🧭 Goldman Sachs Restricts Hong Kong Staff from Claude — A Compliance Lesson for Every Enterprise AI Deployment

Goldman Sachs staff in Hong Kong have lost access to Anthropic's Claude, Bloomberg reported on April 29. The restriction stems from Goldman conducting a strict reading of its enterprise contract with Anthropic and concluding — after consulting directly with the company — that Hong Kong falls outside the territories covered. The change affects software engineers most directly, as Claude has been used primarily for coding tasks on Goldman's internal AI platform. Staff travelling to Hong Kong from other offices are also blocked while in-city; models such as ChatGPT and Gemini remain available in the same environment.

Why Hong Kong specifically?

Anthropic's public disclosure of supported markets does not currently list Hong Kong as a region where either the Claude API or Claude.ai are officially accessible. This is not a sudden policy change — the geographic restriction has existed in Anthropic's terms — but Goldman is the first major financial institution to enforce it publicly. The move arrives against a backdrop of:

• US–China tech tensions — heightened regulatory scrutiny of US AI products operating in or adjacent to mainland Chinese jurisdiction.
• HKMA guidance — the Hong Kong Monetary Authority told Reuters that some banks are assessing additional mitigation measures in response to cybersecurity risks from advanced AI models, particularly following the wider deployment of models with strong agentic capabilities.
• Contract specificity — enterprise AI agreements increasingly enumerate covered territories. Goldman's situation is a canonical example of geographic scope clauses that many enterprise buyers have glossed over.

What this means for enterprise AI teams

Goldman's situation is a preview of a compliance review that every global enterprise should run now, before a Bloomberg article forces the issue. The practical checklist:

The Goldman case also highlights how quickly geographic scope can become a live compliance risk as AI tooling spreads from engineering teams to finance, legal, and operations staff with no central governance layer. A developer tool deployed to one team can become an enterprise-wide exposure when it propagates through a global organisation.

🧭 Anthropic Crosses $30 Billion Revenue Run Rate — and Microsoft's Q3 Earnings Tell the Next Chapter

Anthropic's annualised revenue run rate has surpassed $30 billion, up sharply from approximately $9 billion at the end of 2025, according to data cited by HSBC analyst Stephen Bersey in a note published ahead of Microsoft's fiscal Q3 earnings on April 29. That revenue figure — confirmed by Anthropic via its official X account and covered by Bloomberg — would make Anthropic the fastest-growing AI company in history by the same metric, and represents the first time Anthropic has overtaken OpenAI's $25 billion ARR. The composition is roughly 80% enterprise, with more than 1,000 clients now paying over $1 million annually.

The Microsoft angle: $5B in, $30B of compute rented out

Microsoft invested $5 billion in Anthropic in November 2025 as part of a deal that also sees Anthropic commit to renting $30 billion of Azure compute over the term of the agreement. HSBC raised its Microsoft price target to $593 on the strength of that arrangement, noting that while Azure's outlook still hinges on its broader cloud business, the Anthropic relationship represents a meaningful hedge against Microsoft's high dependence on OpenAI — which accounts for roughly 45% of Microsoft's remaining performance obligations.

The revenue accounting nuance worth knowing

OpenAI's chief revenue officer contested the $30B figure, arguing it is overstated by ~$8 billion. The accounting question is whether Anthropic should recognise revenue from AWS and Google Cloud at gross value (full amount billed through the partner) or net (after the partner's distribution margin). Anthropic argues it is the principal in these transactions — it sets prices and the cloud providers are distribution channels — so gross recognition is correct under GAAP. This debate matters to developers because:

• If gross accounting stands, Anthropic's revenue base is as reported and the compute investment thesis holds.
• If accounting standards shift toward net, reported revenue would drop ~27% — which could affect investment terms and pricing stability over multi-year enterprise contracts.
• Enterprise procurement teams signing multi-year Claude API contracts should ask for a revenue recognition disclosure or at least understand which accounting treatment underpins vendor stability claims.