Compute & Infrastructure

The Department of Defense formally notified Anthropic on Thursday that it has designated the AI firm a supply-chain risk, a label historically reserved for foreign adversaries like Huawei, according to Bloomberg and The Verge. The designation stems from collapsed negotiations over a $200 million contract, with the Pentagon demanding unrestricted access to Anthropic's Claude AI for military applications including surveillance and weapons systems—terms Anthropic refused due to its acceptable use policies. Anthropic CEO Dario Amodei stated the company will challenge the designation in court, calling it unprecedented for a US firm, though he maintains it will not affect the 'vast majority' of customers, per The Financial Times.

Despite the formal designation, multiple sources report that negotiations between Anthropic and the Pentagon have quietly resumed, according to The Verge and TechCrunch. The dispute has exposed tensions between AI safety commitments and national security imperatives, with Bloomberg noting the clash draws fresh attention to the US government's use of commercially available data and AI for mass surveillance. Meanwhile, OpenAI announced it has no restrictions on military use and released new tools targeting financial services, positioning itself to capture Anthropic's defence and enterprise customers during the standoff, per Bloomberg.

The Trump administration is considering requiring permits for every Nvidia and AMD AI chip sale globally, regardless of destination country, according to a draft proposal reported by Bloomberg and The Financial Times. The proposed framework would tie chip export approval to foreign investment pledges in US infrastructure, effectively making Washington a gatekeeper for global AI compute capacity distribution. Under the draft rules, countries seeking advanced semiconductors would need to invest in American data centres, chip manufacturing, or related infrastructure to gain access, fundamentally restructuring the global semiconductor market from a commercial system to a state-mediated allocation mechanism.

The proposal represents a dramatic expansion of US export control authority, moving beyond restricting adversaries to actively managing the entire global supply chain. TechCrunch notes the framework would give the US government a role in every chip export sale regardless of origin country, raising questions about enforceability and allied cooperation. No implementation timeline has been announced, and the draft status means significant changes remain possible before any final rules emerge.

Oracle announced it is eliminating thousands of positions as the company confronts cash flow constraints from its massive AI data centre expansion, according to Bloomberg. The layoffs underscore the capital intensity of AI infrastructure buildout, even for established cloud providers with existing data centre footprints. Oracle has been racing to expand compute capacity to compete with hyperscalers, but the cuts suggest the company is struggling to finance construction while maintaining operating margins. The specific scope of reductions and affected business units was not disclosed.

In contrast, Broadcom CEO Hock Tan projected AI chip sales will exceed $100 billion in 2027, positioning the company as an increasingly credible Nvidia competitor, per Bloomberg. Broadcom's custom AI accelerator business serves hyperscalers building proprietary silicon to reduce dependence on Nvidia's GPUs. The forecast suggests Broadcom is capturing meaningful share in AI training and inference workloads, though Nvidia still dominates the broader market. Semiconductor equipment maker BE Semiconductor dropped 17% after reports that memory chipmakers are evaluating new standards that could delay demand for its packaging technology, according to Bloomberg, highlighting volatility in the AI supply chain even during a broader buildout cycle.

Iranian forces struck Amazon Web Services data centres in the Gulf region during escalating Middle East conflict, marking the first known military attack on a US hyperscaler's physical infrastructure, according to The Financial Times. The attack has rattled regional ambitions to build multibillion-dollar cloud facilities, as Gulf states have been positioning themselves as AI and data centre hubs with significant investments in compute capacity. The strike demonstrates that distributed cloud infrastructure—often promoted for resilience—becomes a vulnerability during armed conflict when data centres are located in or near conflict zones.

The incident compounds existing concerns about energy grid capacity and cooling challenges in Gulf data centres, which already face extreme climate constraints. Regional governments have been courting hyperscalers with tax incentives and cheap energy, but the security risks may now force a reassessment of data sovereignty strategies and physical infrastructure placement. The specific extent of damage and service disruption has not been publicly disclosed by Amazon.

SoftBank Group is seeking up to $40 billion in debt financing to fund its investment in OpenAI, which would constitute the company's largest-ever dollar-denominated borrowing, according to Bloomberg. The loan underscores the extreme capital requirements of AI infrastructure, as even SoftBank—which has significant existing holdings—requires massive external financing to participate at scale. The borrowing comes as private credit markets face growing concerns about valuations and volatility, with Bloomberg reporting that a Blue Owl Capital fund halted quarterly redemptions and Blackstone's flagship private credit fund saw record redemption requests of 7.9%.

The financing structure and ultimate deployment timeline remain unclear, but the scale signals that compute infrastructure and AI model development costs continue escalating beyond earlier projections. SoftBank's willingness to take on unprecedented debt also reflects conviction that OpenAI's market position justifies the leverage, though the company's track record with large tech bets has been mixed.