The US–China AI & Semiconductor Race
Assessment
Two races run at once. Commercially, the US buildout has reached extraordinary scale — Nvidia posted record $81.6B quarterly revenue (up 85% YoY), $75.2B of it data-center, guided to $91B next quarter, and bolted on a $10B Anthropic stake, a Meta deal and a 5-gigawatt CoreWeave commitment, while Jensen Huang signed gigawatt-scale and sovereign-AI deals across South Korea. Capital is chasing it: global AI investment hit a record $344.7B in 2025 (US $285.9B), Google raised $85B in a single equity offering, and Samsung, SK Hynix and Micron each crossed $1T. But the same concentration breeds fragility — a chip sell-off and Broadcom's $300B single-day rout threatened the rally, Fed officials warned the capex boom is inflationary now with productivity gains years away, and progressive Democrats (Sanders, Ocasio-Cortez, Warren) pushed data-center moratoriums and AI taxes. Strategically, China is routing around US chips rather than buying them: DeepSeek V4 runs on Huawei GPUs, Huawei's 'Tau Scaling Law' claims a 1.4nm-equivalent path by 2031, and even cleared H200 sales to Alibaba, Tencent, ByteDance and JD.com stalled because Beijing blocked deliveries to force domestic substitution — Huang says he has 'largely conceded' China to Huawei. The export-control lever is losing grip; the contest is now over who can finance, build and power compute fastest, and over whether the US can keep TSMC's leading-edge nodes as a moat.
Events
- 8 Jun 2026 Nvidia signs gigawatt-scale AI partnerships across South KoreaSeoul
On a Seoul visit, Jensen Huang announced a slate of partnerships with SK Hynix, SK Telecom, Naver, Doosan Group, LG Group and Hyundai Motor Group, spanning advanced memory supply, gigawatt-scale AI cloud construction, sovereign-AI infrastructure, robotics and autonomous mobility. SK Hynix committed to a multi-year technology partnership to develop advanced memory for global AI data centers. The deals are designed to harden Nvidia's supply chain and expand its AI ecosystem across a key US ally in Asia. Coming as Huang has 'largely conceded' China to Huawei, the Korea push reads as Nvidia entrenching the allied side of the divide.
Supply lock-inBinding SK Hynix's advanced memory (HBM) and Korean cloud capacity to Nvidia entrenches a US-centric AI stack across a frontline ally, raising the cost for China of ever displacing it and tightening the chokehold over the memory Huawei's chips also need.Sovereign AISelling 'sovereign AI' infrastructure to national governments converts geopolitical anxiety into demand — every country wanting its own compute multiplies the gigawatt-scale buildout the Fed already warns is overheating.China substitutionAnchoring Korea's chipmakers to the Nvidia stack is the mirror image of Huawei's domestic push: the same week Beijing forces JD.com and Alibaba onto Huawei silicon, Nvidia locks SK Hynix and Hyundai onto its own — the race fragmenting into two hardware blocs. - 5 Jun 2026 Semiconductor sell-off threatens the Wall Street rallyUnited States
A sharp decline in semiconductor stocks endangered Wall Street's prolonged winning streak, driven by sector-specific headwinds rather than a macro shock. The slide raised fears of a broader market correction and underscored how dependent the rally had become on the chip sector. It landed days after Broadcom's $300B rout and amid Fed warnings that AI capex was inflationary, compounding the sense that the AI trade was overextended.
Concentration riskWith the rally resting on a handful of chip names, a sector sell-off becomes systemic — the index has no diversification cushion, so AI sentiment and overall market stability have fused into a single bet on continued data-center spend.Macro linkageA chip correction hits the same household 401(k)s already squeezed by April's 3.8% inflation, transmitting AI-bubble risk straight into consumer confidence and the cost-of-living story the Fed is fighting.Sentiment fragilityComing right after Nvidia's record beat still saw shares fall, the sell-off shows the tape now reacts to expectations, not results — a late-cycle tell that any guidance miss (as Broadcom proved) can erase hundreds of billions in hours. - 4 Jun 2026 Broadcom loses $300B in a single day after a weak revenue forecastUnited States
Broadcom's stock suffered a historic $300 billion rout after the company's quarterly revenue outlook missed analyst expectations. The single-day wipeout was one of the largest ever for a US company and erased hundreds of billions in market capitalization. It signaled a downturn in the semiconductor sector and intensified fears of a broader slowdown in chips and tech. The shock preceded by one day the wider semiconductor sell-off that threatened the Wall Street rally.
Forecast sensitivityA $300B loss triggered by a guidance miss — not an actual revenue decline — shows AI-chip valuations are priced for perfection: even a single soft forecast detonates more market cap than most companies are worth.Contagion triggerBroadcom's rout fed directly into the next-day semiconductor sell-off, demonstrating how one chipmaker's stumble propagates across a sector that the whole index now leans on — idiosyncratic news becoming systemic risk.Counterweight to NvidiaAgainst Nvidia's record $81.6B and parabolic guide, Broadcom is the cautionary data point: demand at the very top (Nvidia GPUs) can stay parabolic while the custom-silicon and networking layer beneath it cools, hinting the buildout is narrower than headline numbers imply. - 4 Jun 2026 TSMC signals chip price hikes and a $165B Arizona buildoutHsinchu, Taiwan
At TSMC's annual shareholders' meeting in Hsinchu, CEO C.C. Wei voiced strong confidence in sustained AI-driven growth across consumer, enterprise and sovereign applications, and signaled openness to raising chip prices to protect profitability. He detailed a $165 billion investment in new Arizona factories and ongoing R&D on ASML's High-NA EUV machines, and announced a 30% annual increase in employee profit-sharing for a second straight year with another 30% projected for 2026. Wei named autonomous vehicles a long-term driver and stressed Taiwan as TSMC's most efficient base. The remarks underscored TSMC's central role in the AI supply chain amid geopolitical tension.
Pricing powerTSMC openly raising prices on leading-edge nodes is the supply side capturing AI-boom rents — a direct margin transfer from Nvidia and every fabless designer, and a tax on the entire Western stack that has no near-term alternative foundry.Lithography moatThe $165B Arizona fabs plus High-NA EUV R&D widen exactly the EUV-lithography gap that US export controls weaponize against China — the same chokepoint Huawei's 'Tau Scaling' architecture is explicitly trying to route around by 2031.Geographic hedgeShifting capacity to Arizona while calling Taiwan its 'most efficient base' is TSMC pricing in Taiwan-Strait risk: the US gets onshore leading-edge supply, but Wei's framing concedes that diversification costs efficiency, keeping the island indispensable. - 3 Jun 2026 Google raises $85B in its largest-ever equity offering to fund AIUnited States
Google completed an $85 billion equity raise — the largest in its history — to finance a major expansion of AI infrastructure and spending. The move signaled an aggressive push to compete head-on in the AI buildout. It landed amid record sector capital flows, days after Anthropic's confidential IPO filing and SoftBank's €75B France pledge, and a day before Broadcom's $300B forecast-driven rout.
Capex arms raceAn $85B single raise shows the buildout is now a financing contest as much as an engineering one — hyperscalers are tapping equity markets at unprecedented scale to fund data-center capacity, the very spend the Fed flagged as inflationary.Dilution signalChoosing equity over cash or debt for AI capex implies even Google won't fund the buildout from its own balance sheet, a tell about how capital-hungry frontier compute has become and how investors are being asked to underwrite a payoff years away.Bubble fuelPouring $85B of fresh capital into infrastructure feeds the same concentration that makes the rally fragile — it deepens the system's dependence on AI demand materializing before the Broadcom-style forecast misses spread. - 2 Jun 2026 Asian memory makers each cross $1 trillion as the AI chip rally broadensUnited States
The AI sector showed no signs of slowing, with Samsung, SK Hynix and Micron each reaching a $1 trillion valuation amid an AI-driven memory rally. The same wave carried Anthropic's confidential IPO filing, SoftBank's €75B ($87B) AI infrastructure investment in France and Nvidia's new RTX Spark superchip for PCs. Samsung averted a strike by agreeing to pay bonuses of up to $370,000 per worker out of semiconductor profits, setting a potential profit-sharing precedent. The cluster of milestones underlined how broadly AI demand was repricing the chip supply chain.
Memory bottleneckSamsung, SK Hynix and Micron each hitting $1T reflects that high-bandwidth memory — not just Nvidia's GPUs — is now a binding constraint on AI compute, so the value is migrating up the supply chain to the three firms that control HBM.Labor captureSamsung paying up to $370,000 per worker to avert a strike shows AI-boom rents reaching the factory floor, and creates a precedent that could raise the cost base across the very memory supply Nvidia's Korea deals depend on.Capital breadthAnthropic's IPO filing plus SoftBank's €75B France pledge in the same beat shows the buildout drawing capital across model labs, sovereign infrastructure and chips at once — the breadth that pushed 2025 AI investment to a record $344.7B. - 1 Jun 2026 Fed officials warn AI investment is fueling inflation with uncertain productivity payoffWashington
Multiple Federal Reserve officials cautioned that AI-driven investment is boosting demand and inflation risk while productivity gains remain uncertain and may take years to materialize. St. Louis Fed President Alberto Musalem and others argued against relying on AI to solve inflation, contrasting with Kevin Warsh's view that AI is disinflationary. Fed Governor Lisa Cook noted AI investment is pushing up prices for chips, equipment and construction. The split exposed a live policy debate over whether the boom helps or hurts the inflation fight.
Demand pullHundreds of billions in data-center capex — Cook specifically flagged chips, equipment and construction — is inflationary now, while the disinflationary productivity gains are deferred, so in the near term the boom worsens the Fed's supply-shock dilemma rather than easing it.Policy cautionBy rejecting AI as an inflation solution (Musalem vs. Warsh), the Fed signals it won't loosen on a productivity promise, keeping rates higher and pressuring the very valuations — Nvidia, the memory makers, Google's $85B raise — driving the buildout.Internal splitThe Musalem-Warsh-Cook disagreement means the AI capex shock has no consensus read inside the FOMC, raising the odds of policy volatility as officials reassess each inflation print against an unproven productivity thesis. - 29 May 2026 Progressive Democrats escalate an anti-AI agenda with data-center moratoriums and taxesWashington
Five progressive Democrats — Bernie Sanders, Alexandria Ocasio-Cortez, Ro Khanna, Elizabeth Warren and Graham Platner — began leading a confrontational Democratic message on AI, proposing data-center moratoriums, taxes on AI companies and worker protections. They criticized AI-linked political money and the environmental impact of the buildout. The push tests whether grassroots skepticism of AI can harden into a lasting political force, opening a left-flank front against the boom.
New cleavageA progressive bloc attacking AI on energy, jobs and concentration cuts against Trump's industrial-policy embrace, turning AI from a bipartisan growth story into a contested partisan issue — even as Sanders separately converges with Trump on capturing AI windfalls.Buildout riskData-center moratoriums target the physical bottleneck — power and siting — so even symbolic proposals raise regulatory uncertainty for the gigawatt-scale projects the race depends on, the same projects already straining grids.Money politicsNaming AI-linked political money makes the buildout's lobbying spend a campaign liability, a channel through which the backlash can constrain firms via disclosure and donor scrutiny well before any moratorium passes. - 25 May 2026 pivotal Huawei unveils a 'Tau Scaling' chip design to bypass US sanctionsChina
Huawei announced a new chip-design principle called the Tau Scaling Law, aiming to produce industry-leading semiconductors equivalent to 1.4-nanometre processes by 2031. Instead of shrinking transistors, the approach reduces signal latency and improves data movement to reach advanced performance. It offers an alternative path amid US export controls that restrict China's access to cutting-edge EUV lithography tools, and underlines Beijing's drive for semiconductor self-sufficiency. The announcement carries direct implications for AI computing and geopolitical leverage.
Strategic end-runBy chasing performance through architecture (latency and data movement) instead of node shrink, Huawei targets the exact chokepoint — EUV-lithography access, the moat TSMC's High-NA spend widens — that US controls were built to exploit.Time horizonA 2031 target signals China is playing a long indigenous-capability game, betting controls buy the US years not permanence — and that domestic substitution, already forced onto JD.com and Alibaba, eventually erases the leverage.Doctrine shiftReframing 'leading-edge' as 1.4nm-equivalent performance rather than a literal node redefines the metric the whole controls regime is denominated in — if effective performance, not transistor pitch, becomes the benchmark, the chokepoint loses meaning. - 25 May 2026 Global AI investment hits a record $344.7B in 2025, US firms leadingUnited States
A report found global AI investment reached a record $344.7 billion in 2025, a 127.5% jump from the prior year, driven by competition among US, Chinese and other firms. The US led with $285.9 billion in corporate investment, followed by China and the UK. The findings highlighted AI's reach across healthcare, agriculture, defense and finance, while noting regulatory divergence — 150 AI-related laws enacted by G20 countries by end-2025. The numbers quantified the scale the Fed and progressive Democrats were reacting against.
US lead, by spendThe US's $285.9B of the $344.7B total — over four-fifths — quantifies the capital advantage behind Nvidia's records and Google's $85B raise, but spend-led leadership is exactly what China's substitution strategy is engineered to make irrelevant.Doubling paceA 127.5% year-on-year jump is the empirical backbone of the Fed's capex-inflation warning: investment more than doubling in one year is the demand-pull pressure Musalem and Cook flagged, with the productivity offset still unproven.Regulatory fracture150 AI laws across the G20 by end-2025 signals the governance layer is fragmenting as fast as the capital is flowing, raising compliance friction that compounds Europe's lag and hands first-mover advantage to whoever regulates least. - 21 May 2026 pivotal Nvidia posts record $81.6B revenue and concedes China to HuaweiUnited States
Nvidia reported record Q1 fiscal-2027 revenue of $81.6 billion (up 85% YoY) and net income of $58.3 billion, with data-center revenue at $75.2 billion, and forecast $91 billion for Q2. It announced a dividend hike to 25 cents, an $80 billion buyback, a $10 billion investment in Anthropic, a major Meta deal and a CoreWeave commitment targeting five gigawatts of AI facilities by 2030. CEO Jensen Huang said demand had 'gone parabolic' in the era of agentic AI but that he has 'largely conceded' the Chinese market to Huawei, assuming zero China data-center revenue in the outlook. Despite the beat, shares fell 1.6% after hours as investors sought hypergrowth and worried about Huawei competition.
Valuation ceilingShares falling 1.6% on 85% growth and a $91B forward guide means the bar is so high even record results disappoint — the late-cycle signal that price has outrun even a genuinely booming business, the same fragility Broadcom later exposed.Single point of dependence$75.2B of $81.6B from data centers, plus a $10B Anthropic stake and a 5GW CoreWeave deal, shows the AI economy funneling through and increasingly being financed by one company — making Nvidia's guidance the de facto leading indicator for the whole race.China write-offHuang openly 'conceding' China to Huawei and modeling zero China revenue is the market admitting export controls failed commercially — Nvidia now prices in the bifurcated stack rather than fighting it. - 14 May 2026 pivotal US clears Nvidia H200 sales to Chinese firms, but Beijing blocks the deliveriesWashington
The US Commerce Department approved around 10 Chinese companies — including Alibaba, Tencent, ByteDance and JD.com — to purchase Nvidia's H200 AI chips, with Trump negotiating a 25% revenue share from the sales, but no deliveries occurred. Beijing reportedly blocked or tightly vetted orders to protect domestic chip development, requiring buyers to demonstrate security procedures and non-military use. Huang visited China as part of a White House delegation, raising hopes for a breakthrough. Amid the uncertainty, Chinese firms pivoted to domestic chips from Huawei.
Reversed leverageWhen the US grants access and China refuses it, the export-control lever inverts — Beijing uses its own market as the chokepoint to force domestic adoption of Huawei silicon, the same substitution Huawei's Tau Scaling roadmap is built to supply.Policy incoherenceApproving H200 sales undercuts the controls regime's denial logic; Beijing's blockade then denies even the commercial benefit and Trump's 25% cut, leaving Washington with neither strategic denial nor revenue.Forced migrationBy stalling deliveries, Beijing pushes Alibaba, Tencent and JD.com onto Huawei chips now — manufacturing the demand that validates DeepSeek V4's Huawei-native design and accelerates the parallel stack. - 7 May 2026 pivotal DeepSeek V4, designed for Huawei chips, challenges US AI dominanceChina
DeepSeek released version 4 of its AI model, built to run on Huawei's GPU chips, demonstrating full Chinese autonomy in AI infrastructure — the culmination of a semiconductor push in which state-backed firms captured 30% of the legacy-chip market and produced 7-nanometre processors for Huawei. The open-source model, with undisclosed training data, threatens US firms like OpenAI and Anthropic, which accuse China of model distillation — claiming DeepSeek trained on outputs from ChatGPT and Claude. It arrived as US tech faced domestic opposition and ahead of major AI-company IPOs planned for late 2026.
Stack independenceA frontier-class model running on Huawei silicon (built on China's own 7nm) proves China can field competitive AI without US chips at all — the single development that most directly devalues export controls and underpins Huang's China write-off two weeks later.Market threatArriving as OpenAI and Anthropic prepare late-2026 IPOs, a credible open-source Chinese alternative pressures US firms' pricing and the valuations underpinning the whole buildout — the same valuations the chip sell-off would soon test.Distillation fightThe OpenAI/Anthropic claim that DeepSeek distilled ChatGPT and Claude outputs reframes the contest as an IP-leakage problem, motivating later proposals to embed Pentagon liaisons in US labs and lock down model access — controls migrating from chips to weights. - 1 May 2026 Pentagon signs new military AI deals with Nvidia, Microsoft and AmazonWashington
The US Department of Defense entered new contracts with Nvidia, Microsoft and Amazon to develop and deploy AI capabilities for military applications. The agreements signaled a deepening integration of commercial frontier-AI providers into defense systems. They gave the buildout an anchor customer outside the commercial cycle just as market scrutiny of AI valuations was rising.
Demand floorDefense contracts give the buildout a non-cyclical customer, insulating the leading AI firms from a commercial slowdown — exactly the kind of slowdown the later Broadcom rout and chip sell-off raised — while binding national security to a few private platforms.Dual-use stakesFolding Nvidia, Microsoft and Amazon into military AI sharpens the export-control calculus: the same firms selling globally now hold defense-sensitive contracts, raising the cost of any leakage to China and the stakes of the H200 approvals.Platform concentrationThree vendors capturing core defense-AI work concentrates national-security capability in the same names that already dominate the commercial stack, deepening the single-point-of-dependence risk Nvidia's earnings exposed.
Background
Nvidia's data-center business is the gravitational center of US markets — record $81.6B revenue with $75.2B from data centers, a $91B forward guide, a $80B buyback, a $10B Anthropic stake and a 5GW CoreWeave commitment by 2030. CEO Jensen Huang says demand has 'gone parabolic' in the era of agentic AI. The rally pulled indices higher but concentrated performance in a handful of chip and AI names, so a sector wobble now reads as systemic.
Financing the buildout has become its own front: global AI investment reached a record $344.7B in 2025 (up 127.5%), led by US firms at $285.9B; Google raised $85B in a single equity offering; SoftBank pledged €75B for AI infrastructure in France; and Samsung, SK Hynix and Micron each touched $1T valuations on AI memory demand. The scale is what the Fed and progressive Democrats are reacting against.
Rather than depend on throttled US imports, Chinese firms are building a parallel stack: DeepSeek V4 designed for Huawei chips (with state-backed firms holding 30% of the legacy-chip market and producing 7nm for Huawei), and Huawei's Tau Scaling Law aiming at 1.4nm-equivalent semiconductors by 2031 via latency and data-movement gains instead of pure transistor shrink. The goal is to make US export controls strategically irrelevant.
US policy oscillates between denial and case-by-case approval. When Commerce cleared ~10 Chinese firms (Alibaba, Tencent, ByteDance, JD.com) to buy Nvidia H200s — with Trump extracting a 25% revenue share — Beijing blocked the deliveries, using its own market access to push domestic substitution and exposing the limits of a controls-only strategy. Huang has 'largely conceded' the Chinese market to Huawei and assumes zero China data-center revenue.
Leading-edge manufacturing still funnels through TSMC, whose CEO C.C. Wei signaled price hikes and detailed a $165B Arizona buildout and High-NA EUV R&D. Whoever controls the most advanced node controls the race's hardware ceiling — which is exactly what Huawei's architecture-first 'Tau Scaling' is trying to route around.
The boom is contested at home: a semiconductor sell-off and Broadcom's $300B single-day loss threatened the rally, Fed officials (Musalem, Cook) warned AI capex is inflationary with unproven productivity gains, and a progressive bloc (Sanders, Ocasio-Cortez, Khanna, Warren, Platner) proposed data-center moratoriums and AI taxes — opening a left-flank political front against the buildout.