Capital & Industrial Strategy
Top Line
Nvidia reported $96.2 billion in Q2 revenue — roughly double year-ago levels — and guided for approximately 70% revenue growth in fiscal 2028, a figure that far exceeded analyst consensus of 45% and signals that hyperscaler AI capex is compounding rather than plateauing.
Nvidia is in advanced talks to acquire Hugging Face at a valuation of approximately $12.9–13 billion, a deal that would give the chipmaker control of the dominant open-source AI model hub and shift Nvidia's strategic posture from hardware supplier to full-stack AI platform operator — subject to regulatory review and not yet confirmed as closed.
Anthropic signed a confirmed $45 billion compute agreement with UK infrastructure startup Nscale, securing roughly 460 megawatts of capacity in West Virginia, while simultaneously deepening its commercial partnership with Salesforce — demonstrating that frontier AI labs are locking in multi-year infrastructure positions at a scale that rivals hyperscaler commitments.
OpenAI's Jalapeño inference chip outperformed Nvidia Blackwell systems on key efficiency benchmarks, while OpenAI separately committed $400 million as the sole investor in a new early-stage AI venture fund — two moves that together illustrate OpenAI's ambition to control both the silicon layer and the startup ecosystem around it.
DeepSeek is in discussions to raise funding at a reported $74 billion valuation, and China's Z.ai confirmed it is the lab behind the benchmark-topping Ox Alpha open model, underscoring that Chinese AI labs are competing on both capital formation and technical output at a scale that demands Western strategic attention.
Key Developments
Nvidia's Blowout Quarter and 70% Growth Guidance Reframes the AI Supercycle Debate
Nvidia posted $96.2 billion in Q2 revenue, beating forecasts, and guided for approximately 70% revenue growth in fiscal 2028 — against analyst consensus of 45%. CEO Jensen Huang stated on the earnings call that 'our demand is much greater than 70%', framing the guidance as a conservative floor rather than a ceiling. The result, reported across Bloomberg, CNBC, WSJ, and Reuters, effectively forecloses the 'AI spending plateau' narrative that had been building in investor circles through mid-2026.
Huang also addressed pointed criticism around Nvidia's practice of providing financing to AI customers — so-called 'circular financing' — defending the investments as 'once in a generation' opportunities with 'limited risk', per CNBC and the FT. Separately, Amazon confirmed it has tripled its Nvidia chip order, adding 2 million GPUs over two years with a partnership structure that extends beyond procurement, per TechCrunch. Nvidia also disclosed it sold a small number of H200 chips into China during the quarter — the first such sales with US government sanction — though volumes fell short of permitted limits, per Bloomberg. Meanwhile, a US investigation into Singapore-based Apex Logistics for alleged Nvidia chip smuggling to China signals that enforcement pressure on export controls is intensifying alongside the partial market reopening.
Nvidia's Hugging Face Acquisition Talks Signal a Platform Land Grab
Nvidia is reportedly nearing an agreement to acquire Hugging Face at approximately $12.9–13 billion, according to Bloomberg citing The Information and confirmed independently by Reuters. This remains an announced negotiation, not a closed transaction, and would be subject to regulatory review. The strategic logic is clear: Hugging Face is the dominant distribution layer for open-source models, hosting repositories used by hundreds of thousands of developers and enterprises. Owning it would give Nvidia privileged insight into model adoption patterns, control over a critical developer community, and a platform to promote its own hardware as the natural runtime for open-source AI.
This move fits a broader pattern of Nvidia repositioning from a component supplier to a full-stack AI platform. Combined with its venture financing of AI startups, its NIM microservices layer, and its DGX Cloud partnerships, a Hugging Face acquisition would extend Nvidia's influence from training hardware into the model distribution and developer tooling layer — territory currently contested by AWS, Google, and Microsoft. Regulators in both the US and EU are likely to scrutinise the deal given Hugging Face's role as a quasi-public infrastructure for the open AI ecosystem.
Anthropic's Infrastructure Commitments Reach Hyperscaler Scale
Anthropic has signed a confirmed $45 billion compute agreement with Nscale, a UK-based infrastructure startup, securing approximately 460 megawatts of capacity at a West Virginia data center development, per sources cited by CNBC, FT, TechCrunch, and Reuters. This follows a string of similar compute deals by Anthropic and signals that frontier AI labs are now committing capital at a scale — and with a multi-year horizon — that was previously the exclusive domain of the major cloud providers.
Simultaneously, Salesforce confirmed an expanded partnership with Anthropic under the 'Claudeforce' branding, with Benioff positioning the combined product as 'a dynamic interface that thinks, reasons, and acts', per CNBC and Reuters. Salesforce stock jumped 12% on the combination of Anthropic stake gains and raised annual forecasts, offering a direct read-through on how equity markets are rewarding SaaS incumbents that credibly embed frontier AI. The Nscale deal and Salesforce partnership together illustrate Anthropic's two-track strategy: secure the compute capacity to train and serve competitive models, while monetising via enterprise distribution partnerships rather than building its own sales infrastructure.
OpenAI Moves on Two Fronts: Custom Silicon and Venture Capital
OpenAI disclosed that its Jalapeño inference chip outperformed Nvidia Blackwell systems on key inference-efficiency metrics, per CNBC and Semafor. OpenAI reportedly used its own AI models to compress the chip design and verification cycle, building the hardware in record time. This is strategically significant beyond the benchmark result: it demonstrates that major AI labs can credibly develop competitive silicon using AI-accelerated design tooling, a capability that Synopsys — which raised its annual forecasts on AI-driven chip design software demand per Reuters — is directly facilitating.
Separately, OpenAI confirmed it is the sole investor in a new $400 million early-stage AI venture fund, per WSJ. Sole LP status gives OpenAI full discretion over portfolio construction with no external governance constraints — a structure that allows it to use the fund as a strategic instrument to seed startups building on its APIs and platform, rather than as a financial return vehicle. Combined with the Jalapeño chip development, OpenAI is clearly executing a vertical integration strategy: controlling the model layer, the silicon it runs on, and the startup ecosystem that builds applications on top.
Chinese AI Labs Accelerate on Both Capital and Technical Fronts
DeepSeek is in funding discussions at a reported $74 billion valuation, seeking capital for R&D and compute infrastructure expansion, per WSJ. This remains an unconfirmed negotiation. Simultaneously, Z.ai confirmed it is the lab behind Ox Alpha, a model that topped multiple benchmarks and leaderboards before its identity was disclosed, with weights to be released as open-source imminently, per TechCrunch and Bloomberg. Z.ai's shares rose as much as 8.6% on the announcement. The stealth release strategy — launching anonymously to accumulate unbiased benchmark results before revealing the lab's identity — is a deliberate tactic to generate credible performance data free from geopolitical discount.
Alibaba's Qwen team also launched Qwen3.8-Flash with a focus on lower training costs, per Reuters, reinforcing the Chinese labs' consistent emphasis on efficiency as a competitive differentiator against US frontier models that prioritise raw capability. The combination of DeepSeek's valuation trajectory, Z.ai's benchmark performance, and Alibaba's cost-efficiency focus suggests Chinese AI labs are pursuing a multi-pronged strategy: compete on frontier benchmarks, compete on open-source ecosystem building, and compete on inference economics simultaneously.
Signals & Trends
Infrastructure Commitments Are Becoming Competitive Moats, Not Just Cost Items
Anthropic's $45 billion Nscale deal, Amazon's tripled Nvidia chip order, and Kioxia's new Japanese fab investment collectively illustrate a structural shift: AI infrastructure commitments are now being made at 5–10 year horizons with contractual scale that forecloses optionality. This is no longer capital expenditure in the traditional sense — it is strategic moat construction. Labs and hyperscalers that lock in compute capacity at today's prices and supply constraints gain a durable cost and availability advantage over later entrants. The implication for investors is that infrastructure access is becoming a prerequisite for frontier AI competition, not a variable input — and that companies without confirmed long-term compute agreements are increasingly disadvantaged regardless of their model quality.
The Inference Layer Is Emerging as the Primary Competitive Battleground
OpenAI's Jalapeño chip outperforming Nvidia Blackwell on inference efficiency, Alibaba's Qwen3.8-Flash emphasising lower training costs, and Nvidia's own inference-optimised product roadmap all point to the same structural shift: as training compute becomes commoditised among well-funded labs, inference efficiency — cost per token served at scale — is becoming the primary margin driver and competitive differentiator. Custom silicon designed specifically for inference workloads (as opposed to training) offers 2–5x efficiency gains over general-purpose GPUs for deployed models. This creates a bifurcated competitive dynamic: training remains Nvidia-dominated, but inference is increasingly contested by custom ASIC developers, opening a vulnerability in Nvidia's margin structure precisely as its revenue is at an all-time high.
AI Security and Identity Are Transitioning from Pilot to Production Deployment
CrowdStrike's 61% year-to-date share gain and 11% post-earnings jump — driven by what it described as AI threat escalation creating a 'Mythos moment' — combined with Okta's 20% post-earnings pop and disclosure that AI-related deals accounted for 30% of new bookings, signals that enterprise security is the first vertical where AI is driving material revenue acceleration at scale rather than in pilot. This is analytically distinct from most enterprise AI adoption narratives: security buyers are not piloting AI tools, they are procurement-mandating them in response to AI-generated threat escalation. Cigna's disclosed AI investment strategy focused on cost reduction in healthcare represents a contrasting pattern — a sector still in structured pilot phase. The divergence between security (production deployment) and healthcare (structured pilots) is a leading indicator of where the next wave of enterprise AI spend will concentrate.
Explore Other Categories
Read detailed analysis in other strategic domains