Vertical Integration Is Not a Magic Bullet to Fix Every Industry

Vertical integration only generates outperformance when it removes real fragility, enables capabilities that modular suppliers can’t deliver, or creates a step change in performance or economics. This is especially true in domains where latency, safety, secrecy, or system coupling make interface failures catastrophic - or system-level optimisation is highly constrained by compartmentalisation. 

SpaceX vertically integrates propulsion, avionics, mission integration, and ground systems because a third-party interface fault can destroy a payload and derail flight cadence. Decades of modularization had layered contractors and interfaces, driving costs sky-high. By collapsing that stack, SpaceX unlocked iteration speed and cost discipline, cutting launch costs substantially.

But integration becomes a liability where performance is already “good enough,” supplier ecosystems are mature, or the learning curve rewards specialization over control. Most automotive OEMs still rely on tier-1 suppliers for EV drivetrains and infotainment because the marginal gain from insourcing doesn’t justify the capital burden and complexity of integration. Even Apple, often mislabeled as a fully integrated company, only pulls functions in-house when supply risk, UX leverage, or IP defensibility demand it. It designs its own chips and OS, but still buys displays and memory from Samsung, LG, and Micron, and only recently moved to swap out Qualcomm modems after years of failed internal attempts.

The right question to ask when contemplating vertical integration is: “Does controlling the stack unlock design freedom, cost compression, resilience, or time-to-market advantages that modules can’t match?” and “can the integrator lock in that advantage through spec control, bottleneck ownership, or a superior learning curve?”

Where those answers are no, modular ecosystems win. Nvidia’s CUDA stack dominates because third-party developers build around it, not despite it. TSMC out-executes vertically integrated fabs by focusing on one layer of the stack and letting others conform to its process rules. Airbus governs a supplier web rather than owning every subsystem because its leverage lies in certification, systems integration, and program architecture, not making every part. 

Integration only wins when it changes the physics, economics, or governance of the system. Otherwise, it's just cost, complexity, and distraction dressed up as strategy.

Unlocking System-Level Efficiencies Requires Technological Readiness

Vertical integration only creates an edge when the underlying technologies in the value chain are already mature enough to be integrated without requiring one or multiple scientific breakthroughs. If a company needs to unlock a step-change in battery chemistry, propulsion physics, chip architecture, or materials science just to make the model viable, you’re no longer dealing with integration engineering risk, but with a deeptech gamble that introduces an existential failure point. As covered in Chapter I.III., this “method over output” approach drives up capital intensity, stretches timelines, and demands a very different type of team and financing structure.

The sweet spot is when the core technologies already exist in silos but haven’t yet been unified into a coherent system. SpaceX didn’t invent new physics, it recombined established propulsion principles, avionics, manufacturing automation, and ground systems into an integrated launch stack. Hadrian isn’t reinventing CNC machines, it’s integrating proven machining hardware with robotics, software, and process control to collapse lead times and tolerances.

What matters is using integration to unlock system-level trade-offs - state-of-the-art components should be available, but fragmented or sub-optimally combined. When integration fills coordination gaps rather than technological voids, the upside is in execution speed and design freedom, not foundational scientific research risk.

Deeper technological innovation can still be pursued later, but it should be an option created by scale and capability, not a prerequisite for the model to work.

Beware of Fata Morganas – Sometimes Fragmentation Is Part of the System Design

Not every messy value chain is a vertical integration opportunity. Some industries only appear broken because outsiders mistake structural, often regulatory constraints for solvable inefficiencies. What appears like fragmentation, margin slack, or coordination failure is often deliberate design - to spread liability, satisfy regulators, capture subsidies, meet geopolitical commitments, or comply with certification regimes.

Technical integrability also doesn’t equal ability to capture value. In most autonomous-trucking models, for example, the majority of value accrues to the fleet operator, not the autonomy stack vendor. 

If you don’t understand why a system is fragmented, you might misread deliberate design decisions as inefficiency and consolidation opportunity. The right questions would be “is the fragmentation accidental, or performing a job we don’t see yet?” The best VIs attack genuine coordination gaps, not physics problems, liability shields, or regulatory moats. Before committing, you have to prove that consolidation unlocks and captures value.

Survivor Bias

Benchmarks for scalability, valuation multiples, and capital intensity in the Chapter II.II. on misconceptions & benchmarking are likely skewed by survivor bias. The VIs that successfully scale - and capture spec power, bottlenecks, and learning-curve advantages - dominate the data and set the narrative. Those survivors show high multiples, low relative equity consumption, and deep moats, but they are likely the exceptions, not the rule.

Just like in all venture-backable categories, most VIs never get there. They burn through capital trying to prove product-market fit and ultimately trade at or below incumbent multiples, but ultimately, I believe that Vertical Integrators have a higher break-out rate than typical SaaS, i.e. a higher % of total VIs actually achieves venture-grade outcomes - we simply need more data and time to show the trend.

Bottom line: Venture Capital is defined by power-laws, so what matters in the end, is that these principles around dilution, scalability, and valuation premiums hold true for the break-out companies, and that appears to be true. And as highlighted in previous chapters and later in the conclusion, there are methods to reduce these risks pre-breakout.

Limited Data Availability

The Cambrian explosion of Vertical Integrators is still in its early years. Outlier data-points exist, but most are mostly non-public information. There are also no standardized KPIs yet to report against, so proper public benchmarks of modern Vertical Integrators will need more time.

This is not unusual and mirrors the early SaaS period - before ARR, CAC, and NRR became part of the vocabulary, investors relied on intuition and pattern recognition. Only later did a shared analytical language crystallize, defining an entire asset class. Vertical Integration today sits in that pre-metric frontier: the inflection point is palpable, but no one can yet measure it cleanly.

There are few dimensions along which performance might eventually be measured - I’m working on a dedicated blog post on this, but here are some early thoughts:

• Throughput Efficiency: Annual output in gross profit per $ of CapEx (+ adjusted for debt/equity mix)

• Automation Leverage: Output in gross profit per unit of human labor cost

• Capital Cycle Time: Time from CapEx deployment to positive cash flow (on a per-site basis)

• And other financial metrics as highlighted in the “Revenue Predictability” segment in chapter II.II.

The defining characteristic of a Vertical Integrator is not its technology stack, but the compression of the capital-to-output loop: the rate at which (equity) capital, energy, and data cycle through production into scalable real-world output. In time, this dynamic will likely serve as the core “scaling law” of the Integration Age.

Overcoming Scaling Bottlenecks

Scaling real world infra requires standardisation - heterogeneity kills at scale. That’s why selling a holistic SaaS suite into existing factories is difficult - each factory has different machines, processes, people. The design of the real-world environment in which the software/tech is deployed is under the control of the Vertical Integrator. Standardizing their real-world infra enables the VI to scale much faster.

VIs can also introduce elegant buffers to absorb complexity. Isembard takes a more decentralised approach to scaling their supply side - rather than few, big factories, they are building a network of small- to medium-sized factories. The benefit being, they can be built faster, cheaper, and reach high utilization more easily. The resulting challenge is quality control across a fragmented supply side. By scaling with a franchise model, the person running the factory is an owner-operator, whose incentive to “run a clean shop” is much higher than a factory manager on a $60k salary. The franchise model also pulls more talent into the industry by lowering the entry barriers, ultimately addressing a potential long-term scaling bottleneck in the form of labor.

Scaling only works if the unit economics work and are proven, not theorized. Too many industrial projects assume economies of scale will magically appear at volume, only to discover that inefficiencies compound rather than disappear. Vertical Integrators earn the right to scale by validating margin structure early. The ability to generate a substantial gross margin uplift compared to incumbents is one of the clearest signs of actual technological leverage. I also like to think of superior unit economics as greater “margin for error”.

Even when unit economics are strong, scale only creates value if the market can absorb it. TAM and competitive dynamics set hard ceilings. If incumbents can easily copy your approach or if your technology moat doesn’t widen with scale then growth simply attracts better-funded competitors. The most powerful Vertical Integrators target markets large enough to absorb billions in annual output, while relying on technological or process advantages that compound as they grow. Scaling is only virtuous when every unit of scale increases your advantage rather than inviting commoditization. Path-dependent layering can build some of the most enduring monopolies, e.g. SpaceX could only build Starlink because they owned the world’s most cost-efficient launch infra.

Geopolitical Caveats

Geopolitical risk is very real but asymmetric. Most serious analysts now see conflict not as global cataclysm but as chronic regional escalations. Russia is increasingly geared for a long, attritional industrial war and remains a threat on Europe’s doorstep - a land war with further escalation into the Baltics and Poland within the next 2-3 years can’t be ruled out. Russia alone has about as many active military personnel as all of Europe combined. Russia is already producing at wartime industrial output and under unified command, whereas Europe suffers from fragmented decision making and delayed industrial scaling.

China poses the broader systemic and economic challenge, but its ability to convert economic scale into military dominance is constrained by some structural headwinds - aging demographics, local-government debt overhang, and capital outflows make expansion ambitions and ability beyond Taiwan unlikely. 

The US remains uniquely advantaged by geography, food and energy self-sufficiency, and naval dominance, making outright military displacement highly unlikely. 

The baseline scenario is not world war, but chronic regional confrontation and economic bifurcation. The strategic priority for the West is sovereign resilience: rebuild industrial capacity to ensure we cannot be coerced and don’t rely on foreign actors for basic needs.

Geopolitical Fracture Might Reverse Hardware Commoditisation

There is a real risk, that the geopolitical fracture above may reverse the very forces that enabled hardware commoditization in the first place - and is a key ingredient to make Vertical Integrators successful. Many of today’s cheap, modular components, such as batteries, sensors, semiconductors, or rare earths, became affordable because supply chains were global, efficient, and open. Rising conflict and export controls may re-nationalize or politicize those inputs, effectively un-commoditizing components, or at least stunting progress on further commoditisation.

This dynamic won’t hit all sectors equally. Vertical Integrators who are building powerful, intelligent, complex systems, providing a real step-change in unit economics are generally less exposed as they have more “room for error” when it comes to their foundational COGS. VIs who are just squeezing out a few % on top of largely commoditised stacks with limited performance gain (albeit that marginal gain may already be valuable), are much more exposed to price/cost fluctuations.

Reshoring much of the component production (see contract manufacturing opportunity in chapter II.I.) is a critical mitigant for this. The bottlenecks that I’m most concerned about are rare earths, refineries, and production equipment. In 2021, China produced 31% of global machine tools, followed by Japan with 13%. In 2024, ca. 40% of injection molding machines were produced in APAC, the majority of which in China.

The Place for SaaS & “White-Collar Vertical Integrators

SaaS is far from obsolete. It will continue to compound in stable industries and non-industrial knowledge work, and many Vertical Integrators will rely on SaaS vendors to fill specific gaps rather than build everything in-house. There is still meaningful opportunity in the industrial domain for point solutions that improve workflows, analytics, or automation, but the upside is increasingly fractionalised as software becomes commoditised and defensibility shifts from pure code to distribution and integration. Category leaders are now harder to spot early and easier to displace - the big question is whether SaaS will continue to compound for long-enough to generate outsized outcomes. SaaS will remain valuable, but in most real-world domains, it will be the tooling rather than the core engine of value creation - and its relative ability to capture value has become more uncertain.

There’s also an adjacent vertical integration opportunity in SaaS and knowledge work - the “White-Collar Vertical Integrator”. Just like a Real-World Vertical Integrator, they pull a software layer over a complex, fragmented, manual value chain, with the key difference that the work product is not physical and not a standardised SaaS product, but a professional service - a story for another day.

Some Bodies Along the Way and Why They Failed

Katerra ($2.2bn raised, bankruptcy 2021) - Vertical Integrator in construction: Attempting to industrialise homebuilding is not a novel idea - many have tried and many have failed (also Might Builders, etc.) - they stumbled not because the idea was wrong but because they overintegrated too deep into the raw materials supply chain from day 1, tried to reinvent methods and supply chains simultaneously and/or scaled prematurely, burning capital before reaching cost parity or delivering scale.

Northvolt ($15bn raised, bankruptcy in 2025) - Vertically integrated battery manufacturing: Northvolt scaled prematurely, struggling to produce stable, saleable cells at scale in their flagship gigafactory (scrap rates reportedly reached 50% at times), leading to missed deliveries and lower revenue. Northvolt expected a $1.5bn state-backed loan guarantee from the Swedish Debt Office, which was suspended in late 2024 due to missed expansion milestones, triggering a liquidity crisis that the company couldn’t recover from.