Metro Multiples: A ranking of top startup ecosystems by return on investment

By: Nate Bek

If you handed me a dollar to invest in a startup, three metros would give me the best shot at returning $10: the Bay Area, Boston, and Seattle.

An Ascend analysis of venture capital efficiency across major U.S. cities reveals that while Silicon Valley maintains its lead in generating returns, Boston and Seattle are closing the gap with strong performance on smaller capital bases. The findings suggest that investors may be overlooking high-performing emerging markets in favor of legacy tech hubs.

The Bay Area leads with a Multiple on Invested Capital, or MOIC, of 10.07. Boston follows at 7.89. The biggest surprise? Seattle, long viewed as a second tier ecosystem, lands third with a MOIC of 7.48.

MOIC, the core metric in this study, divides total exit value by total venture capital invested. It captures actual returns through IPOs and acquisitions, removing the influence of inflated private valuations or unrealized marks. The data includes tech and biotech exits over $100 million between 2000 and mid-2025.

Silicon Valley companies raised more than $200 billion in that time and generated over $2 trillion in exit value. The region’s $2.56 billion average exit was lifted by several outliers, including Meta’s $104 billion IPO in 2012.

Boston startups raised $59.3 billion and delivered $467.5 billion in exits. Seattle companies attracted just $25.1 billion in investment and returned $187.6 billion. Despite its lower volume, Seattle’s average exit — $1.63 billion — was higher than Boston’s and second only to the Bay Area. Its largest deal was Pfizer’s $43 billion acquisition of Seagen.

(Hover over the regions to get their MOIC numbers compared to exit value. Read our methodology below.)

Both Boston and Seattle reflect the long-term advantages of regional specialization. In Boston, life sciences dominate. Proximity to research institutions, regulatory experience, and a deep bench of clinical operators have fueled a steady stream of biotech exits. The $39 billion sale of Alexion Pharmaceuticals to AstraZeneca remains among the largest.

Seattle’s strength lies in enterprise infrastructure. Microsoft and Amazon seeded a generation of engineering talent, much of it concentrated in cloud computing, dev tools, and AI systems. 

San Diego ranked fourth, with a MOIC of 6.95. The metro raised $17.4 billion and produced $120.9 billion in exits. Its biotech sector follows a familiar pattern: fewer bets, longer time horizons, and outsized returns when drugs reach market. 

New York placed fifth at 6.56, on $70.1 billion invested. Its strength in fintech and enterprise software produced consistent returns, though its average exit was smaller.

The remaining metros performed at a lower tier. Washington, D.C. posted a 4.41 MOIC, shaped by government services and defense technologies. Chicago and Los Angeles returned just over 4. Austin, despite its profile as an emerging tech hub, posted 2.62. Miami finished last at 1.79.

The results reveal persistent disparities in capital efficiency. Some of the country’s fastest-growing markets still lag in outcomes, while quieter cities convert dollars with precision. These differences matter more in a constrained funding environment, as LPs push for clarity on where capital is working hardest.

The industry has long favored concentration. Much of that remains rational. Silicon Valley still has unmatched density and depth. But the data suggests the gap is not as wide as capital flows imply. Cities like Boston and Seattle have outperformed on a relative basis for years, and with far less capital in play.

Venture returns are nonlinear. One company can change the trajectory of a hub. But ecosystems that consistently generate high-efficiency outcomes tend to have more than luck on their side. They have technical lineage, operator networks, and capital discipline. Those ingredients don’t guarantee success, but they give each dollar a better chance.

If the goal is to turn one dollar into ten, a few places are doing it better than most. That should inform how money moves next.

If you are looking to get the raw data, contact the head of CB Insights jason.saltzman@cbinsights.com. Thomas Stahura contributed to this report.

---

Methodology

This analysis was conducted to assess capital efficiency across selected U.S. startup ecosystems. Specifically, we sought to estimate where venture capital investment has historically yielded the highest return in terms of exit value.

To do so, we constructed a custom dataset of U.S.-based, venture-backed startup exits with disclosed or estimated values of at least $100 million, covering the period from January 1, 2000, through June 30, 2025. The principal metric used was MOIC (Multiple on Invested Capital), calculated as the ratio of total exit value to total known venture capital raised for each company.

The initial dataset was sourced from CB Insights, which provided foundational exit data. We supplemented this dataset using a web browser automation tool to extract additional information from publicly available sources, including press releases, SEC filings, media reports, and corporate websites. Where automation proved insufficient, we conducted manual enrichment and verification. This process included spreadsheet review, data cleaning, and, in some cases, direct outreach.

We excluded transactions involving non-technology companies. For example, companies in categories such as food and beverage, real estate, or hospitality were removed, even if they appeared in the raw dataset. All companies included had to meet two criteria: (1) the company must have raised verifiable venture capital prior to exit; and (2) the company must have exited via acquisition, public offering, or other liquidity event with a disclosed or credibly estimated value of at least $100 million.

Private equity–backed companies, spinouts, and founder-owned businesses were excluded unless we identified credible evidence of venture capital participation.

We created our own metro area categorization, loosely informed by Startup Genome's global ecosystem rankings. Ten metro areas were selected for inclusion. We made a discretionary decision to include Austin and to exclude Philadelphia, with the intent of focusing on the most active and publicly visible startup hubs. We acknowledge this selection introduces subjectivity into the dataset’s geographic framework.

Biases and Limitations

The authors of this analysis are based in Seattle, Washington, and maintain professional networks concentrated in the Pacific Northwest region. One author is a former technology journalist with enhanced access to local records and contacts. This may have resulted in higher data completeness for Seattle-based companies, including additional exit events and more precise investment totals not reflected in the original CB Insights dataset.

Although we attempted to apply consistent standards across all regions, the availability of region-specific information varied, and our local knowledge likely influenced the dataset composition. This may have introduced a slight bias in favor of Seattle-related data volume and completeness.

This was a two-person project conducted without institutional resources. Data collection and validation occurred over a period of approximately three months, using both automated and manual methods. While extensive effort was made to ensure accuracy, the dataset is not comprehensive, and some exits or funding amounts may have been omitted or misclassified.

This analysis is presented for informational purposes only. It should not be relied upon for investment decisions or used as the sole basis for assessing regional venture capital performance.

Mapping Cascadian Dynamism

It’s hard to label what’s forming across the Pacific Northwest.

These companies don’t fit neatly into a single category. Some are building tactical autonomy, others are advancing compact fusion, modernizing satellite operations, deploying AI-driven robotics, or hardening energy systems. The connective tissue is regional: they’re born out of Cascadia’s unique access to technical talent, physical infrastructure, research depth, and enterprise buyers.

We call this Cascadian Dynamism.

The geography offers structural advantages. Port cities like Seattle and Vancouver provide direct access to logistics, aerospace, and maritime operations. Off-road testing environments are close. The presence of long-range RF labs, defense contractors, space manufacturers, and datacenter operators provides a natural customer base for emerging dual-use and industrial AI startups.

Founders consistently point to the talent layer as the defining factor. Boeing, Amazon, and Microsoft have produced decades of systems engineers, autonomy experts, and cloud infrastructure leads. The Allen Institute for AI, Nvidia’s robotics research, and the University of Washington contribute a steady stream of researchers with deep experience in machine learning, simulation, and computer vision. Fusion and battery chemistry are also heavily concentrated here, with companies like Helion, Zap Energy, Group14, and Sila hiring from a deep regional pool.

Several founders have said, privately and publicly, that this is the only region where their company could be built. It's one of the few places where technical hires understand both AI and physical systems, and where early customers are located within driving distance. The region’s proximity to major datacenter builds, energy utilities, and maritime operations gives startups live environments to test and deploy. Defense buyers and energy procurement teams are actively engaging with startups at earlier stages.

The satellite supply chain is another anchor. Washington produces over half the satellites currently in orbit, thanks to longstanding aerospace manufacturing expertise and a dense network of suppliers. Startups like Starfish, Quindar, and Kymeta are building on that foundation to reimagine in-orbit servicing, control systems, and communications. Robotics startups like Agility are taking advantage of Amazon’s massive warehouse footprint to design and iterate on deployable automation. Other startups are working with research vessels, agricultural testing zones, and off-grid energy projects that would be hard to replicate elsewhere.

Capital is starting to follow the talent. While historically undercapitalized compared to enterprise software, this sector is seeing strong inbound interest. Breakthrough Energy Ventures, Point72, and FUSE are based here. DIU Energy has a foothold in the region. New firms like Conduit and Actuate are being formed around the thesis that complex physical infrastructure is overdue for replatforming —and that Cascadia is one of the few places where that replatforming is already underway.

At Ascend, we have made several investments in this category (four highlighted in the market map), and are on the hunt for more. We call it Frontier AI, or Ai solving the physical world’s hardest problems.

But most of these companies won’t show up in standard SaaS deal flow. Many are operating in stealth, emerging from DARPA projects, research institutes, and spinouts from primes. But the pattern is becoming harder to ignore: engineering-heavy teams, access to buyers, defensible technology, and early federal traction.

There’s a real market taking shape here — one that doesn’t need to chase the latest API trends to build lasting value.

Investing in Clarify

By: Nate Bek & Kirby Winfield

Today, we’re excited to announce our follow-on investment in Clarify, the AI-native CRM platform. USVP and Madrona co-led the $15 million Series A round, joined by Ascend.

Clarify is what we believe the next generation of sales platforms will look like: autonomous, adaptive, and deeply integrated into how modern teams actually work.

Sales reps still spend 70% of their week not selling. CRMs were built as static databases and evolved into bloated admin tools. Even with a growing ecosystem of AI plugins, most remain passive and reactive.

Clarify is an autonomous CRM that eliminates manual entry and reduces operational drag. It prepares call briefs, enriches contact records, records and transcribes meetings, extracts action items, drafts follow-ups, and keeps pipeline data current. The system runs ambiently in the background, removing the need for reps to manage their tools instead of their pipeline.

While everyone else is adding more complexity through bespoke agents that require constant customization and maintenance, Clarify takes the opposite approach. Clarify internalized intelligence into what we call ambient intelligence — you don't manage agents, the CRM is the agent. It's always on, always working in the background

This fits squarely within our SaaS 3.0 thesis: software that applies AI to targeted, high-friction workflows with measurable return.

We backed Patrick Thompson’s last company, Iteratively, from first check through acquisition by Amplitude. We have seen firsthand how he executes. Clarify is his second act, and he came to us early with a clear read on the CRM market shaped by his experience building analytics products at Amplitude and managing GTM operations at Iteratively. He knew where existing CRMs fell short and what a modern system would require. 

“This raise lets us accelerate the vision we’ve had since day one: reimagining the CRM from the ground up — autonomous, flexible, and finally built for the way teams actually work,” Patrick tells us. “What’s even more exciting is that it gives us the green light to go after the entire GTM stack. We’re not just building a better CRM — we’re building the platform we always wished existed.”

Patrick is joined by Austin Hay, previously at Ramp, and Ondrej Hrebicek, who co-founded Iteratively and worked at Amplitude. All three founders have built for high-growth environments and understand how today’s sales and product teams actually operate.

The market for CRM and adjacent tooling is crowded. Salesforce dominates enterprise CRM, while Hubspot has a lock on SMB. Gong focuses on meeting transcripts that get fed into Salesforce, while Outreach does enrichment of customer relationships. Companies also rely on pure transcription tools like Fireflies, Fathom, and Otter. These are large markets, and the incumbents are investing heavily in GenAI. 

But they carry structural baggage. Decades of architecture optimized for manual input and seat-based pricing make it hard to adapt. Bolt-on AI can’t overcome a brittle foundation.

Salesforce and Hubspot remain data silos for many companies, with customer records, activity logs, and pipeline stages locked behind layers of manual upkeep and admin settings. Even AI-enabled versions of these platforms often require reps to tag records, summarize calls, and trigger follow-ups by hand. 

Clarify is stitching this fragmented stack into a single autonomous system. By consolidating functionality and embedding AI from the ground up, Clarify avoids the layering problem that plagues incumbents.

The product is in active use across more than 100 customers, with over 70% engaging weekly. Customers include Paramark, Ravenna, Sift, Volca, and more. Clarify's pricing model gives customers unlimited users on all plans and charges based on usage rather than seats. The company now has around 20 employees, up from 12 at the start of 2024, and is hiring across engineering and go-to-market. Open roles are live on the Ascend Job Board.

Clarify was seeded in Q1 2024 and raised its Series A in Q1 2025. According to Carta, fewer than 9% of startups seeded in Q1 2024 had reached a Series A within four quarters. Clarify outperformed that benchmark.

We are proud to have been the first institutional investor in both of Patrick’s companies and to continue supporting the team as they take on one of the most entrenched and inefficient layers in enterprise software.

Investing in Yoodli

By: Nate Bek & Kirby Winfield

Today, we’re excited to announce our follow-on investment in Yoodli, the enterprise platform for AI roleplays. Neotribe Ventures led the $13.7 million Series A, joined by our friends at Madrona, Cercano, and AI2. 

We navigate the most important moments in life through conversation: closing a sales deal, interviewing for a job, or even prepping for a first date. Clear communication is often the difference between success and missed opportunity.

But most modern tools designed to improve our communication skills fall short. Corporate training still leans heavily on static formats like slide decks and recorded videos, which do little to build the real-time reflexes that define high-stakes conversations. These methods offer information, but they don’t develop skill.

Yoodli is one of the first-movers in a crop of companies innovating in this arena. Through its AI roleplays, users can rehearse high-stakes scenarios, receive real-time feedback, and improve with repetition. It’s an experiential learning tool built on a modern generative AI stack. As Founder Varun Puri puts it, “Yoodli is the batting cage before the big game — but for communication.”

Yoodli started as Project Speakeasy, a lightweight NLP tool that tracked filler words. But it has since evolved into an enterprise platform used by companies like Google, Databricks, RingCentral, Snowflake, BDO, and the University of Washington. These organizations use Yoodli to onboard employees faster, improve sales performance, and reduce the coaching burden on managers and administrators.

Rochana Golani, Databricks’ VP of learning and enablement, told The Information that one-on-one coaching is too expensive and that listening to call recordings doesn’t offer the personalized feedback employees need. Databricks, which employs thousands of sales reps, is now one of Yoodli’s largest enterprise customers.

We invested because we believe Yoodli fits squarely within our SaaS 3.0 thesis — B2B AI applications that target specific workflows and pain points with measurable impact. Communication is an overlooked domain in enterprise software. It’s critical to sales, management, and leadership, yet historically hard to train and assess at scale.

The rise of generative AI has changed that. AI roleplay is a new category that is quickly gaining traction. Salesforce, Gong, and Microsoft are all building in adjacent areas. This validates the space but also signals competition ahead. Yoodli’s edge lies in product depth, speed of iteration, and early enterprise traction. (Read Varun’s take on the competition here.)

We’ve supported Yoodli, a Seattle startup, since its first institutional round in 2022. Kirby met Varun through the AI2 Incubator and wired one of the first checks into the company. Varun previously worked with Sergey Brin at Google and brings the same level of focus to building his team. Co-founder Esha Joshi was a product manager at Apple and brings strong operational depth.

The use cases for Yoodli continue to expand. Sales enablement, manager training, leadership development, performance reviews, public speaking, media training, healthcare simulations — all benefit from practice-based learning. Roleplays can be tailored by persona, difficulty level, language, and feedback style, making them flexible and effective.

With the new funding, Yoodli will grow its team across sales, customer success, and engineering, while deepening its technology to make roleplays more realistic and personalized. You can find those jobs as they go live on the Ascend Job Board. Since our first investment, we’ve helped place multiple team members and advised through early product evolution.

We’re proud to have been early investors in Yoodli and look forward to supporting the team as it grows.

The Information first reported the round, followed by additional coverage in GeekWire.

Subscribe to Token Talk

Hot takes from the most wired-in analyst in AI. Hits your inbox every Wednesday at 11:30AM.

Subscribe

* indicates required

Email Address *

First Name

Last Name

You Let AI Help Build Your Product. Can You Still Own It?

By: Nate Bek

Startup developers should treat their AI coding assistants as copilots, not autopilots.

That’s my biggest takeaway from CenterForce’s Governance & Strategy Summit, held Wednesday at Bell Harbor International Conference Center on a sunny day in Seattle’s Belltown neighborhood. In a room full of legal minds from across the country, my goal was to take away what matters most to startup founders working with AI.

Two high-profile AI legal battles framed most of the discussions on stage: one centered on patents, the other on copyright.

In the patent case, computer scientist Stephen Thaler challenged the U.S. Patent and Trademark Office after it refused to grant patents for inventions created by his AI system. In the copyright case, artist Jason M. Allen sued the U.S. Copyright Office after it denied protection for an award-winning image he generated using AI.

“We have some information from cases that are in different contexts, where multiple humans are debating who is an author of a work,” said Eric Tuttle, a partner at Wilson Sonsini, speaking on stage in a session titled “Copyright in the Age of AI: Navigating New Frontiers.”

Courts are likely to use that standard as a starting point, but how it applies to AI-generated content is still unclear.

Brian McMahon, senior copyright counsel at Microsoft, emphasized a practical takeaway from the latest copyright guidance. While prompts alone are not eligible for copyright, there may still be a path to protection if human input shapes the final result.

“As a company, if you're using the AI tool in your business, so long as you’re not just cranking out the output, throwing that into the stream of commerce… and you're instead taking that extra step and modifying the output in some way, I think that's a good opportunity,” McMahon said.

He explained that if a human’s expressive contribution can be detected in the AI-assisted output, that work may qualify for copyright protection.

Tuttle then raised a key concern for software companies. If AI tools are generating a significant portion of your code, can that code and the software built from it be protected under copyright law? That’s still an open question that has not yet been tested in court, he says.

For startup founders building with AI, I believe this is the crux. If your AI system is generating code, how much of it can you actually claim? How much human input is required before that code becomes yours?

At this moment, I asked the panelists to double-click on this point: If you're a startup, the whole ethos is to move fast, break things. So, then, what is your pro bono advice with AI coding tools? Is it to not move so fast using Cursor or GitHub Copilot that you put yourself at legal risk? Or is it to just be aware of these cases making their way through the legal system and continue building fast with AI?

Here’s what they told me:

Tuttle said AI is being widely adopted and it is unrealistic to expect developers to avoid tools that improve efficiency when writing code. He emphasized the importance of understanding the risks that come with relying on AI, especially in the context of intellectual property.

When it comes to copyright, Tuttle warned that generating substantial portions of a codebase with AI could make it harder to protect that software. He encouraged founders to consider other ways to safeguard their work, such as intellectual property protections, technological measures, and contracts.

Tuttle pointed to the idea of “autopilot” versus “copilot.” The more humans are involved in revising, editing, integrating, and making decisions about the code, the stronger the case for authorship. He added that being able to document that human input could be important if the code ever becomes the subject of a legal dispute.

Glory Francke, head of privacy and data protection officer at GitHub, advised founders to pay close attention to the tools their developers are using.

She recommended sticking with commercially offered versions of AI tools rather than free ones. Commercial tools are more likely to protect user privacy by not retaining prompts or using them for training.

Francke also urged teams to review and configure their settings carefully. For instance, some tools allow you to block code suggestions that match publicly available code, preventing them from being used in completions. Others offer code referencing features that show the origin of suggested code and its license.

Francke encouraged teams to “get to know your admin” and make use of those settings to reduce risk and stay compliant.

While much about AI is still unknown and outside of our control, Francke said companies can take practical steps now by choosing the right tools, using the right settings, and making sure prompts are not being used for training if that is not the company’s intent.

“Use the tools that are there to help protect against the harms that you're concerned about,” Francke said.

Jonathan Talcott, shareholder at Buchalter, stressed the importance of understanding open-source licensing when using AI coding tools. He advised companies to put systems in place to manage open-source compliance, whether through features built into tools like GitHub Copilot or external scanners like FOSSA.

“You want to be making sure that you have some type of tool in place,” he said, to track and address licensing issues before they become legal problems.

On the topic of IP and copyright ownership for AI-generated code, Talcott said much is still uncertain and will depend on how current legal cases are decided.

“That’s going to be impacted by how these cases play out,” he said.

Still, Talcott echoed what others on the panel made clear. Avoiding these tools entirely is not a realistic option for software startups.

“Unless you want to be left behind,” he said.

Disclosures: I’m no lawyer, and this isn’t legal advice. The speakers quoted here also made clear they were speaking for themselves, not their respective companies. Ascend is a client of Wilson Sonsini.

Investing in Avante

By: Kirby Winfield & Nate Bek

Today, we’re excited to announce our investment in Avante, the infrastructure layer for modern employee healthcare benefits. Fuse led the $10 million Seed round, joined by HighSage Ventures. 

The employer-sponsored benefits market is under strain. HR teams are overwhelmed by administrative work, forced to navigate legacy tools that were never built for today’s demands, while employees are making high-stakes healthcare decisions without the information they need. The current ecosystem relies on a patchwork of point solutions that fail to integrate or scale. 

U.S. employers spent more than $1.3 trillion on healthcare last year, making it the second-largest expense after payroll. Despite its size, this market remains under-innovated and inefficient. Benefits are just too important — too expensive — to be run on spreadsheets and outdated portals.

Rohan D’Souza, Avante’s CEO and co-founder, came to us with a clear read on the market. He had spent months in the field talking to customers, brokers, and benefit administrators. The feedback was consistent. Teams want better insights, more automation, and a way to deliver a high-quality benefits experience without adding headcount.

Rohan’s findings align with what we’ve seen. In a survey of VC-backed startups with ties to Seattle, nearly all reported offering health benefits as soon as they raise a priced round. Coverage is table stakes, even for teams under 10. At larger enterprises, the problem only compounds. Dozens of vendors, thousands of employees, and increasing compliance demands create an environment where small inefficiencies turn into massive operational and financial drag.

Company leaders know it matters — for recruiting, retention, and employee well-being — but the infrastructure to support that coverage is often shallow or nonexistent. 

Avante is building the operating layer companies will need to respond to these pressures. The platform unifies critical data across claims, vendors, costs, and employee engagement. It gives employers a complete, real-time view of their benefits program and helps them make smarter decisions. From plan design to vendor selection to wellness initiatives, everything is surfaced through a single interface with AI-powered insights at the core.

The result is a more scalable benefits operation that improves outcomes for employees and drives efficiency for employers. Avante automates support, integrates cleanly into existing systems, and captures employee sentiment to identify gaps and inform strategy.

We invested, in part, because we believe this team is uniquely positioned to build this product and go after this market. 

• Rohan previously served as Chief Product Officer at Olive AI, where he played a key role as the healthcare tech company scaled to a $4 billion valuation. 

• He’s joined by Co-founder Carly Eckert, MD, Ph.D, an epidemiologist and former EVP at Olive, who now leads innovation and impact at Avante.

• Kabir Shahani, executive chairman, founder of Amperity, one of Seattle’s most successful enterprise startups.

• Nick Cecil, founding head of engineering, brings deep technical expertise from his time as director of software engineering at Salesforce

This is a rare combination of healthcare fluency, product rigor, and enterprise software execution.

With the funding, the team is looking to expand across key functions. Avante is recruiting for five positions: Senior Product Manager, Head of Revenue, Software Engineer, Junior Software Engineer, and Analytics Engineer. All roles are live on the Ascend Job Board.

We believe Avante will define a new standard for how companies manage and deliver employee benefits, and we’re proud to be a supporter from day one.

Dispatch from the Desert: Human[X], Robotics, and the AI Power Shuffle

By: Nate Bek

Vegas felt like an odd place for a conference about intelligence. The artificial kind, or otherwise. But maybe that made it the right setting.

The slot machines never stopped blinking. The escalators kept funneling us between velvet-curtained ballrooms and DJs and mimosa towers to the floor of a city built to overwhelm. There’s something about putting thousands of people in a place designed to keep you distracted — and asking them to think about the future.

Human[X] was the first AI conference of its kind, but it didn’t really feel like a debut. It felt more like a checkpoint. I came in with a few themes top of mind: robotics, AI labs turning into real institutions, and how venture is reacting to the next wave. Those threads ended up tying most of the sessions I attended together.

From left: Hugging Face, Co-founder; Rachel Metz, AI Reporter at Bloomberg; Arthur Mensch, Co-founder & CEO at Mistral AI; and Metz.

The most grounded — and maybe most striking — observation came from Thomas Wolf of Hugging Face.

"I'm pretty sure 2025 is going to be the year where robots start to work."

Not in theory. In kitchens. In warehouses. On wheels. Reinforcement learning breakthroughs are beginning to land outside the lab, and there’s a new confidence about it. He also flagged what’s missing: openness.

“They're all closed-source… In the future, where robots are everywhere, I would love a part of them to be open so that we know exactly what's inside. Maybe we even build them ourselves."

That tension came up in other corners, too. Brad Porter, founder of Collaborative Robotics, was clear about the practical challenges of humanoid robots. Drawing on his time at Amazon:

“We needed robots that could work collaboratively in and around humans. It was not clear how you were going to safety rate these 300-pound mechanical things that can fall over really easily."

What stood out to me was how rarely this kind of specificity comes up in general AI discourse. These weren’t hypotheticals. They were design constraints from people who’ve already tried and course-corrected.

Alfred Lin of Sequoia framed it well — almost offhandedly:

“The desert is not designed for humans. Just like highways and space.”

Form should follow function. The future of robotics may feel novel, but its success will come down to how it adapts to spaces we already occupy.

From left: Mike Krieger, CPO at Anthropic; Moderator: Alex Heath, Deputy Editor at The Verge; and Kevin Weil, CPO at OpenAI.

Mistral’s session with Arthur Mensch made this shift clear. He spoke about deploying vision models on the edge.

“When the robots [are] in the outer world, it won’t have a very strong connection… you want to have low-latency decision making.”

It’s a reminder that not every AI breakthrough lives in the cloud. Edge environments are where robotics will have to operate, and some of the more forward-looking labs are already planning around that.

He also spoke about Mistral’s recent partnership with Helsing, and the demand for sovereign infrastructure in Europe:

“There's a lot of demand to have a regional cloud because that brings more sovereignty and leverage from an economic perspective… As it turns out, we're pretty good at operating GPUs.”

Labs are becoming more than R&D centers. They’re shaping policy, forming defense partnerships, and influencing national strategies.

Anthropic’s Mike Krieger talked through another facet of this transformation — the delicate balance between building products and staying aligned with partners:

“I called basically all of our leading coding customers to give them a heads up that we're launching Claude Code… We're hearing from people using both.”

He was transparent. It was a small window into what it takes to navigate product expansion when your customers are also your ecosystem.

There was a lot of quiet confidence in the VC session with Navin Chaddha, Lauren Kolodny, and Martin Mignot.

“We're still not at the peak of the hype cycle,” said Chaddha. “VCs have enough money.”

He cited $307 billion in dry powder. Valuations, in his view, still have room to move. LPs remain focused on DPI. “It’s a financial services business,” he said toward the end — a reminder that beneath the storytelling, this is all still portfolio math.

Martin Mignot pointed to the upside:

“We can have trillion dollar VC-backed companies when they go public… A lot of value is being captured by VCs.”

The liquidity window may be narrow, but the bets are only getting larger. And many of them are moving into real-world infrastructure — robotics, autonomy, defense. 

In between sessions, I kept thinking about how far we’ve moved from the “chatbot moment.” Human[X] was filled with serious people talking about serious problems. Model architectures were mentioned less than operational realities.

Vegas was a strange backdrop. Too loud. Too bright. Too artificial. But also kind of perfect. A place built on simulation, where beneath the surface you start to feel what’s real.

Early-Stage Hiring, Decoded: What 60 Seattle Startups Told Us

Startups are getting smaller. The average headcount for Series A startups dropped to 15.6 in 2024, down from 18.7 in 2019, according to Carta. The shift is most visible in AI startups, which can scale faster with fewer people thanks to new tooling and automation.

Founders are rethinking team composition from day one. When to hire, who to hire, and what roles to prioritize have all changed.

To bring more clarity to the hiring journey, we surveyed more than 60 VC-backed startups with ties to Seattle. The companies range from pre-seed to Series B and span industries, though many are AI-native.

We asked when key team members typically join — from founders to the first sales, marketing, ops, and support hires. The results offer a snapshot of how early-stage teams are actually built in this market.

Here’s what we found.

---

No surprise — every company starts here. In our survey, every startup had a Founder/CEO at formation, and the role remains the anchor through every stage.

Dilution occurs as a startup grows, but the Founder/CEO most often retains at least 15% equity in their startups in the early stages.

Most startups add at least one cofounder early. At most startups, cofounders are in place by the time a company raises its pre-seed round. After that, it's rare to see new cofounders join.

This is often hire No. 2 or 3. Most often, founding engineers join before the seed round, typically on the heels of a pre-seed raise or right before one.

This role tends to come later. Most startups wait until early post-seed or Series A to bring on their first dedicated sales hire. Founders tend to lead early sales themselves.

Marketing is typically a Series A hire. A few outliers bring someone in earlier, especially if they’re selling to consumers or developers, but most wait until they’ve found some product-market fit.

This hire follows revenue. Startups usually add a customer support or success role around the same time as their first reps — often in the Series A window, occasionally just after.

Ops shows up across the board. Some companies bring someone in early to help the team run smoother; others wait until they hit complexity. Most Series A companies in our survey had at least one ops-focused hire.

Early-stage startups in our survey start offering benefits as soon as they raise a priced round. Seed is the most common trigger. Health coverage is table stakes, but perks like wellness stipends and 401(k)s often wait until Series A or later.

*METHODOLOGY: We surveyed 60 verified, VC-backed startups with Seattle ties. Responses were collected in Q1 2025 and included startups from pre-seed to Series B. While this sample is directional and not comprehensive, it offers a useful glimpse into how early teams form. Take the data with a grain of salt, but we hope it brings more transparency to startup hiring in the region.

Booming: An Inside Look at Seattle's AI Startup Scene

By: Nate Bek

Seattle’s AI startup boom isn’t loud, but it’s reshaping the city’s future in real time. 

Startups are multiplying, talent is clustering, and investors are taking notice. “We are, by any measure, the second-most important city in the United States for AI,” says Kirby Winfield, founding general partner at Ascend. 

While Silicon Valley still dominates, Seattle has quietly built an AI battleground — powered by a deep well of technical talent, the cloud giants that fuel modern AI, and world-class research institutions.

On KUOW’s Booming podcast, Joshua McNichols sits down with Kirby to break down the forces driving this surge, the hidden network of startups shaping the industry, and whether Seattle’s AI moment is built to last.

(Join us and consider donating to support local journalism.)

We’ve lightly edited this transcript for brevity, added references, and clarified a fact. Enjoy! — Nate 👾

---

Joshua McNichols: There's an invisible startup boom happening in Seattle right now. 

Kirby Winfield: I'd say, gun to my head, how many startups are operating today in Seattle with AI at the core of their technology? 1000, at least.

Tons of artificial intelligence companies have cropped up in the area, and they're rivaling sectors like health tech and computer software, but the AI startup scene looks a little different than other industries… Seattle's secret AI startup boom could transform our economy. But are we in a bubble, or is it here to stay? I sat down with an insider to find out.

His name is Kirby Winfield, and he's a venture capitalist for this company called Ascend. I started out by asking him to explain to me exactly what he does. 

I take money from investors, and I give that money to startup founders to help them build the future. I think the best analogy I've been able to make for folks who don't know about venture capital is that we're sort of like betters at a race track. There are lots of horses and lots of jockeys, and there are lots of ways to try to handicap the race. You might look at the horse's training speeds, or you might look at the horse's provenance, but in venture capital, it's much more about the jockey than the horse. So my job is to find the right jockey, the right startup founder, and then they take that technology horse and ride it to victory. That's the idea.

And when you're looking at jockeys, you're focusing on AI?

Our fund is entirely focused on artificial intelligence applied to enterprise problems. So basically, businesses selling to businesses technology solutions using AI.

So you're just the guy to talk to. 

I hope so. 

Okay, so can you rank Seattle's place in the AI world? I mean, where does it sit in relation to other cities?

So we typically say that Seattle is HQ2, or the second headquarters of the US for AI. Obviously, Silicon Valley is HQ1. There's just such a density of technology there. But we are, by any measure, the second most important city in the United States for AI. We have one in four of the AI software engineers in the US living in Seattle. We have the second largest population of software engineers in the US period. And about 65% of AI compute usage runs through the cloud. The three largest providers of which are based in Seattle.

But the cloud can be anywhere.

But the talent that runs the cloud and builds the software that allows the cloud to work for AI lives in Seattle.

OpenAI has opened an office here. We've got the Allen Institute for Artificial Intelligence. We've got Microsoft. These are companies I've heard of, but I've also heard that there's a hidden startup scene here. Can you give a couple examples of AI startups rooted in and around Seattle that could really change people's lives?

Absolutely. Probably the most popular one is read.ai. And this is just a comprehensive AI assistant that sits in your email inbox and in your Word documents and on your Zoom calls and coordinates information for you, so that you're always aware of the context of what you're doing and why you're doing it and what may have been related to it before. And this can work for personal situations as well. It keeps track of conversations and historical data that's associated with a given relationship. So this can be useful for, let's say, you're like me, and you have a child with a disability, and you're coordinating across dozens of service providers and caregivers and schools and doctors. That can be challenging to manage on your own and read.ai has software that puts that all at your fingertips. That can be really life changing for people. 

That one speaks to me because it's so complicated even to do a search in my email inbox for the right correspondence. If I had like a little personal assistant tying all those threads together and bringing together other parts of my life outside of email, that could be useful.

Yeah, it really is like having almost a precognition of what the needs of a certain conversation or experience are going to be, and then populating your dashboard, if you will, with all the information you will need before you even know you need it. It's pretty magical.

I've heard that there's not a lot of investor money flowing to startups in Seattle. So is AI an exception to that?

I'd say AI is an exception to that in almost every market in the world right now, because there really is this kind of Cambrian explosion happening, and so there's been a lot of money coming into AI. It helps that we have Microsoft, Amazon, Google Cloud, Apple's HQ2 for technology. We have, I think, 200 now technical headquarters for large tech companies based here. What happens when you have that kind of technical talent base is that those people want to leave those big companies and start their own little companies, and that does attract investment capital. Not only from people like me and other firms in Seattle, but we did some research, and 90% of the investment money in technology companies in Seattle comes from outside of Seattle.

So why is Seattle a good incubator for AI? Are there reasons besides the labor pool?

I think there are. We have probably the greatest independent center for excellence for AI in the United States, if not the world, with Paul Allen's Institute for Artificial Intelligence. That is a one-two punch when combined with the Allen School for Computer Science at the University of Washington. 

We've backed a number of PhDs out of the UW to start their own companies, and those people have huge influence on hundreds and hundreds and hundreds of very talented PhDs and masters students and computer scientists who then go on and leave this market and go to do their masters at Stanford, or go to start a company in Silicon Valley. And when their mentors start a company here, they come back. And so all of these centers for excellence become magnets for talent. That's one really important aspect. 

Certainly the hyperscalers — which is tech slang for Amazon and Microsoft and Google — also create a real bed of support. It's not just about the talent pool. If you're a software engineer who's an expert in AI, you want to apply that talent to big problems. And some of the biggest problems are ones that are experienced by these big companies that are located in Seattle. And so the problem space, if you're experimenting with AI technology, is really rich and dense in Seattle in a way that it's not really anywhere else in the country, maybe outside of Silicon Valley.

We've seen here in Seattle in the past, when an industry is booming, the benefits extend far beyond that particular industry. What kind of impact could AI have on the economy in Seattle, beyond just the companies themselves and making somebody like you money?

I think AI is poised to have an impact that maybe we don't understand on Seattle. An example that I always use is my daughter, who has Down syndrome. She's 15, and she will not be able to drive a car. Seattle is notoriously not a great place for public transit. For her to be able to take an autonomous vehicle, where I don't have to worry about her safety with regards to a driver, that's life changing for her and it'll be life changing for a lot of other people in this market in particular. 

You think about drug discovery. You think about energy. You think about the impact on maritime such as autonomous maritime vehicles. We've been talking about ferries here for 25 years, and the lack of skilled labor, aging of the ferry workforce, and the lack of persistently serviceable vessels. There's no reason we shouldn't have autonomous ferries in the next 10 to 15 years. 

I think a lot of what's going to change in Seattle is going to change a lot of other places, too. But there are those areas where I think we're probably going to see a little bit more impact just because of how our geography is positioned

Quick side note: I know for public transit nerds, they probably would take issue with the idea that public transit is bad here. We got Light Rail opening all over the place, all that kind of stuff. But I take your point. It's a car-oriented city. That's its legacy that it's trying to overcome.

We don't have the time to debate this here. The ridership numbers might suggest otherwise.

A topic for another time… So China rattled the markets recently with news that it's developing new, cheaper AI models like DeepSeek, for example. How could competition from China and other places affect our startup scene here in Seattle?

My perspective on China and DeepSeek, in particular, which is a new large language model, is that it was very disruptive because it actually outperforms some of the big closed-source models, like Open AI and Anthropic. So when it comes to that, I actually get excited, because if these state-of-the-art open source, low cost to free models become ubiquitous, that actually means that there will be a Cambrian explosion of AI applications and agents developed by companies that could not otherwise have existed because it's cost prohibitive to conduct the training of these models for an individual company. There's actually this paradox, Jevon's paradox, when the resource becomes freely available, there's an initial decrease in the volume of use, but then it should exponentially explode, just because it's so broadly available.

So DeepSeek and other cheaper, large language models could actually result in a startup explosion here, whether or not this is good or bad for Microsoft or other big AI companies here .You see this as being good for startups? 

Absolutely. 

One thing AI is making it easier to do is to start companies with very few employees, because you can use generative AI to do a lot of your engineering. So what does the low number of employees needed to start a company today mean for jobs in this region? Could we have an AI startup boom in Seattle that we don't really feel?

I think that's entirely possible. We're already seeing companies that we're backing hiring fewer people to do the same work. *Instagram, when it was bought by Facebook, had 13 employees — and it was bought for $1 billion. And that was before generative AI. So I absolutely think it's not a matter of if, but when, there's a billion-dollar company built by one person. 

Wow. 

That said, I do think one of the only things that's not going to be replaced are functions within a company that are inherently human. I was talking with someone today about the radical increase in AI-generated inbound sales emails. And there are all these startups that have raised hundreds of millions of dollars to build an AI sales representative. And what happens is, as a potential buyer, you realize it's all AI generated, and you delete everything. Before, if I got a cold sales email, I had to in the back of my mind, at least think, "Well, someone took the time to write this email and to notice something about my company or me that made them think that maybe I would buy their product," because they wouldn't do it if they didn't think I would, because their job is to sell their product. So there was this qualification that you could tell it happened by a human and it made you lean in a little bit and say, "Well, maybe not today." But if they email me again in six months or whatever, maybe I'll reply and say, "No, thank you," because I'm a nice person. And if I had their job, which I have, I would want someone to reply. That has been wiped off the face of the earth now. 

So actually what becomes valuable is warm, human introductions and personal networks. Which then begs the question of the return to office in this virtual world we live in and relationship building. And I think human relationships become the capital that's probably the most valuable in the next five years, at least, and I do worry a little bit about what that means for folks who are in the first decade of their career and may not have accumulated that may not know how to.

I've often thought that as AI progresses and becomes indistinguishable from humans, we will just move the bar — redefine what it means to be human, or we'll just keep moving the goal posts and cherish those things, or even fetishize those things that are sort of unique to human. Even if that means being chaotic.

I totally agree. 

We seem to go through these waves where everyone is investing in something, and then there's too much of it, and then the bubble pops. Can we just be in another one of those cycles?

We, without a doubt, are. There's nothing new under the sun, as Shakespeare said. I lived through the commercial Internet boom and bust, and operated startups through it. I lived through the mobile era of boom and bust and the cloud era. And now we're in the AI era, and just like those others, what you get is a very big rush of capital and excitement at the very beginning, when the technology is new. And the challenge is that that capital is chasing ideas that are not very well formed about what the future is going to look like. 

The future gets clearer. As it gets clearer, the costs for the technology tend to come down, which, combining that future being clearer and the cost coming down, results in a completely different landscape than what existed when the wave started to build. So that wave crests, and then the next wave comes. And I think we are just now at the cresting of the first wave, and the beginning of the next wave. And the next wave, I think, resembles the wave that came after the internet bust, where 80% plus of the commercial value on the internet today was created. And so that's what's exciting about investing in AI, at this moment in time is that I do think we're at the beginning of the real growth.