution network, and Kenyan banks eventually partnered with the platform rather than fighting it.
In each case, the innovator built the intelligence layer. The deployment capital came from someone else: an industrial partner, a government program, a project finance lender, or an incumbent who needed the technology.
And if you can't find any of those partners willing to fund the deployment, that might tell you something important about whether your technology is actually the leapfrog you think it is.
A true leapfrog doesn't just change the technology. It changes who pays. ARM's architecture was so valuable that chipmakers funded the fabs. LanzaTech's biocatalyst was so valuable that a steel giant funded the plant. M-Pesa's platform was so valuable that a telecom, a government, and eventually the banks themselves funded the rollout.
The principle extends beyond leapfrogs. Blue Energy isn't commercializing a new reactor. But they designed the plant so that infrastructure lenders could say yes, making nuclear energy project-financeable for the first time. Those design choices are as much of an innovation as any novel technology. (I've written more about this before here.)
In traditional venture capital, finance enables technology. On the frontier, technology enables financing. The founders who internalize that will define the next decade.
Stop catching up, start leapfrogging.
FROM THE BLOG
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.
WORTH READING
ENTERPRISE/TECH NEWS
Anthropic and Pentagon in open war. Anthropic announced it would sue the Pentagon for wrongfully declaring it a supply chain risk. “San Francisco-based Anthropic is challenging a decision by the department and other federal agencies to shift their AI work to other providers, based on a risk designation typically reserved for companies from countries the US views as adversaries. Anthropic wants a judge to remove the supply-chain risk designation and require US agencies to withdraw directives related to it. The company claims it is being shut out for disagreeing with the administration and argues the legal principles at stake affect every federal contractor whose views the government dislikes.” At the same time, OpenAI robotics lead Caitlin Kalinowski publicly quit citing concerns about that company’s willingness to work with the Pentagon on issues like surveillance of American citizens and autonomous weaponry. “AI has an important role in national security. But surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got.”
Appetite for Computer. Nscale raised $2 billion at a $14.6 billion valuation, according to the Wall Street Journal, “a sign that investors continue to see promise in the artificial-intelligence infrastructure startup as it seeks to roll out data centers. The U.K.-based company said Monday that Nvidia, Dell Technologies and Nokia were among the investors that backed the Series C funding round, led by Norwegian industrial group Aker and venture capital firm 8090 Industries. The move shows that investor appetite for companies in the data-center business remains strong despite concerns that tech giants might be overspending on AI.”
HOW TO STARTUP
Pigment’s rise to $100M ARR. Fortune profiled French FP&A startup Pigment, which has doubled ARR three years in a row to get to $100M, proving that triple-triple, double-double might have legs yet. “Pigment’s core product aggregates business data into a single platform and lets teams run scenario models—different oil price paths, shifting tariff regimes—without blowing up fragile Excel sheets or waiting weeks for a specialist to reconfigure a model. Crespo’s latest bet is what she calls Pigment’s biggest leap yet: a “Modeler Agent.” Users describe what they want in natural language, and the agent generates governed, production-ready applications on top of Pigment’s data engine. Early customers say build times have collapsed from weeks to minutes. Figma reported getting to “80% of what they wanted to build from a blank page in minutes,” according to Crespo.”
Cheating on revenue. Cluely, a startup that began life promising to help students cheat on exams, admitted to lying about revenue numbers. The scale of the exaggeration was minor, but it seems to have hit a nerve, as so many founders feel greater pressure than ever to post aggressive growth numbers. Ryan Hoover put up a poll suggesting that many believe dishonesty is increasingly rampant among early-stage startups.
HOW TO VENTURE
Will AI eat VC? Wired covered an AI-powered VC fund called ADIN. “Launched in 2025, ADIN uses AI to replace the human analysts involved in venture dealmaking. Put in a startup’s pitch deck, and out comes a detailed analysis of its business model and founding team, a list of diligence questions and compliance risks, an estimate of the total addressable market, and a suggested valuation. ADIN has about a dozen different agentic investors, each with a distinct persona and investing thesis. Tech Oracle looks at a startup’s underlying technology; Unit Master evaluates the financial fundamentals; Monopoly Maker, loosely based on Peter Thiel, looks for market dominance. When the majority of the agents like a startup, they suggest how much ADIN’s fund should allocate to the deal. The platform does this in about an hour, compared to the days or weeks that it takes an analyst at a VC firm.”
It just keeps concentrating. The Information reported on the continued concentration of the majority of LP dollars into a very small number of mega VC funds. “In fact, the recent funds raised by Thrive, Andreessen Horowitz and Lightspeed amount to $34 billion, or half of all the money U.S. VC funds raised last year. Overall venture fundraising sank 34% in 2025 to just under $68 billion and has plummeted almost 70% from the fundraising frenzy of 2022, according to PitchBook.”
PORTFOLIO NEWS
PORTFOLIO JOBS
Reply