On July 21 Carta published a seed-valuation breakdown with an uncomfortable shape. Median pre-money for seed-stage software in the Bay Area hit $33.3M — a record. New York reached $24.5M, up roughly 80% year over year. The rest of the United States sat at $12.7M, essentially unchanged over the same period. That is a 2.6× premium for the Bay Area and 1.9× for New York, on the same stage, in the same sector, in the same year. The gap isn't the news — everyone knew the coasts pay more. The news is the divergence: New York repriced 80% in twelve months while the rest of the country did not move at all. Read it precisely: this is US-only, software-only, median pre-money, and the underlying data is Q1 2026.
The story everyone is telling about this is that AI ate the seed market, and the data does not support it. Pre-seed is the least AI-saturated stage in venture — roughly half of pre-seed dollars, against 60.7% of all capital on Carta and 83% of SaaS capital. AI's share rises with stage; it does not start at the bottom. Non-AI funding did not fall either, holding flat at about $214B year over year, which makes the share decline a numerator effect rather than a squeeze. Last week we argued the ceiling on what anyone can charge for capability had collapsed ↗ — this week the question is who gets to charge anything at all, and the answer turns out to depend less on what you build than on where you file your incorporation papers.
Carta's July 21 analysis of seed-stage software valuations found the Bay Area at a record $33.3M median pre-money and New York at $24.5M — up about 80% year over year — against $12.7M for the rest of the United States, roughly flat over the same period. That is a 2.6× and 1.9× premium respectively, at the same stage, in the same sector. The caveats matter and are worth stating plainly: this is Carta's own customer base, US-only, software-only, a median rather than a mean, and drawn from Q1 2026 data rather than the quarter just ended. One more caveat cuts at the number itself: Carta reports valuation, not dilution — a higher valuation usually arrives with a larger round, so the share a founder actually sells may be far closer across the two markets than 2.6× suggests. What makes it the week's most useful number anyway is that it is first-party, seed-specific, and measurable — which is more than can be said for the sector-capture narrative it quietly undercuts. Two founders with comparable companies, comparable traction and comparable teams are now raising into markets that price them 2.6× apart. Neither of them is being told which market they are in.
Crunchbase reported on July 23 that roughly $320B — 60% of the $510B raised globally across all stages in the first half of 2026 — went to rounds of $1B or more. The figure that didn't make the headline is the domestic one: in the US alone the share is 73%, about $290B, materially more extreme than the global average implies. This is a cumulative half-year measure of dollars, not a run-rate and not a share of deals. It is the cleanest available statement of what concentration actually looks like: not one sector absorbing everything, but a small number of enormous rounds absorbing most of the money.
CB Insights' State of AI report for Q2 2026, published July 23, found that 89% of AI funding went to 142 mega-rounds — rounds of $100M or more were only about 6% of AI deals yet captured $132.5B. Set that beside the measurement problem underneath most sector-share commentary: AI is a vertical tag, not a mutually exclusive industry, so a single dollar can be counted as AI and as software and as fintech simultaneously — which is why AI share and software share cannot be stacked or compared directly. The discipline check is deal count: AI took 65.4% of US deal value in 2025 but only 39.4% of deal count, and half of that year's total deal value went into 0.05% of completed deals. The sector is not absorbing the market. A handful of rounds inside it are.
Crunchbase's Unicorn Board update on July 22 recorded 34 companies crossing $1B in June, adding more than $110B in value. Ten were frontier AI labs, collectively valued at about $65B, with DeepSeek the highest-valued new entrant at $50B. Sixteen were US companies and eight were Chinese. Read alongside the concentration data, this is the same phenomenon viewed from the other end: value creation clustering into a small number of very large names, in a small number of places, over a short window.
1872 closed a $15M seed round on July 22 led by The O.H.I.O. Fund to build an autonomous steel fabrication factory in Cincinnati. The founding team is three former SpaceX engineers, and the plant uses robotics from Path Robotics, with full autonomy targeted for 2027. The company's own framing — among the largest seeds in Ohio history — is PR language rather than a verified record, and the 320,000-welder shortfall it cites belongs to the American Welding Society. What isn't PR is the structure of the deal: a substantial seed, in the Midwest, led by a fund built specifically to keep that capital local.
The British Business Bank committed £10M to Odyssey Ventures on July 23 through its Regional Angels Programme, alongside Odyssey Discovery I, a $50M pre-seed and seed fund operating between London and San Francisco. The stated thesis is explicit and worth quoting in substance: to keep UK AI and deeptech founders from relocating their headquarters to the United States in order to raise. Note the structure — this is a commitment made alongside the fund via a regional programme, not a £10M vehicle in itself.
Humanoid raised a $152M Series A at a $1.35B post-money valuation on July 21, led by Prime Movers Lab with Schaeffler, Bosch, Fubon FHVC and Aglaé Ventures participating, bringing total funding to about $270M. The company is London-based and vertically integrated, building both the hardware and its own control stack. A note on the figures: the native currency is USD — the €133M and €1.1B numbers circulating in European coverage are conversions, the reverse of the usual error. Its billing as Europe's first pure-play humanoid robotics unicorn is the company's own claim.
Gritt announced $32.4M in total funding on July 21, but the number worth holding is the structure: a $26M Series A led by Obvious Ventures with USV and Active Impact, stacked on a $6.4M pre-seed from First Round, Climactic, Congruent and VSC. The company retrofits robotic arms and AI onto existing jobsite equipment — skid steers, forklifts — beginning with large-scale solar installation rather than construction generally. Combined-total headlines like the $32.4M are among the most common ways early-stage benchmarks get distorted, because they make every comparable round look smaller than it was.
Atoms raised $1.7B on July 22 in a round led by Andreessen Horowitz, with Ben Horowitz joining the board and Bain Capital, Fifth Wall, Chemistry, A*, K5 Global, Abstract, SV Angel and — notably — Uber participating. The company spent eight years in stealth building software, sensors and robotics for physical industries including food, mining and transport. Kalanick's framing of it as a conglomerate built like computers is his own. The under-discussed detail is Uber investing in the new company of the founder it removed.
Bloomberg reported on July 23 that Khosla Ventures is targeting $5.5B across a new set of funds — about $1B for seed, $2B for early stage and a $2.5B opportunity fund. The status matters: this is a target in market, not a closed fund, and fund targets move. Set beside IVP out raising $1.8B for its nineteenth fund the same week, the pattern is capital consolidating into established brands — including the capital that will be deployed at seed. Dedicated seed vehicles were otherwise close to absent from the week's announcements.
Cut Through Venture's Q2 2026 data, reported by Forbes Australia on July 22, found Australian rounds under $5M fell to just 31 in the quarter — a 44% drop from the 2025 quarterly average, and the lowest early-stage deal count since Cut Through began collecting data in 2020. In the same quarter, total capital raised rose 60% year over year to $1.7B. Median pre-seed was $1.3M and median seed $4M. These are Australian figures and likely denominated in AUD, so they are not directly comparable to US data. What transfers is the shape, and it is this week's thesis in miniature: record-low deal count alongside rising capital — fewer doors, each with more money behind it. Reporting attributes part of the squeeze to AI-native startups seeking larger initial rounds.
Augustus raised a $180M Series B at a $1B valuation on July 21, led by Tiger Global with Hummingbird and QED participating, bringing total funding to about $210M. It is not an AI company. It is a federally chartered bank giving international fintechs and banks direct access to US dollar accounts and rails, founded in 2022 and already processing for Kraken. It matters here because it is evidence against the week's dominant narrative: non-AI venture funding held flat at roughly $214B year over year, which makes AI's rising share a numerator effect rather than a drought everywhere else.
The Wall Street Journal reported on July 26 that Nvidia is in talks to guarantee roughly $250B in lease and construction financing so OpenAI can lease a 10-gigawatt facility being built in southern Ohio, with a parallel chip-financing discussion that could reach $350B. Two precision points the coverage often drops: the guarantee excludes chips — it underwrites the customer's capacity to take delivery, not the purchase itself — and the arrangement exists because OpenAI cannot obtain an investment-grade rating on its own. This is also not new. Nvidia has disclosed partner lease guarantees since its Q3 FY2026 10-Q in November 2025, growing to a $3.5B book by January. Michael Burry, who has been short this thesis since last year: “Around and around we go.”
The easy read is that San Francisco and New York are winning and everyone else is losing. The precise one is stranger: the same company is being priced twice. Carta's number is a median pre-money for seed-stage software — $33.3M in the Bay Area, $24.5M in New York after an 80% year, $12.7M everywhere else. That last figure didn't fall. It didn't move at all.
The story attached to this all week was that AI ate the seed market, and the data says otherwise. Pre-seed is the least AI-saturated stage in venture — roughly half of pre-seed dollars, against 60.7% of all capital and 83% of SaaS capital. AI's share rises with stage. Non-AI funding didn't collapse either; it held flat near $214B, which makes the falling share a numerator effect. What actually concentrated wasn't sectors. It was rounds and rooms: 60% of global funding went to rounds of a billion or more, and half of last year's deal value landed in 0.05% of deals.
Now the case against my own argument. This is one dataset — US-only, software-only, median pre-money, drawn from Q1. Carta backfills as customers join, so any year-over-year read on its data compares a fresh number against a matured one. And a valuation gap is not automatically an opportunity; a market can clear lower simply because fewer buyers are in it. The test is dated and close: Carta publishes again in mid-August. If the rest-of-US median moves toward the coasts, this was noise and I was wrong.
But hold what the number is. It is a price gap, not a quality gap — a measure of how many people are bidding, not of what got built. And these are not one market at two prices; they are different worlds. A company in Cincinnati selling to Midwest manufacturers has customers a Bay Area founder can't get a meeting with, twice the runway per dollar raised, and engineers who don't leave in eighteen months for the competitor four blocks away. Those aren't compensations for a lower valuation. They're assets the coastal comparable has no column for.
Which leaves the part worth acting on. The rest-of-US median is flat because nobody has repriced it — the comparable that will define that market has not been set yet. You set it. You set it by building the company that makes an investor argue with their own partnership about the number, in a city where that argument hasn't happened recently. Three former SpaceX engineers just raised $15M in Cincinnati from an Ohio fund; a British state bank is now paying to keep founders from getting on a plane. Those are the early edges of a market being written down for the first time. You can spend this year negotiating against a number somebody else set in a city you don't live in — or you can be the deal everyone in your market benchmarks against next year. One of those is a discount. The other is a comparable.