Deep Tech Companies Are Built Different

A summary of the structural differences between software and deep tech companies.

For the last decade, capital crowded into software because bits scaled faster than atoms. But across the economy, the bottlenecks are becoming physical again. The next set of generational companies will be the ones that use frontier technologies to remove these bottlenecks in huge markets where demand is urgent, incumbents are brittle, and durable moats can actually form. Some will create entirely new categories; others will drag multibillion-dollar industries out of the last century.

You can already see early signals of this. Data centers are straining the grid and creating urgent demand for new power generation, cooling, transmission, and storage. Defense budgets are shifting toward autonomy, munitions, sensors, shipbuilding, and resilient supply chains. Manufacturing capacity is becoming a strategic asset again, not just a cost center to be outsourced. Biology is moving from lab science into programmable therapeutics and industrial-scale production. Robotics is leaving the demo floor and entering warehouses, factories, farms, hospitals, and battlefields.

But deep tech is not just “harder software.” It is a different game, and founders and investors who understand its rules early will have a real advantage. Companies that remove physical bottlenecks will not be built, financed, or evaluated like software companies, and the key differences show up in three layers:

1. People and path dependence

2. Risk and reversibility

3. Capital and value creation

While many of the constraints below make it harder to start, those same constraints make the winners more defensible, better financed, more valuable, and harder to copy.

Stylistic note: I use ‘deep tech’ and ‘hardware’ somewhat interchangeably below, even though the fit is imperfect. The common thread is that these are atoms-heavy companies: startups where the hard part involves some combination of physical systems, specialized talent, regulation, and manufacturing.

People and Path Dependence

• Deep tech pivots are far more constrained. A software company can pivot from dating to video (YouTube) or from games to productivity software (Slack). But a robotics company can’t pivot to nuclear energy or pharmaceuticals without resetting its entire team. That makes it critical to get the early direction right, but it also creates focus. The company is not trying ten products to see what sticks; it is compounding specialized knowledge, technical progress, and customer relationships around one hard problem.

• Early design decisions can make or break a hardware company. Because pivots are hard, the initial idea has to be directionally correct, and the early design has to be thoughtful. Changing something like the length of a robot arm can affect motors, actuators, batteries, manufacturing processes, and supply-chain decisions. The cost of being wrong can be months or years instead of hours with software. But the reverse is also true: good early decisions compound, and a strong architecture can make the product easier to manufacture, cheaper to service, safer to deploy, and harder for competitors to replicate.

• Founder-market fit is essential for deep tech businesses. Many software products can be built with good software generalists. But deep tech products require specialized talent: electrical engineers, mechanical engineers, turbomachinery experts, etc. Missing a key role can slow things down dramatically. These teams are harder to assemble, but once assembled they become part of the moat. This is a recurring pattern in deep tech: the things that make a company harder to start can also make it harder to compete with later.

• Being in the same room is critical for physical products. Software products can be developed remotely if you have well-defined interfaces between components. After all, as soon as someone in New York updates their code, someone in LA can build on top of the new code immediately. Hardware is different. Components have to be developed and tested together, which requires physical proximity — otherwise your iteration speed declines dramatically.
  
  This is why we have clusters like space startups in LA and biotech startups in Boston: these cities have enough local talent density to make building in-person easier in specific categories. These clusters are an advantage because they concentrate talent, suppliers, advisors, customers, and repeat founders in ways that accelerate company progress.

• Deep tech talent is scarcer but more mission-driven. There are approximately four million software engineers in the US, but only 20k RF engineers, 10k nuclear engineers, and 2500 turbomachinery engineers. If you want to hire a top 5% engineer, it’s much harder in hardware because you’re often looking at pools of hundreds or low thousands of people across the entire country. The silver lining is that fewer companies are competing for these candidates than for software engineers, and deep tech companies frequently have inspiring missions that make it easier to recruit specialized engineers from slow-moving incumbents.

• Deep tech companies need different early employees. A software startup can hire generalist full-stack engineers early on and add specialists later. Deep tech companies usually need an extremely specific early team. You don’t hire a generalist biologist or electrical engineer; you hire someone with direct, hard-earned experience in the specific therapeutic pathway, actuator design, RF system, or manufacturing process you’re focused on.

Risk and Reversibility

• Market risk kills software companies; technical risk kills hardware companies. Software products are relatively straightforward to build, but it’s very easy to create something that there’s no market for. I was a software engineer for a decade and don’t mean to imply that software is trivial, but that very few software companies fail because they couldn’t build a marketing tool or mobile app or recommendation engine.
  
  Hardware often has clearer demand at the category level — of course people want cheaper energy or cancer-killing drugs — but much higher technical risk. The market may be obvious in the abstract, while the hard questions are whether the product can be made to work, manufactured economically, deployed safely, and sold through the right channels.
  
  This sounds scary, but technical risk is often more legible than market risk. You can test whether a motor hits a torque target, whether a battery reaches a cost curve, or whether a therapeutic kills a cell type. Deep tech risk can be high, but it can also be attacked with explicit milestones.

• Regulatory friction is the status quo in deep tech, but it cuts both ways. Outside of a few areas like fintech and healthcare, most software isn’t regulated. If you want to build a new CRM, you can just launch it as soon as you’re ready. But most hardware companies face oversight: permitting for facilities, safety certifications, environmental reviews, regulatory approvals, and so on. That makes budgeting and planning harder. But once you clear the hurdle, competitors still have to slog through it. In deep tech, regulatory progress (e.g. FDA approvals for drugs, FAA approvals for drone operations) can lead to major value inflection points.

• Good judgment is worth more in deep tech. A software company can quickly adopt a new database or UI toolkit, but hardware changes can affect manufacturing tools, supply-chain partners, certification paths, and customer deployments. Bad judgment might cost days or weeks in software, but months or years in hardware. That makes the best founders even more valuable: the opportunity is not to avoid hard decisions, but to make critical decisions better than everyone else.

• Deep tech progress is discrete, not continuous. With software, you launch a product and revenue ramps gradually. You might have a goal for your next round like $3M ARR, and success can be measured on a spectrum: you could do okay and only get to $1.5M ARR, do well and hit $3M, or really outperform and hit $5M. Deep tech milestones tend to be more binary: you have a goal like being able to sustain a nuclear reaction or being able to kill some type of tumor cell, and either you are able to do that or you’re not. There are no “we got it half-working” milestones because half-working still doesn’t work. This can make progress feel lumpier, but it also makes risk retirement clearer. When a company tackles a key risk successfully, its valuation can change dramatically because a major uncertainty has disappeared.

• Software has low barriers to entry but weaker moats; hardware has high barriers to entry but more durable moats. Software companies can ship quickly, and many traditional moats are weak or eroding (economies of scale are modest, distribution channels are rarely proprietary, and AI tools make feature replication easier). This means companies can start and grow quickly, but can stall out or shrink just as quickly.
  
  Hardware is the opposite: it often takes two to five years for a deep tech company to build a product that’s commercially ready, which is a big, expensive barrier to entry. But if you can build a sellable product, the moats are very strong: economies of scale are very real, brand and track record are valuable (because given the choice, everyone would rather buy from a company selling its 100th nuclear reactor than from a company selling its first), regulatory friction keeps competitors from catching up quickly, etc. So deep tech companies are much harder to build at first, but the same barriers that slow the company down early can protect it and help it dominate later.

Capital and Value Creation

• Software products typically sell for tens or hundreds of thousands; hardware products sell for millions or tens of millions. Power plants cost millions or billions, large industrial automation systems cost tens of millions, and advanced therapeutics cost $1M/treatment. When your sale price is that high, the structure of your company changes. A software company might spend 30% of its revenue on sales and marketing ($30K CAC for a $100K contract), but a deep tech company can spend less on sales and more on R&D (because a $100M contract doesn’t require anywhere near $30M for sales and marketing).
  
  As a result, hardware companies are more capital intensive upfront when they are proving out technical viability while software companies are more capital intensive later on when they are paying for customer acquisition at scale. Overall capital intensity is similar across the two categories over time, just more front-loaded for hardware and more back-loaded for software. And because of the large contract sizes, a few large customers or deployments can support a lot of company value.

• Deep tech companies have a different financing stack. Software companies can raise equity, venture debt, or revenue-based financing once they have a few million in revenue. Deep tech companies rarely have revenue-based financing in their early years, but they can get equipment financing, inventory financing, project financing, and government grants and contracts. Used well, those tools can save founders a lot of dilution and reserve equity dollars for the highest-risk, highest-upside parts of the business. For example, you can often get hundreds of thousands or even low millions of dollars in equipment financing even at the pre-seed stage, because the lender is underwriting your equipment instead of your progress. (I.e. if you want to buy a $1M machine that can be resold for $900K if you go out of business, you can probably get a good rate for that purchase.)

• Deep tech financing is milestone-based, not metrics-based. A software company raises successive rounds based on its revenue, retention, and usage metrics. Deep tech companies raise based on hitting big milestones and removing key risks: can you build each subsystem? A full prototype? Scale up manufacturing from hand-assembled to partially automated to fully automated? In deep tech, valuations often inflect when major risks disappear. In software, valuations move more continuously with traction.

• The reward for getting deep tech right is much stronger value capture. Software companies face constant competition. Harvey built a great AI product for legal teams. Shortly after, Legora emerged with a competing product and stole a lot of the limelight. Last month Mike — an open source alternative created by one person — came out and already has thousands of GitHub stars. Most valuable software categories see a similar arc of increasing, intense competition. Hardware is different. No one is building a commercially viable nuclear reactor or cancer-killing virus in their garage. It’s just too capital-, team-, and time-intensive to do that. That means if you have a very capable team and are able to tackle a big problem, you can capture a lot of value and have a bigger chance to dominate your market in the long run. Just consider companies like Nvidia (GPUs), SpaceX (space launches), or Illumina (gene sequencers).
  
  The result is a different return profile: harder early company formation, but a better chance that the winner becomes structurally important, difficult to displace, and valuable for a long time.

The “hardware is hard” cliché is a useful warning, but not a useful framework. The better framing is that building in bits and building in atoms operate under different constraints. In deep tech, those constraints can become the source of defensibility.

Software companies usually win by moving quickly through product and distribution uncertainty. Deep tech companies usually win by making a smaller number of harder, less reversible decisions correctly: assembling the right specialized team, choosing the right architecture, retiring the right technical risks, navigating regulation, financing the company with the right mix of capital, and turning their technical progress into a durable moat.

For founders, this means deep tech rewards judgment more than motion. The best founders do not just move quickly; they know where speed matters and where patience is mandatory. They understand which early decisions will echo for years, which risks must be retired before others, and which milestones actually change the company’s odds of success. In software, progress often compounds through iteration. In deep tech, progress compounds through a smaller number of hard decisions made correctly.

For investors, this means asking different diligence questions. In software, the early questions are usually about user love, retention, growth, and GTM motion. In deep tech, the first questions are more foundational: is this the right team for this problem, are the early architecture choices sound, does each milestone retire a meaningful risk, is there a plausible path through regulation and manufacturing, and is the financing plan matched to the development path?

The best deep tech companies will not look like software companies with slower sales cycles. They will have different markers of quality: exceptional founder-market fit, clear technical inflection points, urgent customer demand, and the ability to turn manufacturing, regulation, performance, and supply chains into durable moats.

The key is that deep tech is not just harder. The difficulty is part of the opportunity. Scarce talent, regulatory friction, manufacturing complexity, technical risk, and slower iteration are real obstacles, but they also keep casual competitors out and give the winners a chance to become structurally important companies. When a company solves a hard physical bottleneck, it does not just capture demand; it can reshape an entire industry around itself.

The opportunity today is not to apply the software investing playbook to hardware. The opportunity is to understand the structural differences of the physical economy before everyone else does.

Thank you to Anna-Sofia Lesiv and Julian Shapiro for feedback and advice on this post.