Capital & Industrial Strategy

Intel announced it will join Elon Musk's Terafab project, a planned Texas semiconductor facility intended to manufacture chips for Tesla, SpaceX, and xAI, according to Bloomberg and WSJ. The scope of Intel's contribution remains undefined, but the partnership represents a significant strategic shift: Intel is positioning itself as a contract manufacturer for a customer consortium that could eventually reduce US dependence on TSMC for advanced AI and automotive chips. D.A. Davidson analyst Gil Luria noted Intel has positioned itself to secure the volumes necessary to be profitable, per Bloomberg. Intel's stock rose on the announcement, reflecting investor optimism that guaranteed volume from Musk's enterprises could stabilise the struggling foundry business.

Terafab aims to consolidate the entire chip-making lifecycle under one roof to power next-generation AI and robotics, per Bloomberg. This vertical integration strategy mirrors China's approach to semiconductor self-sufficiency but applies it at the corporate rather than national level. The project carries execution risk — Intel has struggled with process technology delays and cost overruns — but if successful, it could establish a blueprint for how large industrial customers secure chip supply outside the Taiwan-dependent incumbent supply chain.

Anthropic expanded its compute agreement with Google and Broadcom as the company's run-rate revenue surged to $30 billion, according to TechCrunch. The deal includes increased access to Google's TPUs and Broadcom's custom AI accelerators, reflecting the infrastructure demands of scaling enterprise deployments. Broadcom's stock jumped 6% on the announcement, per CNBC, indicating investor confidence that custom silicon deals with leading AI firms will drive sustained revenue growth for the chipmaker.

Simultaneously, Anthropic launched Claude Mythos, a specialised cybersecurity model designed to detect software vulnerabilities, but restricted access to Amazon, Microsoft, Apple, CrowdStrike, and Palo Alto Networks under Project Glasswing, per FT and WSJ. The rollout came days after a source code leak, and Anthropic limited distribution over fears hackers could weaponise the model for cyberattacks, according to CNBC. This dual-use dilemma — where capabilities valuable for defence are equally valuable for offense — illustrates why enterprise AI deployment increasingly requires closed ecosystems and trusted partner networks rather than open model distribution.

Zhipu raised pricing for its most advanced AI model by at least 8%, joining other leading Chinese AI companies in attempting to monetise years of research and computing investments, per Bloomberg. The move reflects a broader industry shift in China from subsidised growth to profit extraction, as venture funding for unprofitable AI startups has contracted and investors demand clearer paths to return. This follows a period of aggressive pricing competition where Chinese AI firms undercut US rivals to build market share, but that strategy appears to have reached its limit as compute costs remain high and customer willingness to pay has not materialised at scale.

The monetisation wave suggests Chinese AI firms are facing pressure to demonstrate sustainable unit economics, likely driven by both investor fatigue and government policy shifts away from indiscriminate subsidies. This could slow the pace of Chinese AI deployment if enterprises prove price-sensitive, or it could accelerate consolidation as only well-capitalised players can afford to continue subsidising adoption. The timing is notable: Chinese firms are raising prices just as US firms are seeing enterprise revenue scale, suggesting the global AI market is bifurcating into distinct competitive environments with different capital availability and customer dynamics.

Family offices are increasingly bypassing traditional venture capital intermediaries to invest directly in AI startups, seeking early-stage exposure to what they perceive as a once-in-a-generation technology shift, according to TechCrunch. Arena Private Wealth noted this trend is transforming passive capital allocators into active participants in AI financing, a structural shift driven by FOMO, dissatisfaction with VC fee structures, and the perception that AI deals are oversubscribed and difficult to access through conventional channels. This mirrors the late-stage private equity dynamic of the 2010s but applies it earlier in the startup lifecycle, compressing the time between founding and large-scale institutional backing.

This capital flow carries systemic risk: family offices typically lack the portfolio construction discipline and due diligence infrastructure of professional VCs, concentrating risk among investors who may be less equipped to absorb losses. However, it also increases available capital for AI startups, potentially accelerating innovation by removing funding bottlenecks. The trend suggests AI's winner-take-all narrative has penetrated private wealth management, creating conditions where capital availability could outpace viable investment opportunities — a classic bubble precursor if fundamentals do not justify valuations.

US AI companies are collaborating to share information aimed at preventing Chinese competitors from extracting capabilities through model distillation, according to Semafor. This represents a significant strategic shift: rivals who normally compete on model performance are acknowledging that intellectual property protection is a collective action problem that individual firms cannot solve through technical means alone. Distillation — training smaller, cheaper models to mimic the outputs of larger proprietary systems — has emerged as a primary vector for capability transfer, allowing well-resourced competitors to approximate frontier model performance at a fraction of the training cost.

The alliance's formation suggests US firms believe technical defences (API rate limiting, output watermarking, adversarial input detection) are insufficient, and that legal or regulatory remedies require coordinated industry action. This could presage lobbying for export controls on model weights, restrictions on commercial API access from certain jurisdictions, or even international agreements on AI intellectual property. The move also acknowledges that the open-source AI movement, while beneficial for research, creates asymmetric advantages for competitors willing to use open models as distillation targets without contributing back improvements.