Deep Tech & AI Infrastructure

The best investments often share a structural similarity even when the surface details look completely different.

Both companies here are building infrastructure. Both are led by second-time founders with successful exits. Both are positioned at inflection points where macro tailwinds meet execution capability. And both require conviction in category creation rather than category capture.

The bet isn't "will this company win an existing market?" It's "will this market exist at scale, and will this team define it?"

Different risk profiles. Different return paths. Same underlying question.

Isembard: Manufacturing for the Reshoring Era

Isembard – Stage: Seed | HQ: London | Sector: Deep Tech / Manufacturing

Building the manufacturing backbone for Western critical industries through MasonOS, a software platform powering a network of highly automated precision factories.

Here's the macro setup: Europe's precision manufacturing sector is facing a demographic crisis. More than a third of UK machinists are over 50. The majority of employers struggle to find skilled labour. Meanwhile, reshoring demand is surging – $650B in planned UK manufacturing investment alone, with 93% of aerospace and defence executives citing national security as their primary motivation.

Skilled labour exiting. Demand surging. Policy tailwinds supporting domestic production. That's a narrow window for technology-enabled manufacturers to capture share from fragmented incumbents.

Isembard's model is distinctive. Unlike Hadrian in the US ($500M+ raised for centralised gigafactories), Isembard uses a franchise-based approach. Lower capex per site. Faster geographic expansion. The Texas factory was debt-funded and is reportedly already profitable.

Alexander Fitzgerald is a second-time founder with an unusual background. After HM Treasury policy work, he co-founded Cuckoo – a challenger broadband company that reached tens of thousands of customers and was voted UK's #1 broadband provider for customer service by MoneySavingExpert. Twice. Cuckoo was acquired by Giganet in September 2022.

But the real origin story comes from Fitzgerald's other role: military reservist with the UK Ministry of Defence since 2016. It was there he experienced 6-8 week waits just to receive quotes for precision parts. That firsthand frustration became the catalyst.

The team fuses "bits and atoms" expertise: ex-Palantir engineers for AI and data platforms, former Rolls-Royce and 3M engineers for industrial hardware.

The execution velocity is notable. Founded October 2024. First factory operational January 2025. First franchise September 2025. US expansion July 2025. That's with 12 employees and seed capital.

The Bear Case

Hardware businesses are hard. Precision manufacturing requires significant operational expertise, equipment maintenance, and quality control. Manufacturing tolerances are unforgiving – unlike broadband, a single quality failure in a flight-critical component could be catastrophic for reputation and customer relationships.

Scaling through franchisees introduces variability risk. AS9100 and ITAR certifications are table stakes for defence contracts, and maintaining those standards across distributed sites is non-trivial.

Hadrian is well-funded and has a different but compelling model. If they execute well, they could dominate the US market and potentially expand to Europe.

What I'd Want to Know

The open question is whether the franchise model creates defensibility or fragility. Centralised manufacturing compounds quality reputation. Distributed manufacturing compounds geographic reach. The trade-off defines the ceiling.

Fitzgerald has demonstrated ability to scale distributed operations while maintaining service excellence – Cuckoo's customer service ratings during rapid growth suggest disciplined operational DNA. The team's Palantir background points to sophisticated data-driven quality monitoring via MasonOS, which centralises production data across all sites.

The strategic investor alignment matters too. Notion Capital is backed by the UK's National Security Strategic Investment Fund and German Federal Government. That's policy-level support for the company's national security mission.

I would revisit the thesis if: franchise expansion stalls beyond current sites, major quality incidents emerge, or the company fails to secure at least one prime customer relationship within 18 months.

Verdict: Bullish. This is a bet on a proven operator building domestic manufacturing capacity for sectors the UK government has designated strategically important. The risks of a hardware-intensive business are real. But Fitzgerald's execution track record and the macro tailwinds create a compelling setup.

Paid: Monetisation Infrastructure for AI Agents

Paid – Stage: Seed | HQ: London | Sector: AI / SaaS

Building the monetisation infrastructure for the AI agent economy – enabling results-based billing for companies whose products eliminate the seats that traditional SaaS pricing assumes.

Traditional SaaS billing fundamentally breaks for AI agents.

Per-seat pricing fails when agents eliminate seats. Per-API-call pricing doesn't capture outcome value. Agent makers face variable LLM costs while customers want to pay for results, not tokens. This creates a structural gap that existing billing platforms – Stripe Billing, Zuora, Metronome, Orb – aren't designed to solve. They meter usage but don't track outcomes or demonstrate value.

Paid is building purpose-built infrastructure for this transition. Results-based billing. Outcome tracking. Value demonstration. The economic plumbing for companies selling AI agents.

The founder is Manny Medina, and his pedigree is exceptional. He co-founded Outreach in 2014 after pivoting from a recruitment startup that was two months from bankruptcy. That pivot produced the leading sales engagement platform: zero to $250M+ ARR, 6,000+ customers, 800 employees, $4.4B valuation at peak. Before that, he was employee #3 on Amazon's AWS team and led Microsoft's mobile division from launch to $50M revenue.

"They didn't really know what to charge." – Medina, on what he heard from dozens of agentic AI startups

This isn't theoretical market research. Medina lived the monetisation problem at Outreach – he watched customers struggle to price software that delivered variable value. After stepping down as CEO in September 2024, he spent months talking to AI startups and found the same pain point everywhere.

The team he assembled reflects the problem: Manoj Ganapathy (founded Invoice IT, acquired by Salesforce; former Head of Product for Salesforce Billing), Raj Dosanjh (YC alum, early Palantir), and Arnon Shimoni (monetisation lead at Pleo and Storytel).

They built the initial platform in one month with two engineers using AI-native development tools. Early customers include IFS, Artisan, 11x, Logic.app, and HappyRobot. The company reports 20-40% revenue increases for customers after implementation.

The Bear Case

Market timing is the primary kill risk. AI agent adoption remains early-stage. If enterprise agent deployment stalls or takes longer than projected, Paid's TAM shrinks accordingly. The platform's value proposition depends on agents achieving meaningful production scale.

Competitive response is real. Stripe, existing billing platforms, and ERP vendors could build agent-specific billing features. Stripe's integration of usage-based billing for OpenAI demonstrates willingness to serve AI customers.

And the valuation is aggressive. $100M+ at seed with no Series A. That's justified by founder pedigree and syndicate quality – Sequoia, Lightspeed, and EQT Ventures all invested pre-Series A – but it creates high expectations for metrics and limits upside if growth is slower than projected.

What I'd Want to Know

The question I'd want answered: does Paid's data layer become a moat, or does billing remain a commodity once outcomes are standardised? If every AI company converges on similar pricing models, the infrastructure becomes interchangeable. If Paid's outcome-tracking creates proprietary benchmarks, they own the category.

Medina has demonstrated the ability to create a category rather than follow one. Outreach defined sales engagement software. He built it during the early, uncertain days of that market. The pattern recognition from that experience – knowing when a category is about to tip – is exactly what you want in a founder betting on nascent infrastructure.

The early integration partnerships matter: Vercel, ElevenLabs, LangChain, LlamaIndex. These create switching costs and ecosystem lock-in before competitors can respond.

I would revisit the thesis if: enterprise AI agent adoption materially slows through 2026, a major competitor launches with meaningful traction, or early customer metrics show poor retention.

Verdict: Bullish. This is a conviction bet on a category-defining founder building nascent infrastructure for the next era of software. The cap table quality provides independent validation. The risk is market timing. The upside is defining how AI companies monetise.

The Connection

These companies look nothing alike on the surface. One manufactures precision parts for aerospace. One builds billing software for AI agents. Different sectors, different risk profiles, different paths to scale.

But the structural similarities are worth noting. Both Fitzgerald and Medina have built, scaled, and sold companies. They've seen what breaks at scale and know how to avoid it. Neither is competing for share in an existing market – both are betting that a new category will emerge and that they'll define it. Reshoring demand for Isembard, AI agent adoption for Paid. Both are positioned at inflection points where structural shifts create demand for new infrastructure.

The syndicate quality matters too. Notion Capital's government backing for Isembard. Sequoia, Lightspeed, and EQT for Paid. Sophisticated investors with relevant expertise have conviction.

The question isn't whether these are good companies. The question is whether the categories they're creating will materialise at scale, and whether they'll capture them when they do.

That's the bet: back proven operators building foundational infrastructure at category-defining moments.