MMC Ventures

The boom in AI cost-cutting startups may just be a 𝑡𝑜𝑘𝑒𝑛 gesture That's the wickedly witty title of Jacob Robbins' latest article for PitchBook, which I enjoyed contributing to. It's a very timely piece on whether AI cost-cutting startups are building real moats. In our recent research on Tokenomics at MMC Ventures, we observed that AI cost control tools are being built into a variety of existing products (FinOps, developer productivity tools, observability solutions, routers/gateways...) so this is rapidly turning into a feature within a broader suite rather than a standalone product. That's why we are much more interested in the underlying infrastructure layer - such as context and memory management, inference optimisation and intelligent routing. If you're building in this space, we'd love to chat! (Links to the PitchBook feature and our report in the comments below 👇)

Enjoyed contributing to S&P Global's latest analysis on GenAI funding trends. One point I keep coming back to: as AI moves into production, token efficiency is becoming as important as model capability. At MMC Ventures, we found that 50-80% of token spend is completely unnecessary. You don't need frontier intelligence for every task, your model is frequently processing (and re-processing) superfluous context, and you may be spending more tokens on verification than on actual generation. Happily, there are fixes for this. We explored this theme in more depth (link in comments) in our recent AI Tokenomics research, where we discuss how to tokenmin while ROImaxxing. Thanks Iuri Struta for including my perspectives in your latest report! It's an excellent overview of the current landscape.

MMC x TUM.ai Dinner 🇩🇪 While in Munich, we couldn’t miss the opportunity to co-host a dinner with TUM.ai and meet some of the standout builders coming out of one of Europe’s top universities. And what a group. Having talked about the talent density in Munich in our previous post, this dinner really brought that to life. The team came away genuinely energised by the calibre, ambition, and technical depth around the table. Huge thank you to Jonas von Seggern and the TUM.ai team, we're excited for more get-togethers like this in the future!

Munich continues to stand out as one of Europe’s strongest technical ecosystems. 🇩🇪 Last week, we headed to Bavaria to host a VC breakfast, bringing together a group of investors for a morning of coffees and conversation on what they’re seeing in the market and how Munich’s founder ecosystem is evolving. Across the conversations, one thing really stood out: the density of local talent in Munich. The city continues to combine deep technical talent with real industrial depth, particularly across AI, automation, robotics, climate, and enterprise software. Founders are building close to research and are close to real customer problems from day one. We’re continuing to spend more time across DACH as we look to meet impressive founders building category-defining companies from the region. A big thank you to Trinity Capital for supporting our trip and to everyone who joined Mina Samaan, Sevi U. & Carla Boichon! Stockholm visit pending...🇸🇪

We’re excited to back Nexcade as it raises a $6M seed round, led by Project A, alongside Connect Ventures and Entropy Industrial Capital! Freight forwarding still runs on a messy layer of inboxes, spreadsheets, rate data and judgement calls, the “shadow TMS” is where the real work happens. Nexcade is building AI agents for that layer. Agents that pull context together, run routine steps, support quotes, tenders and bookings, and bring operators in when it matters. What stood out to us was the team’s understanding of the forwarding desk: purpose-built agents that map to the way the work actually gets done. Proud to partner with Dan Bailey & Tasho Kjosev as they build the AI operating layer for freight forwarding.

𝗔𝗜 𝗧𝗼𝗸𝗲𝗻𝗼𝗺𝗶𝗰𝘀: 𝗛𝗼𝘄 𝘁𝗼 𝘁𝗼𝗸𝗲𝗻𝗺𝗶𝗻 𝘄𝗵𝗶𝗹𝗲 𝗥𝗢𝗜𝗺𝗮𝘅𝘅𝗶𝗻𝗴 AI has entered its usage-based era. As AI-native startups and enterprises deploy agents at scale, token costs are becoming one of the fastest-growing line items in technology budgets. But the answer isn't simply "use fewer tokens," it's how efficiently you convert AI spend into business outcomes. In this report, Advika Jalan, Simon Menashy and Prakriti R. break down: 🔍 The hidden costs most teams overlook (e.g. input tokens rather than output tokens drive the overall cost of agentic coding tasks, or that 60% of the primary cost of agentic software engineering isn’t in initial code generation but in automated refinement and verification) ⚖️ A practical framework for reducing token spend without sacrificing quality (covering everything from workflow design to spend controls that factor in actual AI adoption patterns). 🕸️ The emerging stack for AI cost optimisation: context and memory management, open source models, model routers/gateways, inference providers, as well as cost control and monetisation solutions 🔐 The reasons why security, governance and sovereignty belong in every token optimisation conversation Whether you're building AI-native products or rolling out hundreds of agents across the enterprise, token economics is quickly becoming a competitive advantage. Read the full report: https://lnkd.in/eJAHCMfi

🇩🇪 We're heading to Munich! 🇩🇪 Next Wednesday (July 2nd), we're hosting a breakfast for investors in Munich. We loved getting everyone together in Berlin back in April, so we're back at it, only this time in Bavaria. Spaces are limited, so if you'd like to join Mina Samaan, Sevi U. & Carla Boichon, sign up via the luma link in the comments 🥐 Hope to see you there!

We're excited to announce our investment in Frontier Health as it raises $16 million in seed funding! Healthcare AI grabs headlines for clinical breakthroughs, but some of the highest-impact opportunities lie in the administrative backbone keeping patients moving through care. Last year we published Beyond the Clinic: How GenAI is Reimagining Healthcare Operations, exploring how AI-native startups are transforming non-clinical healthcare workflows. We identified care operations as one of the most compelling areas: solutions that don't just save time, but improve patient outcomes by helping teams proactively manage care pathways. Frontier Health is building exactly this. Founded by Rachel Finegold (who spent six years as healthcare lead at Palantir Technologies across 40+ NHS hospitals) the team has built JUNO, a supportive AI teammate for healthcare admin staff. JUNO integrates into existing systems, handling routine tasks, flagging delays, and keeping patients moving safely through care, with no infrastructure changes required. The results speak for themselves: at East Sussex Healthcare NHS Trust, JUNO saved 221 days of staff time in just eight weeks and reduced median patient pathway time by 22%. As health systems face growing demand with constrained resources, we believe supportive AI will become critical infrastructure for healthcare delivery. Proud to back Rachel and the Frontier Health team alongside Atomico, firstminute capital, XYZ Venture Capital, Tenacity Capital & Avenir. Huge credit to Sevi U. and Charlotte Barttelot who led our healthcare research - links to the full research and announcement in comments. 🏥

After 30+ conversations with CISOs, practitioners, and founders, Jamie Shelton, Carla Boichon and Mina Samaan mapped where the next wave of cybersecurity value is emerging: → Runtime Application Security for agents → Contextual Agentic Security → Structural, infra-led approaches that make bad actions architecturally impossible at the kernel level For founders building in cybersecurity, this research highlights where buyers are feeling pain today, and where new categories are forming. If you're building, buying, or securing agentic systems, we'd love to hear from you. 👇 Link to the report in comments