BriefTechNews
Pre-seed caps jump 33% as deal counts crater: the new math of early-stage funding

Pre-seed caps jump 33% as deal counts crater: the new math of early-stage funding

9 min read · 15 sources

TL;DR
  • Pre-seed investment hit $3.58B across 11,726 deals in Q3 2026, the most capital but fewest deals ever recorded.
  • Median pre-seed caps for $2.5M+ rounds jumped 33% year-over-year, from $30M to $40M.
  • Venture funding now requires $1M to $100M ARR in five years, a 151% compounded annual growth rate.
  • AWS launched credit splitting and transfers, allowing permanent division of credits across linked accounts.
  • Vercel X accelerator offers no-equity entry with a $9M+ credit pool for 40 AI pre-seed teams.

The pre-seed market just did something weird: it raised the most money ever while making the fewest deals on record. Carta’s Q3 2026 data shows $3.58B flowing into 11,726 SAFEs and notes, on pace to finish the year above $4.48B across 14,600+ instruments. That’s the largest quarter by capital and the smallest by deal count simultaneously - a concentration that tells you exactly where the money is going, and who is getting left out.

The numbers behind that split matter more than the headline. Median post-money caps rose across every tier, but the big jump was in large rounds: deals above $2.5M saw caps climb 33% year-over-year, from $30M to $40M. Meanwhile the Western U.S. pulled in 68% of pre-seed dollars, up from 58%, with the Bay Area alone raising $1.7B. If you’re a founder outside that geography, the data says you’re competing for a shrinking slice.

Only 29% of seed rounds convert below the cap, meaning most SAFE investors get their minimum ownership.

Pre-Seed Caps Are the Price, Not the Ceiling

A SAFE defers pricing, but the cap is where most investors actually end up paying. Carta’s data shows only 29% of seed rounds convert below the cap - meaning for 71% of SAFE holders, the cap is the effective price, not a ceiling that gets negotiated down. Seed valuations land above the cap often enough that founders should treat the cap as the likely outcome when negotiating.

The 33% jump in caps for larger deals changes the dilution math. A founder raising $2.5M at a $40M cap gives up roughly 6.25% before any discount; at last year’s $30M cap that same check cost 8.3%. The trade-off is that the bar for what counts as a “larger” pre-seed round keeps rising, and the AI-driven concentration of capital means non-AI founders face a tougher market with caps that haven’t moved as much.

The Indie-Friendly Die-Off Is a Dependency Risk

Source: cassidoo.co ↗

Why are all the indie-friendly products dying at the same time? documents a grim 2026: Mem, Lemon Squeezy, Val Town, Deno Deploy, and Pulley all shut down or got absorbed. It’s a record year for venture-backed shutdowns, and the causes are structural, not random. The easy fundraising environment of 2021 is gone, VC money has shifted overwhelmingly to AI, and AI makes it trivial to replicate a SaaS feature.

The acquisitions are the part that should worry you as an engineer: Stripe bought Lemon Squeezy, Cloudflare bought Deno. When a platform vendor acquires the tool you depend on, your roadmap becomes their M&A strategy. The survivors share a pattern - bootstrapped and profitable, local-first and self-hostable, or API-first platforms that work for humans and agents alike. When you pick a dependency, check whether the business model survives without VC oxygen. Data portability is no longer a nice-to-have; it’s your exit plan.

The New VC Growth Bar: $100M in Five Years, or Don't Bother

Source: saastr.com ↗

SaaStr’s framework for raising venture capital today is brutal: go from $1M to $100M ARR in five years, max. That’s 151% compounded annual growth - and it quietly kills the old T2D3 playbook (triple, triple, double, double, double), which finishes at $72M, 28% short. The clock starts at $1M ARR, not at founding.

The bar is set by public-market economics. Sapphire Ventures’ 2026 report puts the IPO threshold at $400M - $500M+ revenue with 20%+ growth. Figma’s IPO at $749M revenue growing 48% still saw its stock drop 20% when growth fell to 41%. Lower revenue multiples mean you need more revenue to reach the same exit value, and VCs push that requirement back into earlier rounds. Seed rounds still price on team and wedge, but the growth filter is what determines whether you get a Series A at all.

AI Is Now the First Stop in the B2B Buying Journey

Source: growthunhinged.com ↗

Research from Profound’s Prompt Volumes dataset, analyzing 27,374 real user prompts across six SaaS categories, shows the buyer journey now starts with an AI chat, not a homepage. 63% of buyers added a new vendor to their shortlist via AI despite already having candidates, and the chosen vendor was almost always in the buyer’s first AI answer. Pricing is the biggest single topic, especially for coding agents and website builders.

The practical takeaway: 73% of buyers do visit websites after the AI chat, but the shortlist is already set. If an AI model misstates your pricing or features, you’re dropped before you get a pageview. Optimize for AI discoverability - ensure your product is accurately represented in training data and that your site’s content is structured so models can cite it correctly. The author’s own experience is telling: Claude suggested Cloudflare and Lovable, and he trusted the recommendations without further research.

AI-Enabled Roll-Ups Go After the Services Dollar

Source: alexandre.substack.com ↗

A primer on AI-enabled roll-ups makes the case that selling software to service firms leaves most of the money on the table. US accounting spends $6.5B on software versus $220B on services - a 1:34 ratio. Legal is even starker: $8.1B software versus $447B services, 1:55. The playbook: acquire existing firms, run them on your AI platform, and use the gains to serve more clients with the same staff.

Eudia (legal, backed by General Catalyst) and Current (accounting, backed by Thrive Holdings) are pursuing this. The logic is that owning the service outcome forces AI adoption - you can’t ignore the tool when your margin depends on it. Acquisition also serves as cheaper distribution in fragmented markets, like the ~9 - 10k HOA management companies serving 375k community associations. For engineers, the question is whether building vertical software or buying and optimizing service firms is the better path to dominance. The labor budget dwarfs the software budget in every services industry.

AWS Credit Splitting Kills the FinOps Spreadsheet

Source: aws.amazon.com ↗

AWS launched credit splitting and transfers, letting you divide any credit and transfer portions to linked accounts within your payer account. Splits can be by dollar amount, percentage, or even splits, with each inheriting the original’s expiration date and service eligibility. Transferred credits get first-priority consumption in the receiving account. The feature works via Console, CLI, or API, with up to 100 splits per credit family and a $100 minimum per split.

This eliminates the monthly manual chargeback work for FinOps teams - no more spreadsheet tracking of who burned through their startup credit. The catch: splits are permanent with no undo, and partially consumed credits split only the remaining balance. It composes with existing credit sharing and pausing features. If you run a multi-team AWS organization on startup credits, this is the tool you’ve been manually recreating for years.

Vercel X: No-Equity Accelerator With a $9M Credit Pool

Source: vercel.com ↗

Vercel launched Vercel X, a six-week accelerator for AI founders at idea and pre-seed stages, running January 19 to March 2, 2027, in San Francisco. The cohort of 40 teams gets technical mentorship, access to a $9M+ pool of credits from Vercel, AWS, Anthropic, OpenAI, and other partners, plus a Demo Day with investor access. No equity taken. Vercel Ventures will fund winning teams.

Applications close November 29, 2026. You must be a Vercel customer, have no institutional seed round, and be a solo founder or team of up to three. For engineers with early AI ideas, this is a no-dilution path to credits and mentorship - the main cost is six weeks in San Francisco and building on Vercel’s stack.

GitHub Team Gets Free Automated Security Risk Assessments

Source: github.blog ↗

GitHub security risk assessments now run automatically for organizations on the GitHub Team plan, at no additional cost. Organization admins and security managers can review findings covering exposed credentials and code vulnerabilities, see current risk, and prioritize fixes. This extends a feature previously limited to higher tiers.

For teams on GitHub Team, this is a free credential-leak scanner that runs without configuration. The value is the single view: exposed secrets and code vulnerabilities in one place, prioritized by risk. Enable it and review the findings before an exposed API key becomes a breach notification.

AI Agents Are Not Your Next Customers

Source: pawelbrodzinski.substack.com ↗

Despite the hype, AI agents are barely transacting. Uber saw 4 agent-ordered rides in a quarter. Uber Eats had ~500 agent orders out of 320 million in a week. DoorDash had several hundred agent deliveries out of 10 million daily. Airbnb calls agent bookings “very non-material.” Stripe reports 38x growth in agentic purchases in a month - without absolute numbers, which suggests a tiny base.

The author’s argument is that this isn’t technical - MCPs and agent-friendly interfaces exist. It’s psychological: people prefer to make purchases themselves. The infrastructure is ready; the behavior isn’t. For engineers, treat agentic commerce as a future possibility, not a near-term revenue channel. Build the interfaces if you want, but don’t bet the roadmap on agent customers.

Source: searchenginejournal.com ↗

Google Ads is migrating more Search campaign functionality into AI Max. Automatically created assets and campaign-level broad match transitioned in September 2026, with Dynamic Search Ads following in February 2027. Migrations preserve existing settings but enable different AI features per campaign type: ACA transitions with search term matching and text customization, broad match with search term matching, and DSA with all three plus Final URL expansion.

The trap is that two campaigns both showing “AI Max enabled” can have very different automation levels. Audit at the campaign and ad group levels: check search term matching at the ad group level, and text customization plus Final URL expansion at the campaign level. The migrations are rolling out regardless of your preference - the question is whether you know what’s active in your account.

Contexting vs. Context: Why Experts Can't Explain Themselves

Source: cutlefish.substack.com ↗

The tension between “contexting” and expertise is that highly experienced people often can’t describe their decisions because expertise operates in the background. The article shares a product leader who was frustrated by reports’ vague questions, then caught themselves doing the same with their own manager. With AI hype around “sharing context,” the ability to make reasoning explicit matters more - but experts may not need to practice these meta-skills consciously.

For engineers, this is the communication gap that shows up in interviews, design reviews, and incident post-mortems. The person who knows the system best is often the worst at explaining why they made a call. With AI agents increasingly needing explicit context, the ability to articulate reasoning becomes a technical skill, not just a soft one.

The Highest-Revenue Email Looked Like a Personal Note

Source: audiencebridge.io ↗

An ecommerce company’s highest-revenue email was a simple, personal-note-style message with one clear call to action - replacing their usual beautifully designed emails. The offer was identical; only the presentation changed. Freelance email marketer Dylan Bridger found similar results with sponsor placements: ones that feel like part of the publication outperform those that feel dropped in, based on a small test across six newsletters.

The lesson for engineers shipping product updates: plain language beats polished design. The “Direct, Indirect, Direct” approach - announce the offer plainly, then return to direct promotion - outperforms clever copy. In email, as in APIs, the interface that gets out of the way wins.

Get the brief

Liked this one? Pick the beats you care about and get just those, five minutes a morning. Free, no hype, one click to leave.

About Me Author

My name is

BriefTechNews

A daily digest of what actually moved in AI, tech, crypto and fintech, assembled and written with AI, with every item linked to its source. Read More
Tags

You May Also Like