Capital & Industrial Strategy

US pharmaceutical giant Eli Lilly signed a licensing agreement with Hong Kong-listed Insilico Medicine valued at up to $2.75 billion, including $115 million upfront, to bring AI-discovered drug candidates to global markets. The deal represents the largest disclosed pharma-AI transaction to date and marks a strategic shift from earlier AI partnerships focused on research collaboration to licensing arrangements for late-stage candidates. According to Financial Times, global pharmaceutical companies are aggressively searching for new medicines in China, where AI drug discovery platforms have advanced rapidly. Semafor reports the deal follows a pattern of Western pharma de-risking internal R&D by licensing AI-generated candidates that have already cleared early validation hurdles.

The transaction structure—relatively modest upfront payment with milestone-based payouts—reflects pharma's continued uncertainty about AI-discovered drugs' clinical success rates. Insilico's platform generates molecular structures computationally, bypassing traditional medicinal chemistry workflows, but no AI-discovered drug has yet completed Phase III trials. Eli Lilly's willingness to commit nearly $3 billion in total value suggests confidence that at least some candidates will reach commercialisation, making this a test case for whether AI drug discovery can deliver on its long-promised productivity gains.

Anthropic is seeking a court injunction against the Pentagon's decision to designate it a supply-chain risk and ban its models from defence applications, according to Wall Street Journal. The dispute escalated after reporting confirmed Anthropic's Claude model was used during the opening phase of US-Iran hostilities, despite Anthropic's stated objections to military applications involving autonomous weapons or domestic surveillance. Bloomberg's Odd Lots podcast examines how these models are deployed in military contexts and what fully autonomous weapons usage actually entails. The Pentagon's supply-chain risk designation effectively cuts Anthropic out of lucrative defence contracts while its models remain accessible via third-party resellers or cloud infrastructure providers—a regulatory gap Anthropic is likely challenging in court.

Semafor's interview with Drew Cukor, the Pentagon's AI architect, underscores the Department of Defence's view that commercial foundation models are critical infrastructure that cannot be subject to developer-imposed use restrictions. This position directly contradicts Anthropic's attempt to maintain ethical boundaries around military applications. The legal battle will determine whether AI providers can enforce acceptable use policies against government customers or whether national security imperatives override contractual restrictions.

OpenAI shut down Sora, its video generation tool, just six months after public release, marking its first major product retreat since ChatGPT's launch. According to Wall Street Journal, Sam Altman positioned Sora as a vehicle to establish OpenAI as a creative industry pioneer, but the product failed to gain traction beyond novelty use cases. TechCrunch reports the tool's face-upload feature raised immediate data privacy concerns, though OpenAI has not disclosed whether regulatory pressure contributed to the shutdown decision. The timing suggests OpenAI concluded Sora's compute costs—video generation is orders of magnitude more expensive than text—could not be justified by revenue or strategic positioning gains.

The failure highlights structural challenges in monetising generative AI beyond enterprise API contracts and chatbot subscriptions. Consumer-facing creative tools require sustained engagement and willingness to pay for outputs, but early data suggests users treat them as one-time experiments rather than habitual purchases. OpenAI's decision to kill Sora rather than iterate suggests the unit economics were fundamentally broken, not merely requiring product refinement. This contrasts sharply with enterprise adoption, where AI tools are embedding into workflows with measurable productivity gains.

A pro-AI advocacy organisation plans to spend $100 million on US midterm elections, according to Financial Times, positioning the November 8 poll as a battleground over AI regulation. The spending level—comparable to major industry lobbying efforts in healthcare or finance—indicates the AI sector's assessment that legislative outcomes will materially affect market dynamics, either by constraining deployment or by pre-empting stricter state-level rules. The group's formation follows rising public backlash against AI-driven job displacement, as BBC reports that tech CEOs are increasingly attributing layoffs to AI adoption, creating political pressure for worker protections or deployment restrictions.

The electoral strategy likely aims to secure Congressional support for federal pre-emption of state AI regulations—which would favour large-scale deployers—and to block proposals for mandatory AI impact assessments or algorithmic transparency requirements. Venture investors view regulatory fragmentation as a significant drag on exits, as compliance costs vary by jurisdiction and create uncertainty for acquirers. A unified federal framework, even if somewhat restrictive, is generally preferred to a patchwork of state-level rules.