Broadcom signs long-term deal to develop Google's custom AI chips as Anthropic's Claude revenue surpasses $30 billion run-rate, reshaping the commercial AI infrastructure layer that lawyers must now navigate
Broadcom has signed a long-term agreement to develop custom AI chips for Google, deepening the vertical integration of AI hardware and software that underpins the commercial AI infrastructure now central to legal practice and deal-making. Separately, Anthropic — maker of the Claude AI model — has confirmed its annualised revenue run-rate has surpassed $30 billion, up from approximately $9 billion at the end of 2025, a pace of growth that places it among the fastest-scaling enterprise software companies in history.
Anthropic has also announced a commitment to invest $50 billion in strengthening US computing infrastructure, with its primary cloud provider remaining Amazon Web Services (AWS). Claude is now trained and deployed across a range of hardware including AWS Trainium, Google TPUs, and Nvidia GPUs — a multi-vendor approach that introduces complex technology licensing, data processing, and cloud services agreement structures.
For law firms, the rapid commercial maturity of models like Claude — which are being integrated into legal research, contract review, and document drafting tools — creates an urgent need to understand the underlying contractual and liability architecture. frameworks for law firms must now address not just internal use policies but the and governing the AI vendors whose tools are embedded in client-facing workflows. The (which began applying to general-purpose AI models in August 2025) and the both impose obligations on deployers of high-capability models — obligations that law firms using Claude or comparable tools are now directly subject to.
AI governance
technology licence agreements
data processing agreements (DPAs)
EU AI Act
UK's emerging AI regulatory framework
Why this matters
Anthropic's $30 billion revenue run-rate signals that foundation AI models — large-scale AI systems trained on broad datasets and deployable across multiple tasks — have crossed into commercial maturity, meaning law firms can no longer treat AI governance as a future-planning exercise. The EU AI Act's classification of general-purpose AI models with systemic risk (applicable to models trained on more than $10^{25}$ floating-point operations) imposes transparency, incident reporting, and adversarial testing obligations on providers like Anthropic — obligations that flow downstream to deployers, including law firms integrating Claude into client work. The Broadcom-Google chip deal illustrates that AI capability is increasingly concentrated in vertically integrated hardware-software stacks, raising competition law concerns about market dominance that the CMA and European Commission are already examining. For City firms, the practical implication is that every AI vendor due diligence process now requires analysis of the vendor's own regulatory compliance position under the EU AI Act and equivalent regimes.
On the Ground
A trainee on an AI governance matter would mark up data processing agreements to ensure they reflect the specific obligations applicable to high-capability AI models under the EU AI Act, and would draft AI governance policy documentation setting out a firm's internal rules on model use, output verification, and client disclosure. Vendor due diligence questionnaires sent to AI tool providers would also require trainee input, particularly around data residency, model training data sources, and incident response protocols.
Interview prep
Soundbite
At $30 billion run-rate, Claude is no longer an experiment — law firms' AI vendor contracts are now live regulatory exposure.
Question you might get
“Under the EU AI Act, what obligations does a law firm take on when it integrates a general-purpose AI model like Claude into its legal research or document drafting workflows?”
Full answer
Broadcom has signed a long-term deal to build Google's custom AI chips, while Anthropic's Claude has surpassed a $30 billion annualised revenue run-rate — together marking the commercial maturation of the AI infrastructure layer that law firms are now actively deploying. This matters for legal practice because the EU AI Act, which began applying to general-purpose AI models in 2025, imposes transparency and incident reporting obligations that flow to law firm deployers, not just AI vendors. The wider trend is vertical integration of AI hardware and software creating concentrated market structures that the CMA and EC are scrutinising under competition law. My view is that law firms will face their first meaningful regulatory enforcement action related to AI model use within 18 months, making AI governance policy work an urgent priority rather than a future-proofing exercise.