Founders: Han Wang, Hahnbee Lee
Sector: Developer Infrastructure
Location: San Francisco, CA
The Opportunity
AI agents are increasingly taking action on behalf of humans: writing code, calling APIs, answering customer questions, and orchestrating multi-step workflows across software systems. But agents don’t operate on intuition. They operate on context. To reason correctly, plan effectively, and take safe action, agents require structured, trusted, and continuously updated knowledge about the systems they interact with. That knowledge layer is documentation.
Documentation has always been the system of record for how software works. What’s changed is who’s reading it. Mintlify recently analyzed 30 days of traffic across their platform and found that AI coding agents now account for 45% of all documentation requests — nearly tied with traditional browser traffic at 46%. Claude Code alone generated 199.4 million requests in that period. A single AI coding agent is now pulling more documentation than one of the most popular browser and OS combinations in the world. As one customer put it: if a coding assistant doesn’t know about you as a developer tool, you don’t exist.
This shift makes the state of most documentation a genuine strategic liability. Most technical docs are written manually, live in static sites disconnected from the codebase, and fall out of date the moment a new feature ships. The explosion of vibe coding tools has exacerbated the issue: source code is being generated faster than ever, widening the gap between what a product does and what its documentation says it does. Teams patch the problem by standing up Docusaurus instances, or stitching together pipelines from GitHub and a handful of LLM APIs. However, these approaches can cost hundreds of engineering hours and tens of millions of dollars annually, and the output is still a static artifact that agents struggle to reliably parse and reason over.
The companies that will win in an agentic world are the ones whose knowledge infrastructure is accurate, structured, and machine-readable, updating continuously as their product evolves.
The Solution
Mintlify is an AI-native documentation platform that helps developer teams create, maintain, and serve intelligent documentation that scales with their product — for both human and AI consumers.
The platform includes:
Docs-as-code workflow: Mintlify lives natively in the GitHub-based workflows developers already use. Teams can publish beautifully designed, production-ready documentation sites in minutes, without managing their own hosting, theming, search, or deployment infrastructure.
AI-powered authoring and continuous updating: Mintlify drafts documentation from existing codebases and APIs, and automatically detects code changes to propose documentation updates, keeping docs in sync with the product without manual effort.
AI Assistant: Every Mintlify docs site comes with built-in semantic search and an in-docs AI assistant, allowing both humans and AI agents to find answers instantly.
Agent-ready infrastructure: Mintlify hosts an MCP server for each docs instance, enabling tools like Cursor or customer support agents to query documentation in real time. They also generate llm.txt and llm.json files to help LLMs index and reason over documentation accurately, and support skill.md files that tell coding agents exactly how to interact with a product.
Expanding beyond public docs: Enterprises are increasingly using Mintlify to replace tools like Confluence for internal knowledge infrastructure — powering engineering onboarding, customer support, and other use cases where accurate, agent-readable context matters.
The result is a living, continuously updated knowledge layer that serves as the canonical source of truth for how a product works, for every human and agent that needs to know. Critically, humans remain in the loop throughout: when Mintlify’s AI detects a code change and proposes a documentation update, engineers review and approve it before it goes live. This keeps the velocity of automated updating without sacrificing the accuracy that both human readers and AI agents depend on.
Why We’re Backing Mintlify
The market for AI-native documentation is real and accelerating. Half of Mintlify’s monthly active viewership today is already AI agents, and the company expects that to reach 90% in 2026. As AI assistants increasingly mediate how developers discover and evaluate tools, documentation quality has become a direct driver of distribution. Companies like Vercel, Coinbase, and PayPal have internalized this, using Mintlify not just for their public developer docs, but for internal knowledge bases and AI-facing content infrastructure.
What’s more, Mintlify’s founders are exceptional. Han Wang and Hahnbee Lee built Mintlify in response to their own struggles with documentation infrastructure while building products. They’ve created the platform they wished had existed for them, and they’ve assembled a world-class team to support them.
What’s Ahead
Mintlify’s long-term vision extends well beyond documentation tooling. The company is positioning itself as the canonical knowledge layer for the enterprise — the source of truth that powers coding agents, support agents, and internal workflow agents across the entire organization. It’s an ambitious vision, and one we believe they are uniquely positioned to execute given the distribution moat they’ve already built.
We’re proud to partner with Han and Hahnbee as they scale Mintlify into the default platform for how software knowledge is created, maintained, and consumed. And we’re thrilled to co-lead Mintlify’s Series B round alongside a16z.
Welcome to Salesforce Ventures, Mintlify!
]]>
Last month, Alphabet’s acquisition of Wiz officially closed at a $32B valuation — one of the largest technology acquisitions in history and one of the most significant exits in SFV’s history.
We first invested in Wiz’s Series B in May 2021 and continued backing the company through their Series C, D, and E. When we met founders Assaf Rappaport, Ami Luttwak, Roy Reznik, and Yinon Costica, we saw something rare: a team with deep expertise on both sides of cloud security, a product that could be deployed in hours, and a pace of execution unlike anything we’d encountered. Wiz became the fastest cybersecurity company ever to reach $100M in ARR, and then the fastest to $200M.
Wiz’s outcome is a testament to what happens when exceptional founders meet a massive market opportunity and execute with precision and speed. Congratulations to the entire Wiz team. On our blog, we wrote a bit more about why we believed in Wiz and what we believe made this outcome possible.
The Next Generation of Cloud Security
Upwind is redefining cloud security with a runtime-first approach that uses eBPF technology to gain deep, kernel-level visibility into running workloads, filtering out 95% of alert noise by determining which vulnerabilities are actually exploitable in production.
After investing in Upwind’s $250M Series B, our partner Nowi Kallen sat down with Upwind co-founder and CEO Amiram Shachar for a wide-ranging interview.
The Operating System for Modern Legal Work
Legora is an AI-native workspace that automates time-intensive legal workflows — such as contract review, research, and drafting — for both law firms and in-house teams, integrating with tools like Microsoft Word and SharePoint.
Watch: 2025 Year in Review
After publishing our annual Year in Review report, Salesforce Ventures’ Partner Nowi Kallen and Head of Finance Sarah Browne sat down for a conversation on the key takeaways and insights from the year that was.
Shaping What Comes Next
We recently hosted a group of portfolio CEOs for a series of conversations focused on AI and company building. Attendees were treated to a fireside chat between Verity Harding (Author and AI Policy Expert) and Margaret Taylor (VP of Public Affairs and Strategic Relations at Salesforce) on the intersection of technology and global policy, as well as a panel on the future of robotics with CEOs Lindon Gao (Dyna), Tony Zhao (Sunday Robotics), and Salesforce Ventures Managing Director, Emily Zhao.
SFV IRL
Where to find our team in April:
April 6-9: HumanX (San Francisco)
April 15-16: TDX (San Francisco)
April 18-26: Climate Week (San Francisco)
Thanks for reading! If you’re a founder building in our core investment areas, submit an investment inquiry to get in touch. To learn more about Salesforce Ventures, visit our website or subscribe to our Substack.
]]>
Founders: Amiram Shachar, Tal Zur, Lavi Ferdman
Sector: Cybersecurity
Location: San Francisco, CA / Tel Aviv, Israel
The Opportunity
The history of modern cloud security has been defined by three distinct eras.
The Visibility Era (2018-2020): As enterprises “lifted and shifted” workloads to the cloud, the primary challenge was simply seeing what was out there. This era gave rise to the first generation of Cloud Security Posture Management (CSPM) tools — scanners that inventoried assets and checked for basic misconfigurations.
The “Shift Left” & Agentless Era (2021-2023): As cloud adoption matured, friction became the enemy. DevOps teams rejected heavy agents that slowed down deployments. The market swung toward “agentless” scanning — taking snapshots of disk volumes via APIs to find vulnerabilities without installing software. While this solved the friction problem, it created a new one: noise. These tools flagged every theoretical vulnerability, leading to massive alert fatigue.
The Runtime & AI Era (2024 – Present): In the current third phase, near real-time visibility is no longer sufficient. Modern cloud-native environments are ephemeral, complex, and increasingly driven by AI. In this era, security must move from “outside-in” observation to “inside-out” understanding.
Against this backdrop, strategic value is consolidating in the Cloud-Native Application Protection Platform (CNAPP) category — and Upwind has quickly distinguished itself as a technology leader in the space.
The Solution
Upwind is redefining the CNAPP market with a runtime-first approach. Instead of relying solely on static API scans, Upwind leverages Extended Berkeley Packet Filter (eBPF) technology to gain deep, kernel-level visibility into running workloads without the heavy performance penalty of legacy agents.
This “inside-out” architecture allows Upwind to do something that static scanners cannot: determine reachability.
By correlating build-time data with real-time runtime context, Upwind can mathematically prove which vulnerabilities are actually exploitable in a production environment. If a vulnerable library is present on a disk but never loaded into memory, Upwind knows it’s not an immediate threat. This context allows the platform to filter out approximately 95% of alert noise, empowering security teams to focus on the risks that actually matter.
Beyond vulnerability prioritization, Upwind unifies critical security pillars — CSPM, Cloud Workload Protection, Cloud Detection and Response, API Security, and Identity — into a single, coherent platform. It provides a real-time map of network topology and data flows, enabling organizations to secure everything from containers and serverless functions to the rapidly expanding attack surface of AI models and pipelines.
Why We’re Backing Upwind
Our conviction in Upwind starts with the team. CEO Amiram Shachar and co-founders Tal Zur, and Lavi Ferdman bring rare, proven DNA in cloud infrastructure. We knew the team from their time building and scaling Spot.io (acquired by NetApp), where they optimized cloud infrastructure for thousands of enterprises. We stayed in touch because we could sense their great chemistry and alignment when it comes to tech and company building. That experience shows up clearly in Upwind’s product: a security platform built by engineers who deeply understand cloud compute, Kubernetes, and modern DevOps workflows — and who know that security at scale must be frictionless, precise, and native to how infrastructure actually runs.
That foundation matters even more as AI reshapes the cloud. AI is expanding the attack surface, introducing new risks such as model theft, data poisoning, and emerging AI-SPM requirements. At the same time, AI workloads are inherently dynamic and resource-intensive, rendering static, once-a-day snapshots ineffective. Securing modern applications now requires runtime visibility that moves at the speed of the application itself.
We believe this shift will concentrate value in the CNAPP category, which is growing nearly 30% annually and is projected to reach $18.79B by 2029. The next category-defining company will be the one that closes the runtime gap in this market — and Upwind is uniquely positioned to do exactly that.
What’s Ahead?
We’re excited to partner with Upwind on their $250M Series B as they continue to push the boundaries of what’s possible in cloud security. The Upwind team’s energy — combined with their technical chops — is unmatched, and we could not be more excited to work with them.
As AI adoption accelerates and cloud environments become even more dynamic, the need for deep runtime visibility will only grow. Upwind is perfectly positioned to become the system of record for securing these modern applications, giving CISOs and DevOps teams the confidence to innovate faster. We’re thrilled to be partnered in this journey.
]]>
Salesforce Ventures is thrilled to celebrate a monumental milestone: Alphabet’s acquisition of Wiz has officially closed. At an aggregate valuation of $32 billion, this marks one of the largest technology acquisitions in history and one of the most significant exit outcomes in Salesforce Ventures’ history.
We’ve been proud partners to Wiz since May 2021, when we first made a significant investment in their Series B. Over the past four years, we continued to back the company through their Series C, D, and E rounds. Today’s close validates our conviction in backing category-defining enterprise technology companies early and supporting them throughout their journey.
Betting on the Team
When we first met Assaf Rappaport and Ami Luttwak in 2021, they were building something ambitious: an agentless cloud security platform that would redefine how enterprises protect their infrastructure. It wasn’t obvious at the time. The cloud security market was crowded with well-funded competitors and established incumbents. Ultimately, what set Wiz apart was the team.
Assaf, Ami, Roy Reznik, and Yinon Costica brought rare, proven DNA. After serving in Israel’s elite Unit 8200, they built and sold Adallom to Microsoft for several hundred million dollars within three years. They then scaled Azure Cloud Security to over $1 billion in revenue inside Microsoft. They had experience with cloud security on both sides — as a vendor and as a cloud provider — giving them a unique understanding of the problem.
The feedback we heard during diligence was unlike anything we’d encountered before. The team’s credibility in the industry was so strong that Fortune 100 CISOs told us they’d buy from Wiz sight unseen. Former employees of Assaf said they’d work for him again without even asking what the job was.
This was 2021, a white-hot year for venture capital. Deals were closing in days. Assaf took a different approach. Even as the round moved fast, he told us he didn’t care about timing, just about the people joining Wiz on this journey. He said we should feel free to ask for any data or ask any questions, and they would answer everything. His philosophy: no one should join without knowing what they’re getting into.
The Wiz team was wildly ambitious and aggressive in execution. But low ego, down-to-earth, and composed as human beings. It was true N-of-1 behavior.
Product Velocity Like We’ve Never Seen
Wiz didn’t just have a great team. They executed at a pace that redefined what’s possible in the enterprise.
Within one year of incorporation, Wiz had a full-fledged enterprise-grade product. Within 15 months, they were selling seven-figure contracts to Fortune 500 companies. When our investment in Wiz closed in May of 2021, the company was doing under $5M in ARR. Their plan was to 4x by the end of the year. Instead, they more than 20x’d.
Wiz would go on to become the fastest cybersecurity company ever to reach $100 million in ARR — and then the fastest to $200 million, a pace of growth unheard of in cybersecurity. They beat every expectation we had from day one.
What made this possible? Wiz established key innovations that changed the game: agentless architecture, graph-based tools, and cloud security posture management. Their API-based platform provided unmatched visibility and protection across multi-cloud environments. Their research team became second to none, publishing critical vulnerabilities regularly and educating the market. This research focus became a true differentiator that compounded over time, flowing through the product and building unmatched trust with CISOs.
The feedback we heard from CISOs in our network was incredibly strong: Wiz identified more risks, provided richer and more granular data, and moved faster than the competition.
Further, Wiz entered the market at the perfect time. COVID accelerated cloud migrations far beyond analyst predictions. Wiz launched with an agentless approach at the exact moment when the cloud market was exploding and customers didn’t trust agents. The product could be deployed in hours and deliver value in days. That unlock enabled them to become a clear leader.
As Wiz expanded internationally from its headquarters in Israel, our team’s tailored introductions helped Wiz break into several large U.S. enterprises. It was the exact type of partnership we aim to provide: not just capital, but also access, credibility, and strategic support.
Why Wiz Matters
This acquisition is significant not just for Wiz, but for the entire cybersecurity industry. The increasing complexity of cloud environments — accelerated by the rise of generative AI — demands next-generation security solutions. Wiz’s success is a testament to the importance of innovation in this space, and Alphabet’s acquisition validates the immense value of modern, scalable cloud security platforms.
The use of cloud technology continues to grow every year, and we’re seeing a huge step-function jump in cloud resources driven by AI workloads. Wiz is uniquely positioned to expand into cloud and AI security with solutions such as its recently launched AI Security Posture Management module.
Congratulations to Assaf, Ami, Roy, Yinon, and the entire Wiz team on this remarkable achievement. You built something extraordinary: a company defined by trust, transparency, and relentless customer focus. You attracted top talent from around the world and built one of the most respected teams in cybersecurity history.
Wiz’s journey is an extremely rare instance of the right people building the right product in the right market at the right time. When that happens, a team can achieve outlier results. We’re grateful to have been part of Wiz’s journey from the earliest days. This outcome is a testament to what happens when exceptional founders meet a massive market opportunity and execute with precision and speed.
]]>
2025 was a banner year for Salesforce Ventures: 44 new investments and 30 follow-on investments across our core investments themes, as well as a new focus in emerging technologies like robotics, energy, and defense. The pace of innovation was astounding — and it was a privilege to be there alongside the founders driving it.
Behind the Investment: Anthropic
We first backed Anthropic in March of 2023, and we've stood by them through every round since. In honor of Anthropic’s Series G announcement, our team shared why we originally invested in Anthropic, and why our conviction has only grown stronger over time.
Verified Code for Mission-Critical Systems
Code Metal is closing the trust gap for mission-critical industries with a neuro-symbolic platform that marries AI-generated code with formal verification.
AI Needs Interpretability
Goodfire inspects model internals directly — providing a precise, legible signal of a model’s intent and reasoning — enabling users to change a model’s behavior in the ways they want to.
Risk Engineering for the Alliance Economy
Lema is rebuilding third-party risk management from the ground up with continuous vendor security powered by agentic AI that thinks like an attacker.
AI That Works for Japanese Enterprises
Sakana AI is accelerating AI adoption in Japan by developing sophisticated LLMs that combine multiple models to achieve superior performance and efficiency designed for Japanese enterprises.
Bringing AI to Structured Data
Fundamental accesses spreadsheet data with their Large Tabular Model (LTM), delivering accurate predictions with a single line of code without the need for preprocessing, fine-tuning, or retraining.
Enki Toto Named to Business Insider’s 2026 Rising Stars of VC
Congratulations to our own Enki Toto on being named to Business Insider’s 2026 Rising Stars in VC list. Enki has been instrumental in advancing Salesforce Ventures’ mission to back the next generation of enterprise AI, specifically through her work in energy and AI infrastructure — categories essential to how enterprises will scale in the coming decade.
SFV IRL
Where to find our team in March:
March 11: Alex Haskell @ Crafty Counsel’s M&A Day (London)
March 23 - 26: Kevin Wu, Robert Keith, and Kartik Gupta @ RSAC (San Francisco)
Thanks for reading! If you’re a founder building in our core investment areas, submit an investment inquiry to get in touch. To learn more about Salesforce Ventures, visit our website or subscribe to our Substack.
]]>
Founder: Peter Morales
Sectors: AI Infrastructure, Defense Technology
Location: Boston, Washington D.C., San Francisco
The Infrastructure Debt Crisis
The world’s most critical systems — defense platforms, semiconductor toolchains, automotive sensors, aerospace avionics — are built on decades-old code. Much of this software was written in languages like C, C++, MATLAB, and ADA, tightly coupled to CPU architectures and bespoke hardware never designed for modern AI workloads. As compute paradigms shift from CPU-centric to GPU-accelerated and edge-deployed systems, organizations face a compounding crisis: massive technical debt that cannot be rewritten from scratch, an acute shortage of engineers fluent across legacy and modern stacks, and high-stakes environments where correctness matters more than velocity.
Manual code rewrites are slow, risky, and unscalable. Traditional compilers lack semantic understanding. AI coding assistants optimize developer productivity but cannot reason over system-level invariants or provide formal correctness guarantees. Critical institutions — from chip manufacturers to aerospace companies to the military itself — are bottlenecked by their ability to safely translate and evolve software across architectures.
Code Metal: Closing the Trust Gap
Code Metal is tackling one of the most important challenges in software development today: closing the trust gap for mission-critical industries. Particularly as AI-generated code proliferates, the question isn’t whether AI can write code — it’s whether that code can be trusted in environments where failure isn’t an option.
Code Metal has built something fundamentally different: a neuro-symbolic platform that marries generative AI with formal verification. Rather than just predicting the next token, Code Metal breaks programs down into testable, symbolic components to mathematically prove that generated code is functionally equivalent to the source and free of safety violations. This approach serves as a verification system for hardware code, particularly in industries where zero error is the default.
The company’s proprietary research enables Code Metal to solve transpilation problems significantly faster than competing formal tools while dramatically reducing manual implementation effort. Their team — led by co-founders Peter Morales and Alex Showalter-Bucher, who possess deep F-35 and defense systems experience from their time at MIT’s Lincoln Lab — has assembled the deepest private-sector bench in formal methods and safety-critical systems, including the developer of the MATLAB compiler and NASA’s global leader in formal verification research. Further, Code Metal has demonstrated exceptional velocity in its first year of commercialization, securing customers including L3Harris, Raytheon, and the U.S. Air Force.
Our Belief in Code Metal
By the time we met Peter Morales last November, we’d already heard effusive praise about his team and product from dozens of friends, co-investors, and portfolio companies. In less than 2 years, Peter and his team have built a product that engineers working on the world’s most mission-critical systems are already using to automate code translation, validation, and modernization. Spending time with Peter in D.C. confirmed what we’d heard: this was a rare, generational company that we’d feel fortunate to partner with.
We’re thrilled today to be leading Code Metal’s $125M Series B at a $1.25B post-money valuation alongside Accel, B Capital, as well as Shield Capital, Smith Point Capital, RTX Ventures, and others who have partnered with Code Metal from Day 1.
What’s Next
Code Metal is positioned to become the default platform for verified code translation across the edge, AI/cloud migration, and regulated industries. With a rare combination of world-class research talent, formal verification IP, and exceptional early traction across the automotive, aerospace, defense, and semiconductor industries, we believe Code Metal is building pick-and-shovel infrastructure for the next era of AI-native software development.
As AI moves to the edge and into the physical world — and as defense and national security organizations accelerate AI-driven modernization — the ability to safely, verifiably, and rapidly translate code across languages, architectures, and deployment targets becomes strategic infrastructure. Code Metal has built exactly that.
We’re excited to be part of this journey. Welcome to Salesforce Ventures, Code Metal!
]]>
Founders: David Ha, Llion Jones, Ren Ito
Sector: Generative AI
Location: Tokyo, Japan
The Opportunity
There’s a strong desire within the Japanese business community to leverage AI to enhance productivity and drive the next wave of innovation. The opportunity is clear, but realizing it requires solutions purpose-built for Japan’s unique business environment. Global AI platforms often don’t align with Japanese workflows and decision-making processes, while domestic solutions have historically focused on application layers rather than foundational research and infrastructure.
There’s an exciting opportunity to help Japanese industries realize the productivity gains that AI can deliver. What’s needed is a company that bridges world-class AI research with a deep understanding of Japanese enterprise needs.
The Solution
Sakana AI is building that bridge. Founded in 2023, the company develops sophisticated LLMs by combining multiple models to achieve superior performance and efficiency specifically for Japanese enterprises. Rather than offering off-the-shelf products, Sakana focuses on what they call “Layer 3.5” — the critical space between AI models and real-world enterprise applications.
In practice, this means Sakana AI delivers fully realized AI services, combining deep business insight with advanced AI capabilities to craft services tailored to each customer’s workflow. The result is AI that doesn’t just work in theory, but integrates seamlessly into how Japanese companies actually operate.
The company initially targeted two of Japan’s most demanding sectors: finance and defense. Already, Sakana is deploying solutions at MUFG and Daiwa Securities — Japan’s two largest financial companies — building AI systems that leverage deep domain expertise to engineer solutions tailored to these businesses’ specific workflows. Moving forward, the company plans to expand into manufacturing and other sectors, leveraging the expertise gained from these early customers to deploy high-quality solutions at scale.
Why We’re Backing Sakana AI
Sakana AI is led by globally renowned AI researchers David Ha (CEO) and Llion Jones (CTO), alongside Ren Ito, who brings exceptional business development expertise from his background in diplomacy and as former Head of Europe for Mercari.
David Ha’s journey — from Goldman Sachs derivatives trader to Google AI researcher to Sakana founder — gives him a rare combination of technical depth and an understanding of enterprise decision-making. Llion Jones co-authored the seminal “Attention Is All You Need” paper in 2017, the breakthrough that underpins modern AI. Together, they’ve assembled what is arguably Japan’s strongest team of AI researchers and engineers.
This combination of world-class research talent and sophisticated business execution is precisely what sets Sakana apart. Global AI companies haven’t successfully penetrated Japanese enterprises. Japanese AI companies bring valuable market expertise, and Sakana uniquely combines that local knowledge with world-class AI research capabilities to deliver sophisticated AI solutions built with an understanding of Japanese business requirements.
The company’s momentum reflects this unique positioning. Sakana became Japan’s fastest company to achieve unicorn status. At a time when addressing labor shortages and boosting productivity is a matter of national urgency, Sakana AI is leveraging state-of-the-art AI technology to drive operational efficiency and enhance resilience across Japan’s industrial base.
“Salesforce Ventures has been a dedicated investor in Japan’s tech ecosystem since 2011, and Sakana AI represents the next frontier of innovation,” said John Somorjai, President of Salesforce Ventures. “Their rare blend of world-class research and deep enterprise execution makes them a preeminent force in defining Japan’s AI future. We look forward to supporting Sakana’s journey as they scale alongside Agentforce, delivering a new era of autonomous productivity to Japanese industries.”
Looking Ahead
As AI implementation becomes indispensable across all sectors, we’re honored to support Sakana AI, which we believe is at the forefront of AI innovation in Japan.
We’re proud to participate in this Series B round alongside Factorial Funds, MUFG, Khosla Ventures, and Google, among others. We look forward to watching Sakana AI accelerate AI adoption across Japan, providing solutions that deliver meaningful transformation.
Welcome to Salesforce Ventures, Sakana AI!
]]>
The generative AI boom is often framed as a race for raw power — who has the most capital or compute. For us, the story of Anthropic didn’t start with a quest for the biggest cluster. It started with a desire to find the best team bringing AI models to the enterprise.
We first invested in Anthropic’s Series C in early 2023, and we’ve stood by their side through every subsequent round since, including their latest Series G in February 2026. Supporting Anthropic’s journey from early research to their current scale has provided a front-row seat to the incredible pace of change in the AI era.
A Bet on Constitutional AI
In the summer of 2022, OpenAI released DALL-E 2. The visual impact of pixels synthesized to create beautiful, realistic imagery (for that time) made us think something could be afoot in the AI space, and we started researching the market in earnest. ChatGPT’s launch in November 2022 validated our suspicion; generative AI could lead to a platform shift.
Working with the Salesforce AI Research team, led by Chief Scientist Silvio Savarese, we developed a conviction that durable value would initially accrue at the foundation model layer, where deep research expertise creates meaningful barriers to entry. Anthropic was uniquely positioned. The founding team had published seminal papers on scaling laws and reinforcement learning and had helped build GPT-2 and GPT-3 while at OpenAI. They possessed a rare “founder-market fit” and the foresight to focus on long-term technical advantage rather than short-term optimization.
Our initial meeting with co-founder and CEO Dario Amodei took place in early 2023 in San Francisco. Dario spoke passionately about reinventing productivity and knowledge work. While inspired, we remained skeptical: how could an intelligent chatbot revolutionize the economy?
Then Dario laid out the roadmap for how the model would improve beyond its current capabilities and introduced “Constitutional AI,” which felt more like a philosophy for Anthropic than a training method. It uses a set of principles (the “constitution”) to guide a model’s behavior toward being “helpful, harmless, and honest.” Well before AI safety became a central regulatory debate, Anthropic believed frontier AI would only achieve widespread adoption if it was built to be safe and interpretable from the ground up. This stood in stark contrast to other labs chasing near-term usage.
At the time, alignment relied on Reinforcement Learning from Human Feedback, a fine-tuning technique that doesn’t scale well. Constitutional AI was more elegant, allowing the machine to supervise itself through its written principles. After testing this approach with our network of Fortune 500 CIOs and Salesforce product teams, we realized this wasn’t just a safety feature; it was a key selling point. We believed enterprises would value having control over their own model’s “constitution.”
A Focused Approach
During Anthropic’s Series C, the debate over model commoditization was front and center, as it remains today. Our conviction in Anthropic was rooted in the observation that, in an emerging industry where the foundational technology changes daily, the delta between winners and losers lies in execution speed and relentless focus.
Neerav Kingsland, Head of Strategic Initiatives at Anthropic, explained his team’s philosophy at that time: “We’re not optimizing for next week’s viral demo. We’re optimizing for the moment an enterprise asks a model to write sensitive code or analyze a complex medical trial.”
That focus manifested in concrete choices. Anthropic built Claude with Slack in mind — where enterprise data lives — and zeroed in on specific use cases: coding, tool use, and later, reasoning. Within a year of launch, that philosophy was proven — rapid customer iteration revealed Claude’s core differentiators: code generation, customization, and data security. This drove the Claude 2 family of models, showcasing the early coding capabilities now central to their value proposition.
The competitive landscape at the time was intimidating, but we understood what mattered: enterprise revenue is hard to earn, but far more durable. With this understanding, our continued investment in Anthropic’s subsequent funding rounds — from Series D through Series G — has been an easy decision.
Throughout our partnership, Salesforce Ventures has sought to provide Anthropic with not only capital, but also a window into how global enterprises actually function. We stress-tested Claude in demanding environments: data residency in Europe, model steerability under load, etc. We’ve also introduced Anthropic to a network of Fortune 500 leaders who have become deep partners and customers, including Coinbase, Barclays, and Accenture. Today, Claude powers Slack AI and serves as Agentforce’s preferred model in regulated industries.
Neerav later reflected: “Having someone who is invested in you, who you can trust, and who can help you find mutual successes is great, and we got that from the beginning with Salesforce Ventures as they were helping us meet the right people at the right time.”
Market Validation
A core belief in betting on the foundation model layer is that it would become essential infrastructure, and Claude has most certainly become that. The company’s momentum is driven by enterprises selecting AI partners capable of operating in high-stakes, high-ROI environments.
The evolution of Claude Code is the most visible example. It has meaningfully improved developer productivity and compressed development timelines to the point that companies not using it are now at a disadvantage. Claude Code is now often the wedge that gets Anthropic in the door, allowing the team to expand into horizontal and vertical use cases across the enterprise. Anthropic also continues to expand into regulated industries — financial services, life sciences, and healthcare — where the AI workflows customers build with Claude are central to their operations.
In short, Claude Code has taken off, and we’re witnessing firsthand what Dario meant when he spoke of reinventing productivity back in 2023.
Beyond the Hype
In 2023, betting on a research-heavy startup at a meaningful valuation was not obvious, given there was no market precedent of AI research turning into commercial success. Today, Anthropic is the canonical success story of the AI era, proving that you can move from seminal research to a multi-product platform company that captures the world’s imagination. They also proved that a shared commitment to culture and trust can remain substantive even at a scale of thousands of employees.
As Anthropic enters its next phase of growth, we remain steadfast in our belief in this team. The frontier is still moving, and the questions are only getting harder. That is exactly where Salesforce Ventures and Anthropic want to be.
]]>
Founders: Eddie Dovzhik (CEO), Omer Yehudai (CPO), and Tomer Roizman (CTO)
Sector: Cybersecurity / Third-Party Risk Management (TPRM)
Location: New York, NY
Here’s a statistic that every CISO knows all too well: 60% of data breaches start with a compromised third-party vendor. Yet most companies still manage this massive attack surface with spreadsheets and annual questionnaires — tools better suited to the 1990s than today’s interconnected digital economy.
The problem is accelerating. The average enterprise now depends on more than 1,000 external vendors, from SaaS platforms to AI agents. These aren’t just service providers listed in a procurement database. They have direct access to your systems, your data, and your customers. When they get breached, you get breached.
Traditional third-party risk management tools aren’t built for this reality. They rely on “outside-in” security ratings; generic scores based on public data that are often wrong and out of date. They can’t tell you what your vendors are actually doing inside your environment right now. They can’t show you which vendor has access to your sensitive systems and data. And when a vendor gets compromised, they certainly can’t tell you how to shut down the threat before it spreads.
We need a fundamentally different approach. One that treats vendor security as an active threat modeling problem, not a compliance checklist.
Enter Lema: Risk Engineering + Red Team DNA Meets Third-Party Risk
Lema is emerging from stealth after raising $24M across seed and Series A rounds to rebuild third-party risk management from the ground up. Instead of a compliance platform with an AI wrapper, Lema’s focused on a completely new category: continuous vendor security powered by agentic AI that thinks like an attacker.
Founders Eddie Dovzhik (CEO), Tomer Roizman (CTO), Omer Yehudai (CPO) served together as security researchers and R&D leaders in Unit 8200, Israel’s elite cyber intelligence unit. The trio are offensive security researchers who bring an attacker’s mindset to a space that’s been dominated by auditors and checkbox-fillers.
Their deep cyber DNA shows up everywhere in how Lema has been built. The platform forensically analyzes every document your vendors submit and scours public sources for information vendors might prefer to keep quiet. Most importantly, it monitors in real time how vendors actually behave inside your environment — tracking data movement, permission changes, and access to critical assets.
The Tech Behind the Transformation
Lema’s platform runs three core intelligence engines, all feeding into what they call Agentic Risk Engineering: Forensic Artifact Analysis automatically dissects unstructured vendor documents (e.g., SLAs, compliance reports, security questionnaires) to surface hidden risks and contradictions that human reviewers miss.
OSINT Recon continuously gathers intelligence from public sources, dark web monitoring, and breach databases to identify vendor vulnerabilities before they become your vulnerabilities. Most importantly, Blast Radius Monitor provides real-time visibility into exactly what would happen if a vendor got compromised. Which systems would be exposed? Which data would be at risk? Which users would be most impacted? What’s the precise sequence of actions needed to contain the threat?
By combining “inside-out” visibility with “outside-in” intelligence (what’s happening in the broader threat landscape), Lema can assess a new vendor in under five minutes and provide specific, actionable remediation steps.
Why We’re Excited To Back Lema
Lema is tackling a massive and worsening problem with a differentiated approach — and they have the right team to pull it off.
Eddie previously scaled the early product at Noname Security from seed through Series C, learning how to take a technical platform to market in the enterprise cybersecurity space. Tomer and Omer led research and engineering teams in elite cyber organizations, building systems that had to work under the most demanding conditions imaginable.
What’s impressed us most is their go-to-market hustle. The Lema team built their customer base at supply chain and GRC conferences across America. Why? Because that’s where to find the folks responsible for vendor risk. They’re meeting CISOs and compliance officers where they live and learning their pain points. This combination of offensive security chops and bottoms-up market understanding is rare. The team isn’t trying to sell another security rating service. Instead, they’re transforming the old school GRC philosophy into real security engineering.
The market is responding. Lema already counts Fortune 500 companies among its customers, spanning highly regulated industries like financial services and healthcare. As one CISO told us: “Lema removes the compliance theatre of traditional security ratings by providing verified, evidence-backed risk data.” That’s the difference between checking boxes and actually reducing risk.
The Path Forward
Lema is emerging from stealth at exactly the right moment. Companies are drowning in vendor relationships, and AI agents are proliferating across the enterprise — each one representing a new potential attack vector. Regulators are demanding more rigorous third-party risk management. And traditional TPRM tools are failing to keep up.
We backed Lema at seed when they were three relentless founders with a vision but no product. Watching them transform that vision into an industry-leading platform that customers genuinely love has been remarkable to witness. We believe Lema is fundamentally changing how enterprises secure their supply chains — moving from reactive compliance to proactive Risk Engineering.
The future of cybersecurity requires vendor ecosystems built on trust and backed by real security, not just paperwork. Lema is building the foundation for that future.
We’re excited to be along for the journey.
]]>
Founders: Eric Ho, Daniel Balsam, Thomas McGrath
Sector: AI Infrastructure
Location: San Francisco, CA
More than three years into the generative AI era, enterprise customers care more about the ROI they see from their AI investments than ever. In a recent UBS survey of 130 IT execs, “unclear ROI” was the most commonly cited challenge to enterprise AI adoption. From our own observation, a big driver of ROI concern is that enterprise customers cannot steer AI models to behave reliably, consistently. For these reasons and more, we’re excited to partner with Goodfire as they build the foundational infrastructure for understanding and intentionally designing modern AI.
Goodfire is solving for the lack of AI reliability and control over training using interpretability. While traditional approaches rely on “black-box” methods, Goodfire inspects model internals directly. This approach provides a precise, legible signal of a model’s intent and reasoning, moving the industry from guessing at behavior to measuring and precisely shaping it so that users can change a model’s behavior in the ways they want to.
Goodfire creates a powerful new feedback loop for the enterprise. Insights from Goodfire can be used to guide model training, fine-tuning, and alignment strategies, helping teams improve reliability and safety and design their models without relying solely on prompts or black-box training methods. Goodfire’s platform offers insights into a model’s inner workings and enables scientists to directly extract novel discoveries from powerful scientific AI models, as demonstrated by their recent identification of a novel class of Alzheimer’s biomarkers with Prima Mente.
Recently, the company developed new methods to precisely target parts of a model’s internals, enabling more exact modifications to a model’s behavior. Unlocking the ability to precisely design and debug a model’s behavior will enable more AI use cases to be put into production and open up the possibility of working more reliably with different models, including open source ones.
Goodfire was founded by Eric Ho, Daniel Balsam, and Thomas McGrath. Eric and Daniel built one of the first applications in applied AI through their prior company, RippleMatch, a recruitment automation platform. Thomas, Goodfire’s Chief Scientist, is a pioneer in interpretability, having co-founded the interpretability team at DeepMind, and continues to be a highly respected researcher in the field. In just a few years, Goodfire has built a team consisting of many of the leading researchers across interpretability. We believe Goodfire is the best independent team doing interpretability research and building real products on top of that research to democratize access to interpretability for enterprises and model labs alike.
Goodfire is building critical infrastructure for the next phase of AI deployment, and we’re proud to support Eric, Daniel, Thomas, and the entire team as they turn interpretability into a production-ready standard for the industry.
Welcome to Salesforce Ventures, Goodfire!
]]>
Founders: Jeremy Fraenkel, Marta Garnelo, Gabriel Suissa
Sector: Enterprise AI
Location: San Francisco, CA
While LLMs have transformed how enterprises work with text, a large majority of enterprise data still sits in spreadsheets — a format that’s stubbornly resistant to AI. Traditional methods for converting table data like XGBoost require extensive preprocessing, break with missing values, and demand months of specialized data science work for each use case, rarely achieving high accuracy.
Fundamental has built a Large Tabular Model (LTM), doing for structured data what Claude does for natural language. Their foundational model delivers accurate predictions with a single line of code, requiring no preprocessing, fine-tuning, or retraining. Early results are compelling: in fraud detection testing, their NEXUS models outperform traditional methods across use cases like churn prediction, drug discovery, and demand forecasting in industries including financial services, healthcare, and energy.
Since co-founding Fundamental in 2024, CEO Jeremy Fraenkel has assembled a world-class research team to tackle one of the largest unsolved problems in enterprise AI. The Fundamental team brings rare expertise across frontier AI research and enterprise go-to-market from organizations like DeepMind, Isomorphic Labs, and JP Morgan.
After building in stealth for 18 months, we’re excited that Fundamental is formally launching NEXUS and announcing a strategic partnership with Amazon Web Services to put its model in the hands of enterprise customers. NEXUS can consume raw tabular data and autonomously understand its structure, relationships, and patterns — no feature engineering or custom model training required. The result? Materially higher predictive accuracy compared to conventional machine learning, delivered with dramatically less effort.
We’re thrilled to partner with the Fundamental team as they work to redefine how LLMs are built and excited to have co-led their Seed and to be participating in their Series A.
Welcome to Salesforce Ventures, Fundamental!
]]>
Building the Nervous System for Multimodal AI
The next generation of AI doesn’t just read and write — it sees, hears, speaks, and acts in the world. But building these systems is fundamentally different from building traditional web applications. Voice and multimodal agents must operate across telephony, web, and mobile environments; process streaming audio and video; respond in milliseconds; coordinate multiple AI models in real time; manage turn-taking and interruptions; and reliably scale thousands of long-lived, stateful sessions.
Until now, the infrastructure required to make this possible simply didn’t exist.
Enter LiveKit.
Founded by Russ d’Sa and David Zhao, LiveKit is building the nervous system for multimodal AI. Their end-to-end platform gives developers everything they need to build, deploy, and operate real-time AI agents — from SDKs for voice and video frontends, to a backend framework for agent logic, to a global cloud runtime that enables autoscaling, observability, and production-grade reliability.
LiveKit is already powering some of the most demanding real-time AI workloads in the world, with customers including xAI, Meta, and Spotify — a signal of both technical excellence and deep trust from leading builders.
What stands out most to us is the team’s developer-first ethos. Russ and David have a rare clarity of vision around where multimodal AI is headed, and, more importantly, what abstractions developers will need to build in that future. LiveKit doesn’t just make real-time AI possible; it makes it accessible, composable, and scalable for teams of all sizes.
We’re thrilled to partner with LiveKit on their Series C as they become the foundational infrastructure layer for the next era of AI-driven applications.
Welcome to Salesforce Ventures, Russ, David, and the entire LiveKit team.
]]>
Founder: Grant Demaree
Sector: Defense Technology, Vertical AI
Location: Honolulu, HI
Modern military operations move extremely fast.
Yet coordination still relies on fragmented emails and static documents. In today’s conflicts, that information gap isn’t just inefficient — it’s a strategic risk.
Onebrief is modernizing military operations by building the definitive system of action for planning, collaboration, and global integration. Its platform replaces manual, document-heavy processes with a real-time, collaborative environment purpose-built for defense. By synchronizing planning data across every echelon, Onebrief enables staff officers to move at the speed of relevance. Armed with the most important context, they can spend less time formatting information and more time making high-stakes decisions.
What’s impressed us most is Onebrief’s product-market fit within the most demanding environments. The platform is already in active use across the Joint Staff and major U.S. Combatant Commands, supporting complex operations spanning the globe. Workflows that once took days now happen in minutes, maintaining alignment as conditions shift.
The acquisition of Battle Road Digital further extends Onebrief’s functionality and market leadership. By integrating simulation directly into the planning cycle, Onebrief allows leaders to stress-test and refine plans in a single, closed-loop system. This moves the industry away from static plans toward a living digital environment.
Onebrief is led by founder and CEO Grant Demaree. As a former U.S. Army officer with the 101st Airborne, Grant brings the unique empathy and context of a seasoned operator required to navigate defense complexity. He and the Onebrief team have built a product that is revolutionizing defense operations, earning genuine trust from the people on the front lines.
Salesforce Ventures is proud to invest in Onebrief’s Series D. We believe Onebrief is the inevitable operating system for modern command and control, and a generational company in the defense technology space. We’re thrilled to partner with Grant and his team as they define the future of global security infrastructure.
Welcome to Salesforce Ventures, Onebrief!
]]>
Founders: Anton Osika and Fabian Hedin
Sector: Artificial Intelligence
Location: Stockholm, Sweden
The high cost, slow pace, and technical complexity of traditional software development creates a steep barrier to entry for non-technical personas (e.g. founders, product/design leaders, and hobbyists) trying to bring their ideas to life. Fortunately, AI is transforming and democratizing software development — and Lovable is one of the companies leading this revolution.
Lovable’s mission is to equip the 99% who cannot code through an intuitive product experience. The company’s product architecture and UI/UX prioritize simplicity and accessibility, enabling non-technical users to easily build websites and applications through prompts while abstracting away the nitty gritty of software application building. Consequently, Lovable empowers domain experts to directly express their creativity and execute on business opportunities instead of having to rely on busy developer teams.
Led by experienced founders Anton Osika (formerly co-founded Depict) and Fabian Hedin (formerly founded Tentium and TenFAST), Lovable launched only a year ago but has already achieved incredible traction. To date, 25M+ projects have been created using the Lovable platform, and Lovable consistently rates highly on UX/UI customer surveys. We believe Lovable is quickly establishing itself as a leader in AI application development targeted at non-technical users, and are thrilled to invest in their Series B fundraise.
Welcome to Salesforce Ventures, Lovable!
Founders: Issei Takino (CEO) and Rosen Diankov (CTO)
Sector: Robotics
Location: Tokyo, Japan
The Opportunity
Manufacturing and logistics operations worldwide face intense pressure, driven by labor shortages, rising wages, and the increasing complexity of on-site operations. Facilities must now manage greater SKU variability, shorter planning cycles, and unpredictable demand, all while maintaining throughput, safety, and accuracy.
Traditional automation systems, with their fixed programming and rigid infrastructure, were never designed to accommodate this level of variability. When product dimensions, packaging, or workflow priorities shift, these legacy systems struggle to keep pace. This forces essential, repetitive tasks (such as depalletizing, case handling, kitting, and sorting) to remain dependent on human labor.
As these physically demanding roles become harder to staff, productivity plateaus and operational costs rise. The core challenge is no longer just labor; it is the lack of adaptable, intelligent automation that can match real-world conditions. Closing this gap is crucial for manufacturers and logistics providers to scale and remain competitive.
The Solution
Companies are eager to adopt robotics, but traditional deployment is challenging. Today’s robots often require painstaking, expert programming for every movement. If a product is slightly out of place, the robot stops, requiring an engineer to intervene. This makes automation risky and expensive.
Mujin is changing that.
Mujin’s technology enables robots to understand their surroundings and respond nimbly, much like a human worker. Instead of manual programming, robots equipped with Mujin intelligence recognize different products and adjust their actions automatically, ensuring continuous productivity with less engineering effort.
Mujin also removes the barrier of system complexity. While traditional facilities might use robots from several manufacturers—each requiring complicated, unique integration—MujinOS seamlessly coordinates robots from multiple global manufacturers under one intelligent platform. One smart system, many robots, working as one.
From unloading pallets and building mixed-SKU loads to coordinating automated vehicles, Mujin enables full-flow automation across the entire warehouse or factory. By making robots easier to deploy and maintain, Mujin is helping businesses scale automation faster than ever before.
Why We’re Backing Mujin
As the wave of physical AI rises, Mujin stands out as a company defining the future of intelligent robotics globally. Mujin’s strength lies in its combination of visionary leadership and deep technical expertise. Issei Takino’s intimate industry knowledge, paired with Rosen Diankov’s pioneering research in robotics from Carnegie Mellon University, has transformed advanced concepts into practical, production-ready solutions.
This commitment to precision and reliability meets the stringent requirements of the world’s most demanding industrial environments. Mujin’s technologies are already deployed by leading multinational manufacturers and logistics providers, positioning the company as a frontrunner in the global shift toward intelligent, autonomous operations.
Mujin’s innovations address universal pressures—rising labor costs, shrinking workforces, and accelerating complexity—enabling organizations worldwide to boost productivity, enhance safety, and establish more resilient, future-ready operations.
What’s Ahead?
As the wave of physical AI rises globally, we’re honored to support Mujin, a company emblematic of Japan’s strength in robotics technology.
Mujin’s solutions are already driving implementation within many global enterprises, and their business acceleration is anticipated to grow further as the world focuses more intensely on physical AI. We’re proud to join this Series D funding round alongside NTT Group, Qatar Investment Authority (QIA), and Mitsubishi HC Capital Realty, and contribute to Mujin’s global expansion and the transformation of the industrial sector.
Welcome to Salesforce Ventures, Mujin!
]]>
Founder: Daniel First
Sector: Vertical AI
Location: Brooklyn, NY
The Opportunity
Manufacturing has become increasingly complex.
Products contain thousands of components, electronics are becoming increasingly sophisticated, and intricate supply chains span the globe. Despite this complexity, customer quality and issue detection remain largely reactive processes dependent on manual investigation and siloed data streams (oftentimes across dozens of platforms). Poor product quality costs manufacturers billions of dollars each year, and each day a poor quality product goes undetected exacerbates the downstream impacts.
Today’s field quality engineers face a daunting challenge: detecting emerging issues across millions of units in the field, investigating root causes across disparate data sources, and preventing catastrophic outcomes, such as recalls. The current customer quality toolset isn’t built for the scale of modern manufacturing. When these legacy tools do work, they’re capable of catching known issues, but fail to identify novel, emerging problems that don’t fit predetermined patterns. By the time these issues surface in the broader organization, the damage is already done. Brand reputation suffers, recall costs mount, and customer trust erodes.
What manufacturers need is an intelligent observability layer that can synthesize unstructured data across the enterprise product lifecycle and surface emerging issues at their earliest warning signals. Manufacturers require AI that can support their best engineers, detect subtle patterns, and empower teams to investigate issues at the speed of thought.
The Solution
Axion provides an AI-powered observability platform that transforms how manufacturers detect, investigate, and resolve quality issues impacting customers. Rather than forcing engineers to waste precious time manually piecing together insights across fragmented systems, Axion creates a quality-centric digital twin of failure modes that synthesizes data across the messy universe of manufacturing data — CRM systems, ERP databases, IoT, PLM records, service tickets, complaint logs, warranty claims, and so on. This unified view enables customer quality teams to run sophisticated anomaly detection, perform root-cause analysis in minutes, and solve emerging failure modes.
Axion’s AI is capable of discovering entirely novel patterns, correlating seemingly unrelated events, and flagging signals that huge teams of human analysts may never connect. Quality teams at companies like Cummins, Baxter, Medtronic, Boeing, and SharkNinja use Axion as their quality command center. What used to require weeks of SQL queries and spreadsheet forensics now happens in real-time within Axion’s single, intuitive center console.
For manufacturers operating at scale, where every percentage point of quality improvement represents millions in cost savings and brand equity, Axion provides the observability infrastructure that enables proactive quality intelligence. The result is faster issue resolution, prevented recalls, and confidence in product quality.
Why We’re Backing Axion
When we met Axion’s founder, Daniel First, we were immediately struck by his unique perspective on manufacturing quality. During his time at McKinsey, he witnessed firsthand how even the most sophisticated manufacturers struggled with quality observability. Daniel’s experience helps inform product decisions at Axion, from the platform’s architecture to its interface, designed to help engineers operate with maximum efficiency.
The market timing for Axion is exceptional. Manufacturing is undergoing a profound digital transformation with unprecedented volumes of data being generated, and AI is finally mature enough to make sense of it all. As the product lifecycle becomes increasingly complex, manufacturers can no longer afford to rely on reactive quality management. Axion is well-positioned to become the essential infrastructure for this transition.
Further, Axion has earned the trust of a customer base that doesn’t adopt new technology lightly. During our market research, we received fantastic feedback from users, citing the multi-million-dollar cost savings provided by Axion and the record time-to-value. As one executive put it, the platform delivers “eight engineers in a box.”
In short, we believe Axion has the potential to define the future of manufacturing intelligence — starting with quality intelligence. In time, we see Axion integrating across the production lifecycle. In Daniel and his team, we’ve found the rare combination of domain expertise, technical sophistication, and customer obsession required to tackle one of manufacturing’s most valuable unsolved problems.
What’s Ahead
Salesforce Ventures is thrilled to lead Axion’s $37M Series B alongside Bessemer Venture Partners, Inspired Capital, Amplo, and Schneider Electric.
Since our investment, Axion has continued its exciting growth trajectory, signing flagship customers and expanding its product surface area into other parts of the manufacturing value chain. We can’t wait to see what’s next for Daniel and his team.
Welcome to Salesforce Ventures, Axion!
]]>
Founders: Robin Rombach, Andreas Blattmann, and Patrick Esser
Sector: Generative AI
Location: Freiburg, Germany
The Opportunity
Over the past few years, the promise of AI-generated media has become clear in the stunning, one-off outputs we see daily. The problem? AI-generated media is still confined to the consumer space, failing to achieve true enterprise traction. As core AI technology improves, we anticipate an inflection point where generative media shifts from an exciting novelty to an essential, non-negotiable component of creative workflows.
In this future, the competitive advantage belongs to those who can iterate fastest. When quality and design are key competitive differentiators, the company that can compress the create-review-refine cycle wins, enabling:
Marketing teams that test hundreds of variations in days, not weeks.
Creative agencies that collaborate with clients in real-time.
E-commerce and social media platforms that generate product imagery on demand.
However, the generative media market is still in its infancy and needs to deliver on three key promises:
Creative professionals need precise control and brand coherence for robust implementation in their workflows. Most models optimize for artistic novelty rather than reliable consistency. Teams can’t rely on a tool that generates slightly different aesthetics each time (drift is a real problem) and is not consistent with the brand’s identity.
The infrastructure for deploying these models needs to improve in terms of speed and scale. When generating tens of thousands of digital assets, latency and compute costs become business metrics.
Models must be flexible and trustworthy. Creators, enterprises, and developers require infrastructure with managed APIs for convenience, open weights for customization, and local inference for privacy, without sacrificing quality.
The Solution
Black Forest Labs (BFL) is building the foundational infrastructure for production-grade generative images through its proprietary FLUX family of models (e.g., FLUX 1, FLUX 2, FLUX Dev, FLUX Kontext, FLUX Pro). BFL’s approach is to build architecture around what forward-thinking enterprises, creators, and developers need to bring their imagination to life.
On precision and brand coherence, FLUX delivers hyperrealistic image generation with exceptional prompt adherence and character consistency. Their suite of editing tools (Fill, Depth, Canny, Redux) provides the precise, non-destructive control that maintains brand consistency across iterative cycles. In terms of speed and enterprise economics, BFL’s models are architected from the ground up to be production-ready, delivering the sub-second latency that real-time creative collaboration demands, while making enterprise-scale deployment economically viable. On infrastructure flexibility, BFL has made its FLUX family of models available across multiple channels, including API, open-source, licensing, and browser via Playground. This unique tiered approach means developers get genuine choice, enterprises get full control, and organizations can match deployment to their specific needs without compromising on quality.
Equally important, the company marries precision and rigor with a global perspective on responsible AI development. BFL has embedded responsible AI and content safety as core business principles. The team employs rigorous dataset filtering, post-training safeguards, and content moderation, utilizing tools from organizations like the Internet Watch Foundation (IWF). This commitment to safety ensures that enterprises and creators alike can deploy their models with confidence.
Why We’re Backing Black Forest Labs
Led by CEO Robin Rombach, Black Forest Labs’ founding team is made up of the original co-creators behind latent diffusion, adversarial distillation, and Stable Diffusion: groundbreaking research papers that have shaped the visual AI field. In fact, their work has influenced the core architectures of popular models, such as DALL-E and Sora, among others. At an incredible pace, they’ve translated their work into a leading research lab for visual intelligence, and we believe BFL will be the source of the next major advance in visual research. The team’s track record of innovation and respect within the industry has helped to attract and retain world-class talent, enabling BFL to build one of the fastest-growing companies in the world.
What’s most impressive is how quickly BFL has achieved market leadership across multiple channels. FLUX models are among the most downloaded (400 million+) and used globally across Hugging Face, fal AI, Replicate, and Together AI. At the same time, BFL has secured enterprise partnerships with blue-chip companies like Adobe, Meta, and Microsoft.
Lastly, the market dynamics are unmistakable. As organizations accelerate their shift from experimental generative media to production-grade systems, they need infrastructure that can be trusted at scale. BFL is uniquely positioned here, offering a multi-channel solution that empowers enterprises and developers to select the optimal deployment model without sacrificing any of their core requirements
What’s Ahead
BFL is poised to become the foundational platform upon which the next generation of creative tools and enterprise workflows are built. Generative AI is expanding beyond text and code into the visual world — where seeing, understanding, and creating images will drive the next wave of applications in design, media, and human-computer interaction. BFL is at the forefront of this shift, building the visual infrastructure for the AI age.
We’re thrilled to co-lead Black Forest Labs’ $300M Series B round with Anjney Midha at AMP, alongside existing investors Andreessen Horowitz, Northzone, Creandum, Bain Capital Ventures, and others.
Welcome to Salesforce Ventures, Black Forest Labs!
]]>One powerful example of that belief in action comes from G2. In 2023, G2 formally joined Pledge 1%, committing 1% of the company’s equity to charitable causes, with a focus on expanding access to STEM education for underrepresented communities worldwide. Their example showcases what it means to blend purpose with action and how a business can have a lasting positive impact in more ways than one.
A Commitment Rooted in Kindness
Over a decade ago, G2 was founded on a set of values: Performance, Entrepreneurship, Authenticity, and Kindness, or “PEAK” for short. Regardless of how the business evolved, founders Godard Abel, Tim Handorf, Mike Wheeler, Mark Myers, and Matt Gorniak agreed that G2 would remain rooted in these ideals.
After Salesforce Ventures invested in G2 and invited them to take the pledge, the founders agreed it would be a perfect opportunity to live their value of Kindness.
“At G2, we’re driven by Kindness and the desire to create real, lasting impact through compassionate leadership,” says Abel, G2’s CEO. “Joining Pledge 1% was a natural extension of this ethos, offering a structured way to make philanthropy part of G2’s DNA.”
The G2 team chose to center their philanthropic efforts on expanding access to STEM education for underserved communities. By opening doors to the tech ecosystem for those who might otherwise be left out, Abel and co. believe they can help ensure that the next generation of innovators has the opportunity to bring their ideas to life.
“By concentrating on this specific area, we’re ensuring our efforts can scale with intentionality,” says Abel. “We want to help build the next generation of software experts and builders.”
In practice, G2 structured their pledge around a handful of commitments:
The company has pledged to donate 1% of its equity to charitable organizations focused on improving access to STEM education for underserved populations.
The company has already begun its philanthropic efforts by volunteering with and donating to organizations such as Chicago Tech Academy, Ennovate Labs, Girls Who Code, and the American Cancer Society, aiming to deliver a positive impact in communities both in the U.S. and globally.
Through its G2 Gives program, G2 donates $10 to organizations like Girls Who Code every time a user submits a software review on the G2 platform.
Employees at G2 are granted 20 paid volunteer hours per year to volunteer with a philanthropic organization of their choosing.
To promote buy-in for this program, Abel, Handorf, Wheeler, Myers, and Gorniak agreed to personally pledge 1% of their own equity in addition to 1% of the company’s corporate equity.
“We wanted to set the tone from the top,” Abel explains. “This wasn’t just a corporate move; it was a personal one.”
That leadership example quickly resonated across the organization. Abel says Chief People Officer Priti Patel helped champion the initiative internally, ensuring that G2’s philanthropy efforts became a shared mission embraced across teams and geographies.
The Result: Engagement and Growth
G2’s philanthropic efforts to date have already raised over $1 million for charitable organizations through its G2 Gives program and resulted in thousands of employee volunteer hours across G2’s 600-person team. Further, it’s having a real impact on the ground. As part of G2’s pledge, Handorf has made multiple trips to Nigeria to serve as a guest teacher at a local school and speak at LAUTech University about entrepreneurship — witnessing firsthand how encouraging entrepreneurship and financial support can open doors to opportunity.
The G2 team shares these benevolent efforts via a philanthropy page on its website, blog, and via social media posts to amplify their work to customers and employees.
“Our Pledge 1% commitment reinforces our culture of purpose and kindness,” says Abel. “It reminds employees they are part of something bigger — a company committed to making a difference.”
Outside of increased employee engagement, Abel says customers and partners have taken notice of G2’s commitment to values-driven leadership.
“Our commitment signals to customers and partners that we are a values-driven organization,” Abel says. “We often collaborate with our partners to raise funds for G2 partner organizations and help them scale their impact.”
That alignment of purpose and partnership reinforces G2’s long-term vision — a company that helps others grow while growing itself responsibly.
Advice for Founders: Don’t Wait to Start
When asked what he’d tell early-stage founders considering Pledge 1%, Abel’s advice is simple:
“Don’t wait. Purpose is most impactful when embedded early. Committing something now — even just 1% — sends a strong message to your team and community. It’s a small stake today with the potential for massive impact tomorrow.”
For founders who feel unready to make a full financial commitment, Abel encourages starting small — through mentoring, volunteering, or supporting local nonprofits — and letting those efforts evolve into a long-term giving framework.
“Anchoring philanthropy in your company’s foundation helps shape culture and scale impact as you grow,” Abel says.
_
To learn more about the Pledge 1% program, visit their website >>>
]]>
Leadership: Josh Araujo (CEO) and Scott Sanders (CGO)
Sector: Defense Technology
Location: Clarksburg, MD
The Opportunity
Across service branches, the U.S. Department of Defense manages a vast fleet of over 400,000 ground vehicles. Modernizing this fleet to be autonomous would enable more troops to be out of harm’s way while maintaining operational readiness and effectiveness. However, this transformation requires a new kind of company. Traditional defense contractors, while experts in hardware, are not built to deliver the agile, software-defined, and interoperable platforms that autonomous operations demand. This has created a critical gap and a $10B+ opportunity for a company that can deliver autonomy at scale.
The Solution
Forterra is building the connective tissue for modern operations by successfully delivering both the hardware and operating system for autonomous platforms.
Instead of creating a single, proprietary vehicle, Forterra provides a full-stack platform that can be retrofitted onto existing vehicles or integrated into new ones. Forterra has over 100 patents across five core components, with additional integrated technologies that expand the ecosystem:
AutoDrive: A software-enabled hardware kit that allows vehicles to navigate autonomously in the most complex off-road and remote environments.
TerraLink: A powerful command and control (C2) interface that allows human operators to manage entire fleets of autonomous vehicles.
OASIS (Open Architecture Standard for Interoperable Systems): Forterra’s open architecture platform and interoperable mission module, a new operating standard. It enables the government and third-party developers to build and deploy new capabilities on top of the AutoDrive platform, thereby preventing vendor lock-in.
Vektor: A software-defined communications and data-brokering layer built for the edge, ensuring reliable performance in degraded or low-bandwidth conditions.
goTenna: Technology that delivers interoperable mesh networking devices and software that provide secure, off-grid communication for defense and public safety missions.
Why We’re Investing in Forterra
We believe Forterra is building a category-defining company with the rare ability to field interoperable autonomy at scale. Our conviction is driven by their world-class team, undeniable market traction, and defensible, ecosystem-first platform.
The company is led by an exceptional team that possesses elite go-to-market execution, deep technical expertise, and a two-decade foundation of technical excellence, established by founder Alberto Lacaze and CTO Joseph Putney. CEO Josh Araujo brings a powerful combination of experience as a U.S. Marine officer and an experienced defense investment banker. He is joined by fellow Marine and Chief Growth Officer Scott Sanders, an early and pivotal employee at Anduril who is widely regarded as one of the best GTM leaders in the defense community. Scott Philips — former CTO of Vannevar Labs and an early Palantir employee — also recently joined the Forterra team.
Forterra has achieved what few others in the defense technology space have — securing contracts with meaningful programs of record, including those within the U.S. Marine Corps and U.S. Army, and deploying their technology in the field.
This traction demonstrates Forterra’s ability to move beyond R&D and deliver real, fielded capabilities, including commercial pilots for yard truck autonomy, demonstrating Forterra’s dual-use capabilities. The team is also building a powerful ecosystem through partnerships with major primes, such as Volvo Defense, BAE Systems, and Raytheon’s RTX, that solidify Forterra’s position as a significant node for any mission in need of autonomous capabilities.
What’s Next
Looking ahead, we see Forterra positioned to become the default, next-generation prime contractor for ground operations and logistics. The long-term vision is for Forterra’s OASIS platform to become the department-wide, interoperable operating system for autonomy. This involves scaling deployment across a variety of programs across the U.S. Army, U.S. Marine Corps, and other service branches to modernize an increasing share of the DoD’s ground vehicle fleet. As a leading provider to the DoD for autonomous missions, Forterra is well-positioned to capture the growing commercial market for off-road autonomy as well.
Salesforce Ventures has invested in Forterra as part of its $238 million Series C funding round. We are excited to partner with Josh, Scott, Alberto, and the entire Forterra team as they redefine autonomy and build the connective tissue for modern operations.
Welcome to the Salesforce Ventures portfolio, Forterra!
]]>
Founders: Zach Dell, Justin Lopas
Industry: Energy & AI
Location: Austin, TX
The Opportunity
The unprecedented surge in demand for AI compute has created a critical bottleneck in the U.S. energy grid. Following a decade of flat power demand, there’s an urgent need for an estimated 130 new gigawatts (GW) of energy to support AI innovation. At the same time, traditional power generation has seen increasingly extended lead times, with a power plant taking 5+ years to come online, and utility-scale storage taking 3+ years to build. This pace is simply not fast enough to meet AI’s energy demands. Power constraints are cited by hyperscalers and utilities as their number one pain point, slowing the pace of AI development and threatening U.S. competitiveness around AI.
This existential challenge for the U.S. grid requires a solution that’s fast, reliable, and scalable. Aging infrastructure and interconnection wait times, which can stretch as long as decades, risk significantly delaying planned data center projects. The market not only needs to add capacity, but it needs to deploy that capacity at an unprecedented speed to ensure the foundational resource for the AI boom — electricity — is readily available. This is the critical problem Base Power is solving: delivering reliable and affordable power to the core pillars of the economy being reshaped by AI: hyperscalers, utilities, and homeowners.
The Solution
Base Power is a vertically-integrated distributed energy platform that leverages proprietary battery hardware, software, and retail electricity services to stabilize the grid and meet energy demand quickly. This vertical model gives Base Power control over costs, improves reliability, and dramatically accelerates deployment, differentiating it from traditional players. The company’s core innovation lies in the use of distributed energy through battery storage. Each battery takes less than a day to install, and in aggregate Base can install enough batteries to reach the same power capacity as a utility scale battery in only 3-6 months — a fraction of the time required for conventional power plants or utility-scale storage.
Base’s business model today is to be a Retail Electricity Provider (REP) in Texas, installing proprietary batteries in customer homes. This model allows Base to lean on the price fluctuations of the grid for energy arbitrage, charging batteries from the grid during low-peak times and selling power back to the grid or to the home during high-peak times — thereby improving both reliability and affordability for consumers. To date Base Power has avoided over 2,400 outages for its Texas customers, improving grid stability across the state.
Long term, Base has the ability to not only serve customers in deregulated markets such as the Electrical Reliability Council of Texas (ERCOT), but expand its focus to regulated utilities and hyperscalers to provide a distributed power plant. For utilities, Base offers a fast path to grid capacity: the company has early proof points with both Bandera and Guadalupe Valley Electric Cooperative (GVEC). For hyperscalers, Base can provide a 100 MW utility-grade power asset in just 3-6 months, offering a capacity lifeline that alleviates grid congestion, accelerates data center interconnection, and improves homeowner resiliency in the region.
Why We’re Backing Base Power
Base Power was founded in 2023 by Zach Dell (CEO) and Justin Lopas (COO), who had a shared vision for building a vertical solution in the energy sector. The executive team is exceptional, with a unique blend of experience from companies known for rigorous execution and vertical integration, giving Base Power a strong foundation in hardware, software, and energy development. Dell is a former investor at Thrive and Blackstone, while Lopas led manufacturing at Anduril and was a lead manufacturing engineer at SpaceX. Dino Sasaridis, Head of Hardware, led the Powerwall engineering team at Tesla. Cole Jones, Head of Growth, ran GTM for Starlink. Dana Paz leads deployments and formerly led a team of manufacturing engineers at Anduril. And, Jared Greene, who leads Software, formerly developed the laser mesh for Starlink. This one-of-a-kind team’s ability to seamlessly marry software with innovative, deployable hardware is a massive competitive advantage and a de-risking factor in a complex, capital-intensive industry.
What’s Ahead?
Base Power has the potential to become a national distributed utility infrastructure leader, directly solving the energy scarcity problem for the AI-driven economy. As the company expands its utility and hyperscaler engagement, we expect its multi-pronged approach — providing grid stabilization through Virtual Power Plants (VPPs) and offering direct, fast energy supply for data centers — to dramatically accelerate Base’s growth and market impact. Base Power’s success will prove that distributed energy resources are the most viable near-term path to rapidly scaling reliable energy capacity across the U.S.
We’re thrilled to partner with the Base Power team as they build a new power company driving reliable and affordable energy alongside fellow investors Addition, Thrive, Valor Equity Partners, Lightspeed, and others.
Welcome to Salesforce Ventures, Base Power!
]]>