On October 21, I moderated a panel at Future of AI 2026 titled “Can Start-ups Move Faster Than Aerospace and Defense Giants?” My initial answer was: “Of course they can.” Building has never been faster. This was never the problem.

The problem starts after the demo.

Speed no longer sets you apart. AI is the great equalizer. Google’s CEO said in April that about 75% of the company’s new code is now generated by AI, up from roughly 25% in 2024. A team of five can now build what once required an engineering department. The giants have access to the same tools.

Capital has noticed. According to PitchBook, defense-tech venture deal value hit a record $49.1 billion in 2025, up from $27.2 billion in 2024. Israel is part of this wave. The Defense Ministry said that its orders from defense start-ups reached about $330 million in the first half of 2026. This is double the figure from the same period last year.

However, building never slowed this industry down. In July, the US Government Accountability Office reported that the Pentagon’s 104 costliest weapon programs, representing more than $2.4 trillion in planned investment, take over 12 years on average to deliver initial capability. Faster coding does not close that gap. Closing that gap requires faster certification, security review, validation, and accountability.

A visitor speaks on his mobile phone in front of an ad for military AI displaying a drone at the Berlin Security Conference, on November 3, 2026 in Berlin.
A visitor speaks on his mobile phone in front of an ad for military AI displaying a drone at the Berlin Security Conference, on November 3, 2026 in Berlin. (credit: John MACDOUGALL / AFP via Getty Images)

What the giant is actually buying

It’s important to understand that a giant is not buying your model. It is buying reduced risk. In aviation, a software error is not a bug ticket. It is an incident report, or worse. Everyone in the room is asking the same quiet question: If this fails, who answers for it?

A program manager, an engineer, and a CISO each carry part of that answer, and none of them wants to learn your technology at the expense of their career.

The rules are still being written, which makes the question harder.

In November 2025, the European Union Aviation Safety Agency published its first regulatory proposal on AI in aviation, on the trustworthiness of AI systems. It has not been adopted, and a second proposal is planned for 2026. 

Regarding cybersecurity, airborne systems already answer to established standards such as DO-326A/ED-202A. If your product touches an aircraft, you will be asked which standard you are built to meet. “We’re working on it” will not be a good answer.

How to win the demo

Here are some key points on how to win the demo:

First, understand the pain before you walk in. Do not arrive to interview the buyer about their problems. Arrive already understanding them. Read their public tenders, their incident reports, and their annual filings. Open with their problem, in their language.

Bring a working prototype, not a slide deck or a mock-up. A deck or mock-up asks the giant to imagine. A prototype lets it judge. Even a narrow prototype, running on realistic data, beats 20 slides. Also, show the giant how it fails, not only how it works. A team that can explain its edge cases earns more trust than one that only demonstrates the happy path.

Key point. Make the first impression count, because you get one chance. A demo that fails in the room can close a door for years in an industry where the circle of decision-makers is small. Take the time to build something really good before you knock on the door.

Find all three doors. The user feels the problem. The buyer holds the budget. The gatekeeper, often the CISO, DPO (data protection officer), or the security and quality functions, can stop the process. Win the user. However, engage with the buyer and gatekeepers early.

Security and trust come first. Arrive with your security file. Where does your model run? What data does it touch? Who can access it? What happens when it is wrong? A third-party risk assessment can be a great selling point. A young company has no track record. Meticulous documentation has to do what a track record does for an established vendor.

Finally, define success before the demo. Agree on one use case, one metric, one timeline, and what happens if you hit it. Keep in mind that the most expensive words in this industry are, “Let’s see how it goes.”

From demo to deployment

This is where most start-ups get stuck. A pilot that impresses and then goes nowhere is the common outcome. MIT’s NANDA initiative reported in 2025 that about 95% of organizations it studied saw no measurable P&L return from generative AI initiatives.

The cause was not model quality. It was how the technology was woven into how the organization actually works. That study covers enterprises broadly, not defense. However, the pattern will be familiar to anyone who has watched a promising pilot stall.

The same study holds good news for start-ups. Solutions bought from specialized vendors succeeded about 67% of the time, while internal builds succeeded only about a third as often. Buying beats building internally. A giant that insists on building everything itself is playing against the odds. That is your opening, if you can get through the gate.

Even well-designed pipelines leak. The Pentagon’s Defense Innovation Unit made 450 prototype awards between fiscal years 2016 and 2023, and reported that 51% of completed prototypes transitioned to production, according to the GAO. Nearly half did not.

In the Pentagon’s Middle Tier of Acquisition pathway, which is meant to deliver in about five years, GAO found that 18 of 40 programs entering between 2018 and 2025 started with immature technology.

The lesson for founders is to state your maturity honestly, and to say what remains to be done. Overpromising is one of the fastest ways to turn a pilot into a postmortem. Plan the conversion before the pilot begins: who signs the production contract, which budget line it comes from, which certification it uses, and who owns the integration. Try to write the path to deployment into the pilot agreement itself.

Plan your runway for a sales cycle measured in quarters, and potentially years. Dual-use products help here. A commercial version of your technology can earn revenue while the defense process runs its course, and it gives the giant a second reason to take you seriously. You don't depend on them to survive.

The giants have homework too

Closing the gap is a joint effort. Giants should name a single internal champion for every start-up pilot, define what a pilot must prove to move forward, and agree on security and compliance review terms up front.

In August 2026, the Defense Ministry’s director general said the ministry does not see start-ups and large manufacturers as an either-or choice. This is the right framing. The giants need the start-ups’ speed, and the start-ups need the giants’ ability to certify, integrate, and field.

The bottom line is do your homework. The giants will move faster for a start-up that has already done the work: understood the problem, proven the product, answered the security questions, and mapped the road to deployment.

Speed gets you to the meeting. Trust gets you the contract.