Capital & Industrial Strategy
Top Line
OpenAI missed internal revenue and user growth targets ahead of its IPO, triggering a broad selloff in AI-linked equities including Oracle and CoreWeave, while the company publicly pushed back claiming its consumer and enterprise businesses are 'firing on all cylinders' — the disconnect between private targets and public positioning is the central tension investors must resolve before big tech earnings this week.
Google signed a confirmed contract with the Pentagon to provide AI for classified military work, stepping directly into the space Anthropic vacated by refusing DoD use cases involving domestic mass surveillance and autonomous weapons — a strategically consequential split in how frontier labs are positioning themselves relative to national security.
OpenAI ended Microsoft's exclusivity and immediately expanded to AWS, with Amazon Bedrock now offering OpenAI's latest models including Codex — a structural shift in the frontier model distribution market that erodes Microsoft's competitive moat and gives OpenAI direct leverage over cloud pricing.
Beijing ordered Meta to unwind its $2 billion acquisition of Chinese AI startup Manus, effectively killing the deal and signalling that China will block outbound transfers of AI talent, data, and IP regardless of deal structure — raising the risk premium on all cross-border China AI transactions.
US power equipment spending for AI data centers is projected to surge from $2.6 billion last year to $65 billion by 2030 according to Wood Mackenzie, while a major Oaktree-owned data center operator has paused Middle East investment decisions amid the Iran conflict — illustrating both the scale of infrastructure buildout and its geopolitical fragility.
Key Developments
OpenAI Revenue Miss vs. Bullish Narrative: Market Credibility at Stake
The Wall Street Journal reported that OpenAI fell short of internal sales and user growth targets, sending AI-infrastructure stocks sharply lower — Oracle and CoreWeave led the selloff, given their deep revenue exposure to OpenAI compute commitments. The market reaction reflects a structural concern that has been building for months: whether the hundreds of billions being committed to AI infrastructure by hyperscalers is anchored to durable end-demand or to optimistic projections that are already slipping. OpenAI responded publicly that its business is 'firing on all cylinders' and cited growth in enterprise and a nascent advertising business, but the gap between that messaging and the reported internal targets has not been resolved. Bloomberg, CNBC
The timing is critical: OpenAI is racing toward a structural conversion from nonprofit to for-profit entity, with an IPO on the horizon, and its commercial trajectory directly determines the justification for the $40 billion funding round it closed earlier this year. AI investors surveyed by Axios remained broadly bullish, framing the miss as a timing issue rather than a demand problem, but the selloff indicates that public market sentiment is more conditional. The upcoming earnings from Microsoft, Meta, Alphabet, and Amazon will be the next credibility test — capex guidance from those four companies effectively sets the floor for how much AI infrastructure spending is truly committed versus discretionary. Axios, Reuters
OpenAI Ends Microsoft Exclusivity, Expands to AWS — Distribution Power Shifts
One day after restructuring its relationship with Microsoft to remove exclusive cloud rights, OpenAI announced that its latest models — including Codex — are now available on Amazon Bedrock. AWS simultaneously launched a new OpenAI-powered agent service. The sequencing is notable: Microsoft agreed to loosen exclusivity, and within 24 hours Amazon was offering OpenAI products competitively. This was not an opportunistic move by Amazon — the deal required negotiation and was clearly in progress before the Microsoft agreement was finalised, suggesting OpenAI has been systematically building its multi-cloud distribution strategy. TechCrunch, Financial Times, CNBC
For Microsoft, losing exclusivity weakens the primary competitive differentiator that justified its multi-billion dollar investment in OpenAI — Azure was the only place enterprise customers could access frontier OpenAI models at scale. For OpenAI, multi-cloud distribution is an IPO-readiness move: it diversifies revenue, reduces single-counterparty dependency, and signals to public market investors that the business is not structurally subordinated to Microsoft's commercial interests. WSJ analysis suggests this restructuring may ultimately benefit both parties by reducing tensions that had built around the exclusivity arrangement, with Microsoft retaining preferred pricing and compute rights even as distribution widens. WSJ
Google-Pentagon Deal: National Security AI Market Crystallises Around Compliant Labs
Google has signed a confirmed agreement with the US Department of Defense to provide AI — including Gemini — for classified military applications. The deal was confirmed by the Pentagon's AI chief and reported by both Bloomberg and TechCrunch, and came directly after Anthropic declined to permit DoD use of its models for domestic mass surveillance and autonomous weapons systems. Google's internal researchers protested the deal, echoing the employee activism that led to Project Maven's cancellation in 2018 — but this time Google's leadership proceeded. Bloomberg, TechCrunch, CNBC
The Pentagon AI chief's public statement that relying on 'one model is never a good thing' signals that the DoD is actively cultivating a multi-vendor AI supply chain — a procurement posture that benefits labs willing to accept national security use cases and disadvantages those, like Anthropic, that maintain hard constraints. This creates a structural bifurcation in the frontier AI market: labs that accept defense and intelligence work will compete for a growing federal AI budget, while safety-focused labs that decline will be limited to commercial enterprise markets. The national security AI procurement market is large enough — and growing fast enough — that this is now a material strategic decision for any frontier lab seeking scale.
Beijing Blocks Meta-Manus Deal: Cross-Border China AI M&A Now High-Risk
Chinese authorities ordered Meta to unwind its confirmed $2 billion acquisition of Manus, the AI agent startup that attracted significant attention earlier this year as a potential challenger to Silicon Valley models. The intervention was not a regulatory delay but a direct order to reverse a completed transaction — an extraordinary exercise of state power that goes beyond standard merger review. Beijing's stated concern centres on the outbound transfer of AI talent, data, and intellectual property to a US entity, and the action was framed explicitly as a warning to other Chinese AI startups considering similar arrangements. Bloomberg, CNBC, Reuters
The strategic implication for investors is significant. Chinese AI startups have historically been attractive acquisition targets for US tech companies seeking talent, proprietary datasets, and technology developed outside Silicon Valley's cost structures. Beijing's intervention effectively walls off that pipeline and raises the cost of any deal with a Chinese AI company — even those headquartered internationally — by introducing unwinding risk after capital is deployed. For Meta specifically, the failure is particularly costly given the $2 billion price and the reputational damage of a forced divestiture. For the broader M&A market, expect deal risk premiums on any cross-border China tech transaction to reprice materially.
AI Infrastructure Capex: Supply Chain Winners Emerging, But Geopolitical Fragility Increasing
Multiple data points this week confirm the AI infrastructure investment wave is propagating deep into supply chains. Victory Giant Technology reported 28% year-on-year revenue growth in Q1 on printed circuit board demand for AI servers, while Asian chip supply chain companies are posting gains of up to 770% according to Bloomberg's supply chain analysis. Seagate issued an upbeat quarterly forecast citing AI-driven data storage demand, lifting storage stocks broadly. SK Hynix and Samsung are pushing customers toward long-term supply contracts for high-bandwidth memory as acute shortages persist — a structural shift that, if it holds, would significantly reduce the boom-bust volatility that has historically characterised memory chip markets. Bloomberg, Bloomberg, Financial Times, Reuters
Against that backdrop, Pure DC — an Oaktree Capital-owned data center operator — has confirmed it is pausing investment decisions in the Middle East due to regional uncertainty from the Iran conflict. This is a confirmed operational pause, not speculation, and it illustrates the geopolitical fragility of the Gulf data center buildout that had been positioned as a major growth market for AI infrastructure. Wood Mackenzie's projection of US power equipment spending reaching $65 billion by 2030 — a 25-fold increase from 2025 — underscores that the domestic US infrastructure investment remains on track, but international deployment is increasingly subject to conflict risk and political disruption. Bloomberg, CNBC
Signals & Trends
Frontier Model Distribution Is Becoming a Multi-Rail Utility — Pricing Power Shifts to Model Providers
The OpenAI-AWS deal, coming one day after Microsoft's exclusivity was dissolved, signals that frontier AI models are transitioning from exclusive cloud differentiators to multi-platform utilities — available wherever enterprise customers prefer to operate. This mirrors the historical trajectory of databases and middleware: initially exclusive, then commoditised across cloud platforms, with pricing power consolidating at the layer that controls the model rather than the infrastructure. For hyperscalers, this is a structural margin threat: they are increasingly becoming compute and networking providers for AI workloads rather than differentiated AI solution vendors. For OpenAI and other frontier labs, multi-cloud distribution is a leverage play heading into IPO — broader distribution means more revenue data to present to public market investors and reduces the risk of any single cloud partner extracting concessions. The strategic question is whether Google, which controls both a frontier model and a cloud platform, can resist this commoditisation dynamic longer than Microsoft — or whether Gemini's cloud-native integration advantage will also erode as enterprise customers demand model portability.
China's AI Capital Controls: A New Category of Deal Risk
Beijing's forced unwinding of the Meta-Manus acquisition should be understood not as a one-off political intervention but as the operationalisation of a policy framework that has been building since the 2021 data security law cycle. China is treating advanced AI capabilities — models, training data, and the engineers who built them — as strategic assets subject to export controls analogous to physical dual-use technology. This creates a new category of deal risk that M&A lawyers and fund managers have not previously had to price: completed transaction unwinding risk in the Chinese AI sector. Venture funds with Chinese AI portfolio companies, particularly those structured with offshore holding vehicles, need to reassess the liquidity assumptions underlying those positions. The assumption that a Chinese AI startup could be acquired by a US strategic buyer as an exit pathway is now materially impaired. This also has implications for US firms that have been hiring Chinese AI talent — the policy signal suggests Beijing may seek to restrict individual mobility as well as corporate transactions.
The AI Labour Paradox: Data Annotation Workforce Contracting as Compute Costs Dominate
The reported layoff risk facing over 700 Covalen workers in Ireland who were training Meta's AI models — combined with the Nvidia executive's public statement that compute costs currently far exceed employee costs in AI development — points to a structural transition in how AI is built. The early AI development model was labour-intensive: vast numbers of human annotators, raters, and trainers were required to generate the supervised learning signal that models needed. As models become more capable of generating and evaluating their own training data through synthetic data generation and reinforcement learning from AI feedback, the human annotation workforce is contracting. This is not a labour cost optimisation story — it is a fundamental shift in the production function of AI. Capital is substituting for labour at the data generation layer, which means the AI development supply chain is becoming more capital-intensive and less labour-intensive simultaneously with the infrastructure buildout. The political and regulatory implications of this transition — particularly in jurisdictions like Ireland with strong worker protections — have not yet been fully priced by the market.
Explore Other Categories
Read detailed analysis in other strategic domains