Ascend logo

Ascend

AI, Software, Robotics · Preseed

Ascend is a pre-seed venture capital fund that invests early in technology founders building companies in AI, software, and frontier technologies such as robotics and intelligent infrastructure. It provides capital, operating experience, and network support to help founders solve hard technical problems and turn new possibilities into products people need. Its mission is to back the next generation of technology entrepreneurs from the first check forward.

Facts

HQ
Seattle, WA, US
Sectors
AI, Software, Robotics, Autonomy, Defense, Intelligent Infrastructure
Founded
2019
Employees
2–10
Members see funding amounts.
.
Funds raised
3 funds

Ascend series: Seattle’s ‘Third Act’ — physical AI startups on Lake Union signal a shift toward software-enabled hardware

The piece reports from Lake Union to argue Seattle is entering a third industrial era—'physical AI'—where hardware-enabled intelligent machines mix with legacy maritime and aerospace industries, exemplified by startups like Overland, Anduril, Brinc, and Endurance Energy. The authors say Seattle has the talent, infrastructure, and institutions to lead but needs more risk-taking, investor capital for hardware, and pro-innovation policy to fully capitalize on this shift.

Ascend finds 60+ non-Seattle–headquartered C-suite execs live in Seattle area, signaling concentrated tech leadership

Ascend identified more than 60 C-suite executives who live in the Seattle metropolitan area while working for technology companies headquartered elsewhere, spanning AI, cloud, cybersecurity, fintech, and enterprise software; many previously held senior roles at Amazon, Microsoft, Google, or similar employers. The research, compiled from public sources, argues that executive residence signals concentrated talent, networks, and potential startup activity in Seattle and will be updated over time.

1 commentary post

Token Talk 38: Author Warns AI Chatbots Can Foster ‘AI Psychosis’ by Validating Delusions and Lowering Epistemic Standards

The article argues that modern AI chatbots can validate and amplify users' delusions—what the author calls “AI psychosis”—by reflecting and reinforcing fears, lowering epistemic standards, and creating frictionless interactions that deepen maladaptive thinking, sometimes with tragic outcomes. It links historic computer paranoia to today’s systemic risks from LLM hallucinations, model training conflicts, and design choices that encourage overreliance.

Seattle’s UW research, Amazon alumni and military adjacency fuel startups building high-stakes, regulation-ready AI for biotech, finance and government

A Seattle high-school intern argues that strict school AI rules illuminated why AI must be disciplined in high-stakes domains, and describes how Seattle's research institutions (notably UW and the David Baker ecosystem), Amazon alumni, and military adjacency produce founders building accountable, regulation-ready AI for biotech, finance, and government.

Ascend joins SNR Capital in $2M seed for Limelight to automate personality-led B2B marketing

Ascend invested in Limelight’s $2 million seed round alongside lead investor SNR Capital; Limelight provides an AI platform that automates creator discovery, activation, and ROI measurement to scale personality-led B2B marketing. The company plans to expand into an autonomous AI-driven B2B marketing workforce, and founders David Walsh and Youngjae Ji bring deep go-to-market and technical experience.

From Galveston 1900 to GraphCast: AI models like DeepMind’s pinpoint Hurricane Melissa’s unexpected Category 5 intensification, reshaping forecasting

The article traces the evolution of weather forecasting from early guesswork and the 1900 Galveston hurricane through Richardson’s numerical ideas and mid-20th-century supercomputing to modern AI systems like DeepMind’s GraphCast that now outperform physics-based models for many forecasts. It highlights AI’s speed and recent success predicting Hurricane Melissa’s rapid intensification while noting limitations (smoothing extremes) and the emerging hybrid AI–physics frontier.

Author Thomas Stahura warns Sora 2’s invite‑only AI video and cameo feature fuels a ‘slopapocalypse’ that blurs fiction and reality

Sora 2, an invite-only world-model AI for video, rapidly generated highly convincing short clips and a cameo feature that lets verified users insert themselves or others, blurring lines between fiction and reality. The author warns this surge in synthetic, engaging content — dubbed the "slopapocalypse" — will make curation, verification, and sustaining genuine creativity urgent priorities as AI video becomes ubiquitous.

Thomas Stahura explains how scaling laws, FlashAttention, Mixture of Experts and speculative decoding cut memory costs and sped inference

The article traces how model and hardware co-evolved since GPT-2—scaling laws drove huge dense models that demanded high HBM bandwidth, prompting software and architecture innovations like FlashAttention, quantization, Mixture of Experts (MoE), and speculative decoding to reduce memory traffic and speed inference. These techniques shift bottlenecks between memory, communication, and on‑chip compute, and could eventually make powerful models more efficient and even runnable on smaller devices.

Token Talk 32: Technological Telepathy — EEG startups and invasive ECoG/Neuralink arrays push BCIs from lab demos to real-time speech and image decoding

The article surveys the current state of brain-computer interfaces (BCIs), comparing non‑invasive EEG systems (like Emotiv, Neurable, OpenBCI) with invasive methods (ECoG, Neuralink) and explains the signal-processing pipeline that maps brain waves into actionable commands. It highlights experimental successes—real-time speech decoding and image reconstruction using invasive recording plus AI—arguing the challenge is now engineering and data science rather than impossibility.

Author argues AI isn’t collapsing into a bubble despite MIT and McKinsey studies showing limited enterprise ROI; employee use of personal LLMs driving ground-up adoption

The author reflects on the current state of AI after visiting Berkeley and hacker houses, arguing that progress has slowed from explosive expectations to steadier, research-driven improvements—especially in agents and tooling—while media claims of an "AI bubble" often misinterpret survey timing and employee adoption of personal AI tools. Studies cited (MIT July 2025, McKinsey July 2024) show limited enterprise returns from formal GenAI initiatives, but widespread employee use of personal LLMs suggests ground-up traction despite shifting public sentiment and fatigue.

Token Talk 29: Thomas Stahura Says GPT-5 Won’t Kill Enterprise SaaS — Moats Still Live in Data, Compliance, and Integrations

The author argues GPT-5, while technically strong and capable of rapid prototyping, won’t make enterprise SaaS obsolete because the real value and moat lie in data, reliability, compliance, integrations, and post-demo workflows. Open models are closing the gap with closed ones, OpenAI released two open-source gpt-oss models, and the market should expect SaaS vendors to survive by owning the ‘messy middle.’

USVP and Madrona co-lead $15M Series A in Clarify to build an autonomous AI-native CRM

USVP and Madrona co-led a $15 million Series A in Clarify, an AI-native autonomous CRM founded by Patrick Thompson that automates call briefs, transcriptions, action items, follow-ups and pipeline maintenance. Clarify, seeded in Q1 2024, now serves 100+ customers, has ~20 employees, uses usage-based pricing with unlimited users, and aims to replace legacy CRM workflows dominated by Salesforce and HubSpot.

Governments and Cloud Giants Ramp Up Quantum Spending as VCs Stay on Sidelines Ahead of Hypothetical ‘Q‑Day’

The article argues that while governments and cloud giants are rapidly funding and preparing for a potential ‘Q‑Day’ when quantum computers could break current encryption, most venture capitalists view quantum as a long‑horizon, noninvestable science play lacking near‑term commercial buyers; corporate VCs, defense primes, and hyperscalers are the primary active investors today.

Neotribe-led $13.7M Series A in Yoodli to Expand AI Roleplay Enterprise Platform Serving Google, Databricks and More

Neotribe Ventures led a $13.7 million Series A in Yoodli, an AI roleplay enterprise platform that helps users rehearse high-stakes conversations with real-time feedback; existing investors including Madrona, Cercano, AI2 and Neotribe (who previously invested) highlighted Yoodli’s enterprise traction with customers like Google, Databricks, RingCentral, Snowflake and BDO. The funding will expand sales, customer success, and engineering while improving personalization and realism in roleplays to address communication training gaps at scale.

Fuse leads $10M Seed in Avante to build AI-powered infrastructure for employer healthcare benefits

Fuse led a $10 million Seed round, joined by HighSage Ventures, in Avante—an infrastructure platform that centralizes benefits data and uses AI to help employers manage employee healthcare more efficiently. Avante, led by CEO Rohan D’Souza and co-founder Carly Eckert, plans to expand its team and hire five roles as it scales its platform to reduce administrative burden and improve benefits outcomes.

Survey of 60 VC-backed Seattle Startups Finds Series A Headcount Averaged 15.6 in 2024; Hiring Peaks for Sales, Marketing, and Support Shift to Post-Seed/Series A

A survey of 60 VC-backed Seattle-area startups (pre-seed to Series B) finds smaller early teams—average Series A headcount fell to 15.6—and shows common hiring patterns: founders/cofounders first, founding engineers before seed, sales and marketing around post-seed/Series A, ops sprinkled earlier, and benefits typically starting at priced rounds.

People

Online