The Brief Brief
the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·18 AUG 2026 ·TUESDAY ·2 MIN READ ·5 STORIES

Tuesday: leases, run rates and one bad YAML

An 8-gigawatt lease lands partly on a former uranium enrichment site, Anthropic adds $18B of run rate in two months, and an autonomous agent walks out of Snowflake's Jira.

01 / The Day

TUESDAY 18 AUG 2026, ranked

05

OpenAI leases 8GW in Ohio; Nvidia backstops $105B

OpenAI signed a 20-year lease with SoftBank subsidiary SB Energy for a campus with around 8 gigawatts of IT capacity at PORTS-Pike in Ohio, a site that sits partly on a former Department of Energy uranium enrichment facility, with Nvidia guaranteeing the residual value of the first phase.

  • 20-year SB Energy lease for ~8GW of IT capacity; the first 800MW is slated to come online in 2028
  • Nvidia guarantees up to $105B of residual value on phase one's 4.25GW, plus $1.5B into SB Energy
  • Nine major tech firms now carry roughly $3T of mostly AI-related off-balance-sheet obligations
Why it mattersResidual-value guarantees are the mechanism by which compute risk moves off the tenant's books and onto the chip supplier's.
01 the-decoder.com↗

Anthropic's run rate hits $65B, up $18B in two months

Anthropic closed July at a $65bn annualised revenue run rate, up from $47bn in May and $9bn at the end of 2025, Bloomberg reported on Monday.

  • $9B at end-2025, $47B in May, $65B at end-July; investors expect $100-120B by the close of 2026
  • Last valued at $965B in late May on a $65B round; expected to reach public markets possibly as soon as this fall, seeking $2T or more
  • OpenAI's run rate reached $40B, double its $20B at the end of 2025
Why it mattersGrowth at that slope is the entire argument for the two-trillion ask, and the reason nobody is asking how run rate is defined.
02 techcrunch.com↗

Stripe to buy OpenRouter for more than $7B

Stripe is reportedly acquiring model-routing gateway OpenRouter for more than $7bn, roughly five times the $1.3bn valuation it took three months ago.

  • $1.3B valuation and a $113M Series B in May; a reported $7B-plus exit in August
  • OpenRouter fronts more than 400 models for 8 million users, pitched by its CEO as 'Stripe for AI' to prevent lock-in with individual providers
  • The read: Stripe already handles high volumes of latency-sensitive, high-availability requests, which is what token routing is
Why it mattersThe neutral layer between applications and labs turns out to be a payments problem, and it is now being priced like one.
03 the-decoder.com↗

Wiz's Red Agent turned one YAML into Snowflake's Jira

Wiz's autonomous Red Agent found and exploited a GitHub Actions script-injection flaw in Snowflake's public .NET connector repo, exfiltrating a Jira API token. Copilot Autofix had contributed a separate fix to a different workflow file in the same pull request.

  • An issue title containing a single quote broke out of an echo in jira_issue.yml; the agent read its own syntax error and rewrote the payload
  • The token authenticated as qa@snowflake.net, reading engineering, security-compliance and bug-bounty projects
  • Live 18 June when PR #1218 merged, found and patched 23 June, credential rotated 24 June
Why it mattersThe agent did the interesting part: it debugged its own failed injection and retried. Wiz says it is unclear whether the vulnerable change was AI-assisted, which is the honest version of a much louder claim.
04 wiz.io↗

Groq raises $350M at half its old valuation

Groq took $350m led by investment firm Disruptive at a $3.5bn valuation, down from $6.9bn, and recast itself from custom-silicon designer to an Nvidia-based inference cloud.

  • $3.5B now against $6.9B in September 2025; Nvidia's participation in the round is planned, not closed
  • Nvidia hired founder and CEO Jonathan Ross and other top talent as part of a $20B licensing deal paid out to investors
  • 13 data centres across North America, Europe, the Middle East and Asia Pacific, scaling from 54MW to more than 200MW in 2027
Why it mattersThe company that spent years arguing GPUs were the wrong shape for inference is now renting them out.
05 techcrunch.com↗
Be subscriber #013 the weekly ten, every friday by email · no spam · unsubscribe anytime
Past days

The Daily Archive

/TUE 18 AUGTuesday: leases, run rates and one bad YAML/MON 17 AUGGigawatts and a shredder/SAT 15 AUGThe price of the future/FRI 14 AUGFlash, Flash, Flash/THU 13 AUGMath and model races/WED 12 AUGThe COO exits, a model does maths, and Nvidia bankrolls the buildout/TUE 11 AUGAnthropic Day/MON 10 AUGAgents went rogue, China swept video, and 500 towns said no/SUN 09 AUGSunday shakeup: Hassabis pivots to AGI, Moonshot climbs the benchmark/FRI 07 AUGByteDance goes to 10T; Dean goes independent/THU 06 AUGCourts, coders, and coordinated agents/WED 05 AUGPlatform wars, agent incidents, and $10B in compute/MON 03 AUGMonday morning models/SUN 02 AUGThe agent problem compounds/SAT 01 AUGThe Rules Are Coming/FRI 31 JULOpenAI cuts Luna 80%; Anthropic discloses model breaches/THU 30 JULThe Accountability Edition/WED 29 JULWednesday, 29 July — Accountability morning/TUE 28 JULChips fall, labs take positions, agents get a cost formula/MON 27 JULKimi-K3 Ships, Gemini Goes Three-Way, and HuggingFace Speaks for Itself/SUN 26 JULThe Labs' Actual Positions