The full issue. Week of August 11–17, 2026. 13 signals, each with a Founder Signal. The week's headline AI-infrastructure numbers were announcements of capacity, not commitments of capital — and the risk behind them moved outward until it reached insurers who say they are near their limit.
Signal of the Week. WSJ report: Nvidia now expected to guarantee under $120B, down from the $250B previously discussed, and only for the project's first phase — after investors raised concerns about its risk exposure.
Second source on the guarantee cut. The 10-gigawatt Ohio project is developed by SB Energy, a SoftBank subsidiary; OpenAI is still negotiating a binding lease for the full build.
Primary source, announced Aug 10. Treats GPU compute as a financeable asset the way a power plant or aircraft fleet is. Memoranda of understanding, explicitly subject to definitive agreements — announced capacity, not committed capital.
Mobilize and deploy $250B through lending, investments, capital markets and advisory over 18 months ending July 4, 2027. Unlike the week's other large numbers, this one carries a deadline — which makes it checkable.
Scope detail: three categories — digital infrastructure, energy and power, and core infrastructure including grid, water and critical minerals.
The constraint nobody announces. AIG's Eric Andersen: data centers need coverage from project finance through to operations, and the buildout is 'absolutely maxing out the P&C insurance industry in terms of the limits that are required.'
Second source on the insurance ceiling. Concentrating $10–20B of value on a single site pushes past what traditional insurers can comfortably underwrite alone.
Featured signal. Primary source. An executed agreement with a named counterparty and figure — but the NVIDIA HGX B300 capacity it buys does not arrive until Q1 2027. Even committed compute is a forward contract.
Featured signal, and the week's clearest counter-example: when a buyer genuinely wants an asset, money commits in weeks. Roughly 5.4x the $1.3B valuation OpenRouter carried at its Series B in May 2026.
Second source. Bloomberg notes the final price could still change. OpenRouter routes traffic between AI models — the position between a developer and whichever model they use this month.
Featured signal. Led by Coatue with Blackstone, MGX, T. Rowe Price and Sixth Street Growth. 42% above the $134B mark set in February, on a revenue run-rate past $7B growing over 80% year over year.
The detail worth keeping. Here the constraint was not the availability of capital but how much a company was willing to absorb — the opposite problem to the one most founders have.
Igor Babuschkin's full-stack AI company, ~$5B valuation two months after founding, with strategic investment from NVIDIA and AMD Ventures. The chip vendors are again investing in the companies that buy their chips.
Second source on River AI, and useful context on where the week's largest checks landed.
An announced valuation meeting a reported income statement. Would surpass SpaceX's $1.77T June listing; one investor called $2T low and suggested $3T.
The arithmetic underneath the target: $559M adjusted operating profit on $10.9B of Q2 revenue, a margin of about 5.1%, and not yet profitable annually. At Nasdaq-100 multiples a $2T company needs roughly $59–79B in annual profit.
Small next to the week's headline numbers and considerably more informative: contracted revenue with a counterparty whose credit quality is stated.
$582.3M in Q2 revenue. Set against announced programs measured in hundreds of billions, this is the scale at which the industry is actually collecting cash rather than pledging it.
Worth distinguishing from everything else in the issue: an LP commitment to a venture fund is a legally binding obligation to fund capital calls. Not a memorandum, not a target.
The India leg of the same raise. New funds have deployment clocks, and the first year of a fund is the most permissive period a founder will ever meet.
Disclosure is the story. Most of the week's very large numbers cannot be converted into a multiple because no denominator was published. This one can — roughly 7.7x — so a founder can judge it.
The other end of the same market that produced trillion-dollar valuation targets this week.
Returning capital rather than spending it to zero is an honest ending, and it preserves the one asset a founder keeps across companies: whether investors would back them again.
Explore more in Finance →