The Daily Brief · Issue 3 · 18 August 2026Anthropic tells investors its revenue run rate hit $65 billion, and Nvidia backs OpenAI's Ohio data centre with up to $105 billion
Money dominated Monday: Anthropic's growth numbers leaked ahead of its IPO, Nvidia put its balance sheet behind OpenAI's biggest build-out, and Groq raised $350 million. Cursor launched a GitHub rival on the same day GitHub went down for hours.
BusinessAnthropic's annualised revenue reaches $65 billion
Anthropic told investors its annualised revenue run rate (monthly revenue multiplied by 12) reached $65 billion in July, CNBC reports, confirming a Bloomberg scoop. That is up from $47 billion in May and roughly seven times its pace at the end of 2025. A source said preliminary second-quarter revenue was $11.5 billion, about 14 times the same quarter a year earlier. For comparison, CNBC says OpenAI's run rate recently hit $40 billion. Anthropic filed confidentially for an IPO in June and is valued at $965 billion. Anthropic declined to comment.
Why it matters: Anthropic is now earning more than OpenAI on this measure, mostly from business customers, which strengthens its hand before a record IPO.
Source: CNBC
InfrastructureNvidia commits up to $105 billion to back OpenAI's Ohio data centre
Nvidia said on Monday it will provide up to $105 billion of financial support for a huge OpenAI data-centre campus in Pike County, Ohio. The package includes a $1.5 billion investment in SB Energy, the SoftBank affiliate building the site on a 20-year lease to OpenAI, plus backing for about 4 gigawatts of development opening between 2028 and 2030. It acts as a safety net if OpenAI's finances weaken. The deal follows Nvidia's plan with Goldman Sachs, Blackstone and others to raise $500 billion to finance GPUs. Critics call this 'circular' spending; Nvidia's backers call it a moat.
Why it matters: Nvidia is increasingly financing its own customers, which keeps chip demand high but ties its fortunes more closely to OpenAI's.
Source: CNBC
Developer toolsCursor launches Origin, a code-hosting service, as GitHub suffers a long outage
Cursor began rolling out Origin, its own place to store code and review pull requests, to paid users on Monday. About three and a half hours later GitHub had a global outage lasting 6 hours 42 minutes, VentureBeat reports, with error rates near 20% on pull requests and the API. Origin does not ask teams to leave GitHub: it mirrors connected repositories, keeps GitHub as the main copy and syncs comments both ways. Vercel, Depot and Buildkite integrations shipped on day one. Cursor has not published Origin's pricing or data-handling terms, and it is switched on unless enterprise admins opt out.
Why it matters: Company admins should check whether their code is being mirrored to a new host by default, now owned by SpaceX.
Source: VentureBeat
FundingGroq raises $350 million at a $3.5 billion valuation
Groq, the inference-chip company whose technology Nvidia licensed in December 2025, raised $350 million at a $3.5 billion valuation. That is about half the $6.9 billion it was valued at in September 2025, before Nvidia hired its founder and senior team. Groq argues the old and new companies are not comparable. It now runs as an inference cloud (a service that runs other people's models), mixing its own chips with Nvidia systems. Groq says it has 13 data centres, over six million developers and plans to grow from 54 megawatts to more than 200 megawatts in 2027. Those figures are self-reported.
Why it matters: The drop in value shows what is left of a chip challenger after Nvidia takes its technology and people.
Source: Implicator.ai
ChipsGoogle reportedly works with AMD on a future TPU with built-in CPU cores
Google may be working with AMD on part of its tenth-generation TPU, its in-house AI chip, according to a SemiAnalysis client note reported by Tom's Hardware. The idea is a chip package that puts general-purpose CPU cores next to the AI cores. That would help with reinforcement learning, the trial-and-error training used for reasoning and agent models, which needs more ordinary computing than classic training. Google's latest TPU 8i systems already pair one Axion CPU with every two TPUs. Tom's Hardware stresses the report is unconfirmed and its own analysis is speculative.
Why it matters: Agent training is changing what AI chips need, which could open doors for AMD and other suppliers beyond Nvidia.
Source: Tom's Hardware
ResearchLinear data: teams with coding agents now open three times more pull requests
Project-tracking company Linear published data from tens of thousands of software teams. Between January and June 2026 the share of users active on AI features more than doubled in every job type; product managers rose from 12% to 34%. AI now writes just under half of all new issues (task tickets) in Linear, up from fewer than one in a thousand two years ago. Teams that connected a coding agent went from 21 to 65 pull requests a week over two years, while teams without one went from 8 to 10. Linear notes it cannot see AI use outside its product.
Why it matters: It is one of the largest real-world datasets showing coding agents change output, though it measures volume, not the value of the work.
Source: Linear