Capital & Industrial Strategy
Top Line
AMD and Anthropic have signed a confirmed deal for AMD to sell tens of billions of dollars in AI server chips to Anthropic and invest up to $5 billion in the startup — the largest AMD-Anthropic transaction to date, structurally tying chip supply to equity ownership and signalling Anthropic's determination to reduce dependence on Nvidia.
Alphabet has raised its 2026 capital expenditure guidance to $195–205 billion, causing free cash flow to turn negative and triggering a market selloff, as investors weigh whether the cloud backlog of $514 billion justifies a spending rate equivalent to Sweden's GDP.
Samsung has secured a confirmed $200 billion chip supply contract with Broadcom, the largest single foundry deal on record, reinforcing the accelerating vertical integration of AI infrastructure supply chains.
The US government has accused China's Moonshot AI of stealing Anthropic's model weights and accessing banned Nvidia chips to build Kimi K3, threatening sanctions in what Washington has characterised as industrial-scale IP theft.
IBM has cut its annual revenue growth forecast after mainframe sales fell 42%, with enterprise customers redirecting budgets toward AI infrastructure rather than traditional IT refresh cycles — a structural demand shift with broad implications for legacy enterprise vendors.
Key Developments
AMD-Anthropic Deal: Chip Supply Meets Equity — A Strategic Realignment Against Nvidia
AMD has confirmed a deal to supply Anthropic with tens of billions of dollars in AI server chips — described as a 2-gigawatt compute commitment — and will invest up to $5 billion in Anthropic equity. Reuters and WSJ both confirm the deal is signed. This is not a letter of intent — it is a binding commercial agreement. The strategic logic is layered: Anthropic secures diversified compute at scale, reducing concentration risk on Nvidia H100/H200 supply chains; AMD secures a lighthouse customer whose $47 billion annualised revenue run rate (per TechCrunch) makes it a credible volume anchor for AMD's MI300/MI350 roadmap; and the equity stake aligns AMD's financial upside with Anthropic's trajectory in a way pure chip sales would not.
This deal is Anthropic's latest in a string of infrastructure commitments this year and fits a pattern of frontier labs locking in multi-year compute supply while simultaneously raising the capital cost of entry for challengers. The concurrent announcement of Claude Opus 5 — with a cost-capability toggle — signals that Anthropic is also competing on inference economics, not just raw capability, a direct response to pressure from cheaper Chinese models.
Alphabet's $205 Billion Capex Commitment: Cloud Backlog Justifies Spending or Signals Bubble?
Alphabet raised its 2026 capex guidance to $195–205 billion, doubling year-on-year, and reported free cash flow turning negative in Q2. Bloomberg and WSJ report shares fell on the announcement despite a cloud revenue beat and an 82% jump in Google Cloud revenue. The bull case rests on the $514 billion cloud backlog — contracted but unrecognised revenue that provides forward visibility — and Gemini's approach to one billion monthly users. Sundar Pichai pushed back on competitive positioning concerns, citing enterprise AI adoption acceleration. Reuters
The bear case is equally coherent: OpenAI's total infrastructure commitment has now reached $750 billion through 2030 per TechCrunch, and the aggregate capex across hyperscalers implies that if AI revenue growth decelerates even modestly, balance sheets across the sector will be materially stressed. The market's reaction to Alphabet's results — selling on a beat — reflects investor fatigue with what Bloomberg describes as growing fears of an AI bubble. The critical question is whether the $514 billion backlog converts to recognised revenue on a timeline that justifies current capex velocity.
Samsung-Broadcom $200 Billion Foundry Deal: AI Infrastructure Vertical Integration Accelerates
Samsung Electronics has confirmed a contract worth more than $200 billion to manufacture chips for Broadcom, the largest single foundry supply agreement on record. Bloomberg and Reuters both confirm the deal. The strategic significance extends beyond Samsung's foundry ambitions: Broadcom has been building custom AI accelerator (XPU) business with hyperscalers including Google and Meta as an explicit alternative to standard Nvidia GPU procurement. This deal locks in manufacturing capacity for that pipeline at scale, and represents a direct challenge to TSMC's effective monopoly on leading-edge AI chip fabrication.
For Samsung, the deal is a critical credibility signal after years of yield and process struggles at its foundry division. For Broadcom, it secures supply chain optionality. For the broader AI infrastructure market, it reinforces a trend toward custom silicon — where hyperscalers co-design chips with partners like Broadcom rather than buy off-the-shelf Nvidia products — which has long-term implications for Nvidia's pricing power and market share in the data centre segment.
Moonshot AI Allegations: US-China IP War Escalates to Chip Access and Model Theft Claims
The US government has formally accused China's Moonshot AI of improperly accessing Anthropic's model weights and using banned Nvidia chips to build the Kimi K3 system, threatening sanctions. Semafor and Bloomberg both report the White House has characterised the activity as industrial-scale IP theft. The allegations are currently unproven in a legal sense — they are government accusations, not adjudicated findings — but the policy response is already in motion. Washington's framing of Kimi K3 as a stolen rather than independently developed model is politically significant: it allows US authorities to escalate export controls and sanctions without conceding that Chinese labs have organically matched US frontier capabilities.
The geopolitical dimension compounds an already fractured AI safety cooperation landscape. Reuters reports that US-China tensions are threatening bilateral AI safety cooperation, reducing the likelihood of any coordinated approach to frontier model risk. Meanwhile, the FT notes that Chinese AI talent is increasingly choosing to build at home rather than migrate to Silicon Valley, suggesting the talent dynamic is also shifting.
Enterprise AI Adoption: Winners and Losers Emerging Across the Stack
The Q2 earnings cycle is producing a clear bifurcation in enterprise AI beneficiaries. ServiceNow raised its annual subscription revenue forecast again on AI agent demand, with CEO Bill McDermott explicitly positioning the platform as governance infrastructure for autonomous AI deployment. Reuters Nokia also beat Q2 profit estimates on AI-driven network infrastructure demand. By contrast, IBM cut its annual revenue growth forecast after mainframe sales collapsed 42%, with its CEO acknowledging that customers are redirecting capital toward AI infrastructure rather than traditional enterprise IT. WSJ reports IBM's consulting business is also under pressure as clients pause discretionary spend pending AI-driven workflow redesign.
The buyout market is beginning to reprice this dynamic. Francisco Partners co-founder DJ Deb, raising a $21 billion fund, is explicitly targeting 'SaaS-pocalypse' dislocations — software assets trading at distressed multiples because AI disruption fear is being priced in ahead of actual revenue impairment. FT reports this as a contrarian thesis: Deb argues AI will not kill software wholesale but will create acquisition opportunities among over-sold assets. This is a meaningful capital allocation signal — private equity entering the AI disruption trade from the long side on legacy software.
Signals & Trends
DeepSeek Funding Pause Signals Chinese AI Capital Markets Are Not Immune to Narrative Risk
DeepSeek has told prospective investors it is suspending its second fundraising round, days after comments attributed to founder Liang Wenfeng about US-China AI competition went viral, per Bloomberg. The timing — concurrent with the Moonshot IP theft allegations and escalating US sanctions threats — suggests that Chinese AI labs now face a dual risk: US regulatory action that could restrict their access to compute and customers, and domestic investor caution triggered by heightened geopolitical visibility. DeepSeek's open-source positioning had been a strategic differentiator, pitching stability and accessibility to developers globally as US frontier labs restrict API access. A prolonged funding pause would create a vacuum in the open-source competitive dynamic precisely as that positioning was gaining commercial traction. This is a weak signal worth monitoring: if other Chinese labs follow with funding pauses, it would indicate that the Moonshot allegations have created a broader chilling effect on Chinese AI capital formation.
Venture Capital Is Concentrating, Not Dispersing — Khosla's $5.5 Billion Raise Continues the Mega-Fund Trend
Khosla Ventures is in confirmed talks to raise $5.5 billion — its largest fund in two decades — per Bloomberg. Combined with a16z leading Travis Kalanick's Atoms robotics round at $1.7 billion and Menlo's Anthropic thesis playing out at scale, the pattern is clear: AI venture capital is concentrating into established franchise funds with proven frontier model access, rather than dispersing to seed-stage generalists. This structurally disadvantages new entrants without existing portfolio relationships to frontier labs, and means LP capital is effectively being allocated not just to venture as an asset class but to specific fund managers with embedded network advantages in the AI ecosystem. The implication for startup founders outside the top-tier network is increasing difficulty accessing the check sizes needed to compete on compute-intensive applications.
Government Industrial Strategy Is Shifting from Subsidy to Procurement Mandate
South Korea's president hosted a US tech summit explicitly calling for a 'new AI era', while the UK's new AI minister confirmed chips and drones will be central to Andy Burnham's reindustrialisation agenda, per FT. Japan's AI strategy and Taiwan's adjacent tech positioning are also being actively repositioned, per FT. The pattern across multiple governments is a shift from passive subsidy regimes — which characterised the 2022–2024 period — toward active procurement mandates and strategic partnership frameworks that tie AI infrastructure investment to national industrial policy goals. The Google-Nvidia partnership with German robotics startup Microagi on factory humanoids, per Semafor, illustrates how US hyperscalers are positioning themselves to capture government-directed industrial AI spend in allied markets. For investors, the actionable signal is that government procurement is becoming a meaningful revenue channel for AI infrastructure providers in non-US markets, and deal structures are increasingly being shaped by industrial policy conditionality.
Explore Other Categories
Read detailed analysis in other strategic domains