On August 4, Bending Spoons agreed to buy Airtable at an enterprise value of $1.285 billion — about 2.7x its roughly $480 million of ARR, which is still growing over 20% a year. The business was never the problem. In December 2021 the same company raised $735 million at an $11.7 billion post-money valuation, and at any realistic multiple that mark required several billion dollars of revenue to clear. It had four hundred and eighty million.
What makes this week instructive is not the markdown — the secondary market had already repriced Airtable to about $4 billion earlier this year. It is how the gap got resolved. Bending Spoons' SEC filing records that before the purchase agreement was signed, assets and liabilities relating to the “Hyperagent” business line were transferred out to a separate company — the AI platform Airtable launched in February, kept by its founder. The proven past was sold at the market's price to clear a $1.4 billion preference stack; the uncapped future left in a new vehicle with a clean cap table. That is the shape of an unclearable mark being unwound. And in the same seven days, fourteen companies closed billion-dollar rounds — the most in any month on record — each one a new promise about a multiple nobody has been asked to pay yet.
Bending Spoons will acquire Airtable in an all-cash deal valuing it at an enterprise value of $1.285 billion, which together with Airtable's net cash-and-cash-equivalents balance implies an equity value of approximately $2.25 billion, subject to closing adjustments. The gap between those two numbers is roughly $965 million of cash the company never spent. Airtable had raised about $1.4 billion since 2012 — some $1.2 billion of it during 2020–21 — and peaked at an $11.7 billion post-money in December 2021 on a $735 million Series F led by XN, with Silver Lake, Salesforce Ventures, T. Rowe Price, Franklin Templeton and MSD alongside. The preferred is 1x non-participating, so the preference stack clears and the Series F recovers its money and nothing more: $735 million in, $735 million out, over roughly five years. The company Bending Spoons is buying is healthy — 500,000+ organizations, 80% of the Fortune 100, ~$480M ARR growing over 20%. But it is not the whole company that existed a week ago. The SEC filing states that “assets and liabilities relating to the ‘Hyperagent’ business line were transferred by the Company to Hyperagent Inc.” before the agreement was signed — the agentic AI platform Airtable launched in February 2026, retained by founder Howie Liu. Neither the acquirer's press release nor most of the coverage mentions it.
Global startup funding reached $65 billion in July, up 100% year over year, and fourteen companies raised billion-dollar rounds — the highest count in any single month on record, per Crunchbase. The largest was a $10 billion investment in Blue Origin, its first external financing. Every one of those rounds sets a mark, and every mark is a claim about a future multiple. Airtable's $11.7 billion was such a claim once, made in a month that also felt like evidence.
Situational Awareness, the fund run by Leopold Aschenbrenner, saw total assets fall to roughly $10 billion from about $45 billion at the start of July after leveraged bets on public AI infrastructure names went against it, narrowly surviving margin calls and selling most of its public book to Citadel. In early August it invested $400 million in Source Foundry — a chipmaking startup reportedly valued at $5 billion, previously backed by Sequoia, founded by Stanford researchers Abdulmalik Obaid and Joe Burg to challenge ASML's lithography monopoly — bringing its total in the company to about $500 million. The public book repriced in weeks. The private one did not have to.
Jeff Dean (chief scientist, 27 years), Sanjay Ghemawat, Oriol Vinyals and Quoc Le are leaving Google to found Discovery Loop, an independent public benefit corporation aimed at automating machine-learning research and engineering before moving into hardware design, drug discovery and clean energy. Google is a founding investor and cloud partner. Radical Ventures and Khosla Ventures are co-leading a seed round with Lightspeed and Kleiner participating — but the round has not closed and no valuation has been disclosed, so any figure you see attached to it is not yet a fact.
Nielsen agreed to acquire DoubleVerify in an all-cash deal at an enterprise value of about $2.15 billion — $13.60 per share, a 30% premium to the 60-trading-day volume-weighted average price as of August 5. Nielsen measures who watched; DoubleVerify verifies that an ad was seen by a real person in a brand-safe environment. The deal is financed with cash and debt from Barclays, BofA and Citi, and is expected to close by the first quarter of 2027 subject to shareholder and regulatory approval. Because DoubleVerify was public, its price had been repriced every trading day for years — so the acquirer had to pay above it, not below.
Decade emerged from stealth with an $85 million seed round backed by Greenoaks, Benchmark and Diffusion, to build an AI-native wealth advisory business in Brazil pairing each client with a senior human adviser and a proprietary model that retains context across conversations. It was founded by Vitor Olivier, formerly Nubank's chief technology officer, and Hyperplane founder Felipe Meneses. The company describes it as the largest seed round ever raised by a Latin American startup — that is Decade's own characterization, not an independently verified record.
Volta launched from stealth with $300 million raised across its seed and Series A at a $2.4 billion post-money, co-led by Andreessen Horowitz and Altimeter with Nvidia, Michael Dell's family office, Azora and Matter participating. The same day, Bitdeer announced that its Tydal Data Center subsidiary signed a 16-year colocation lease with Volta's Tydal subsidiary for 121 IT megawatts in Norway, worth $4.7 billion in contracted revenue — up to roughly $8.0 billion over 24 years if the renewal option is exercised, with Dell as technology provider. A credit backstop of approximately $1.3 billion is anticipated via letters of credit from J.P. Morgan affiliates and another global institution. Volta says the end customer is a leading AI lab; it has not been officially confirmed.
The Financial Times reported that Google has assembled roughly $200 billion of interconnected contracts to deploy more than $150 billion of AI chips for Anthropic, uniting Google, Broadcom, Apollo, Blackstone, Morgan Stanley and several crypto-mining companies. The mechanism is the point: because Anthropic has no credit rating, the risk is distributed rather than underwritten — Google, itself an Anthropic investor, guarantees the data centres, while Broadcom commits to purchasing the chips it sells and helps finance them. This instrument class is not new. Vendor financing at semiconductor scale dates to 2023, and chipmakers have disclosed partner-lease guarantees in filings since late 2025. What is new is the size.
Inevitable AI Group, founded by Nimrod Lehavi and Ofer Bar-Or, disclosed a $6 million pre-seed led by Aleph — the Tel Aviv firm that backed monday.com and Melio early. The plan is explicit: take software companies worth between $1 billion and $10 billion, rebuild their products from scratch as AI-native, run each with a single founder and no staff, and undercut the original on price. Lehavi expects the cost of these products to fall by “90 to 95%”. For context, Forbes notes the early-2026 software selloff nicknamed the “SaaSpocalypse” wiped an estimated $285 billion of valuation in two days.
A survey of 396 enterprise organizations found one quarter delaying or cancelling AI initiatives because of unforeseen costs, with nearly half reporting that AI spending surprises had escalated to the board and one third imposing emergency spending freezes. Two important caveats travel with this number and should not be dropped: it comes from Mavvrik, a vendor that sells AI cost-governance software and therefore has an interest in the finding; and the fieldwork was conducted in April and May 2026, so the report is current but the data is roughly three months old.
Liquid AI released LFM2.5-2.6B, a 2.6-billion-parameter on-device agentic model pre-trained on roughly 34 trillion tokens, with a 128,000-token context window and native tool calling. The company reports 220 tokens per second on an Apple M5 Max, about 30 on a smartphone, and under 2.5GB of memory to run — beating models nearly four times its size on tool use and instruction following. Base and post-trained checkpoints were published with open weights on Hugging Face.
Sarvam AI closed a $75 million extension of its Series B led by Nvidia, valuing the Bengaluru sovereign-AI company at over $1.5 billion. The extension sits on top of an earlier $234 million Series B led by HCL Technologies. Sarvam says it is the first Indian company to bring Nvidia on as a strategic investor — a narrower claim than “Nvidia's first Indian investment,” which is not accurate given Nvidia's other commitments in the country. The capital goes toward foundation-model training and a trillion-parameter model built in India.
TikTok is laying off 250 employees and closing its Nashville office, leased in 2024 and home to part of its content-moderation organization; the site closes October 5 and affected staff receive severance. A spokesperson for the TikTok USDS Joint Venture said the move was to “streamline our operations and better align our teams for long-term growth.” Reporting places the cuts against a broader industry shift toward using AI to detect and remove violent or explicit content — though the company did not name automation as the cause.
Airtable's outcome was shaped in December 2021. Not by the product, which is fine, and not by the market, which is buying. By the number on a term sheet.
An $11.7 billion post-money is not a compliment. It is an obligation, and the obligation is arithmetic: at any multiple a buyer would pay, that mark needed billions in annual revenue behind it. The company built four hundred and eighty million, growing over twenty percent — a genuinely good business. The promise was larger than the category. Wanting the big number was never the mistake; not knowing what it obligated was.
So watch how it got unwound. The AI business moved into a separate company before the sale. The proven half went to a buyer at the market's real price, and the proceeds cleared a preference stack that returned the last round exactly what it put in. Read that generously and a founder recovered the only thing still uncapped. Read it plainly and the most valuable asset left just before a sale in which preferred got 1× and common — including employees who took years of below-market salary — got the remainder. Both come from the same filing.
The tempting conclusion is that the exit market broke. It did not. Last quarter set records on both sides — twenty-four companies acquired above a billion, thirty-two listings above it. Buyers are paying. The constraint was never the window. It was what you promised on the way in.
Which makes this answerable today. Run the division above against your own post-money. If the revenue it implies extends what you already do, your price is a target. If it describes a different company, your price is a ceiling — and the fix sits underneath the number, not years away at a sale. You often cannot choose your valuation — term sheets and your lead's ownership math set it. You have far more room on the structure, and structure decided this one: Airtable's preferred was 1× non-participating, the only reason the stack cleared. A 2× participating stack would have left common with nothing. Argue that line harder than the headline. Fourteen companies took a billion-dollar mark last month, and for some it was right — the honest cost of my argument. Less capital is less capital, and whoever raised more may outspend you. My wager: if the median public software multiple is above six by August 2027, those marks clear and I was wrong.
The best reason to hold terms you can defend is not caution. It is that a clearable mark leaves the interesting thing inside your company instead of forcing a carve-out later. Somewhere in your industry is work still done by hand because it was never worth automating — no round raised against it, no comparable, unpriced because nobody has built it. That is what you keep building toward. You do not get there by being valued correctly, but by staying free enough to still own it when it works.