Meta's Multibillion-Dollar Google TPU Deal, ASML's EUV Breakthrough, and Rare Earth Shortages Are Reshaping AI Compute

Three developments this week collectively represent the most significant shift in AI compute economics since Nvidia's H100 became the default currency of artificial intelligence infrastructure. Meta signed a multibillion-dollar deal to rent Google's Ironwood TPUs. ASML demonstrated a 1000-watt EUV light source that could boost chip output 50 percent by 2030. And rare earth suppliers reported that allocation-based sales of critical minerals have become the norm as yttrium and scandium shortages deepen.

Each story is significant on its own. Together, they reveal an AI chip supply chain being restructured across its entire depth — from raw materials to chip design to manufacturing throughput.

Meta Breaks the Nvidia Monopoly Assumption

The most immediate shock came from Meta's decision to rent Google's custom TPUs for training and deploying next-generation large language models. The deal, reported by The Information and confirmed by multiple outlets, involves Google's latest Ironwood TPUs — the seventh generation of Google's custom AI silicon.

The deal's significance lies in its context. Within the same month, Meta committed multibillion-dollar orders to Nvidia for Vera Rubin GPUs and to AMD for Instinct MI400 chips. Meta is not replacing Nvidia — it is deliberately constructing a multi-vendor supply chain that prevents any single chipmaker from holding leverage over its AI roadmap.

Google's ambition is to capture up to 10 percent of Nvidia's data center revenue. A separate joint venture with an unnamed investment firm will lease TPUs to additional customers, creating a TPU-as-a-service layer that competes directly with Nvidia's cloud GPU ecosystem.

Meta's own custom MTIA chips have experienced technical delays, making the external diversification more urgent. But even if MTIA were on schedule, the strategic logic of multi-vendor sourcing would remain compelling at Meta's scale of AI investment.

For a deeper analysis of how this deal fits into the broader AI chip diversification trend, see today's feature article.

ASML's 1000-Watt Milestone Eases the Manufacturing Bottleneck

While Meta reshuffles the demand side of the chip equation, ASML addressed the supply side with a breakthrough that could relieve one of the AI industry's most persistent constraints.

ASML researchers demonstrated a new EUV light source capable of reaching 1000 watts — up from the 600-watt systems currently in production. The technology fires three lasers at 100,000 tin droplets per second, nearly doubling the current rate, to generate more intense extreme ultraviolet radiation.

The practical impact: wafer processing throughput jumps from approximately 220 wafers per hour to 330 — a 50 percent improvement. Since every leading-edge AI chip (Nvidia, Google TPU, Apple, AMD) requires EUV lithography, and each EUV machine costs upward of $300 million with multi-year lead times, this throughput gain is the economic equivalent of adding new fabrication capacity without building new fabs.

The 1000-watt system is targeted for deployment around 2030, but the successful demonstration in 2026 gives chipmakers like TSMC, Samsung Foundry, and Intel the confidence to plan capacity expansions around the projected capability.

ASML remains the sole commercial manufacturer of EUV lithography systems. No competitor — including Chinese and American rivals — is remotely close to matching this performance. The breakthrough reinforces ASML's monopoly while simultaneously easing the production constraints that have limited AI chip supply.

Rare Earth Shortages Add a Third Pressure Point

Beneath the chip design and manufacturing stories lies a more fundamental vulnerability. Aerospace and semiconductor suppliers reported this week that shortages of critical rare earth elements — particularly yttrium and scandium — are worsening despite eased trade tensions with China.

China controls approximately 60 percent of global rare earth mining and 90 percent of processing capacity. Limited import volumes, difficult licensing regimes, and competing demand from both AI and defense sectors are forcing suppliers to prioritize their largest customers. Smaller chip companies and equipment manufacturers are being turned away entirely.

These materials are not optional luxuries. Neodymium magnets power cooling systems. Yttrium appears in chip packaging ceramics. Gallium enables compound semiconductors. Germanium is essential for advanced transistor architectures. A shortage in any of these constrains production regardless of design innovation or manufacturing throughput.

The rare earth bottleneck adds strategic value to the chip diversification trend. Google's TPUs and Nvidia's GPUs draw on partially different materials supply chains through different foundry and packaging partners. A disruption affecting one supply chain may not equally impact the other — making multi-vendor sourcing a form of materials risk hedging as well as price competition.

Also This Week

Samsung Galaxy S26 AI-First Launch. Samsung unveiled the Galaxy S26 with expanded AI capabilities and integrated Google Gemini for multi-step task automation — ordering food, booking rides, and managing workflows. The company is positioning smartphones as primary AI inference devices, with the on-device versus cloud processing split becoming a key differentiator.

Bipartisan AI Standards Bill. U.S. senators reintroduced a proposal supporting voluntary AI standards, benchmarks, and transparency guidelines, along with national laboratory testbeds for AI and frontier technologies. The bill focuses on maintaining American AI competitiveness through standards infrastructure rather than restrictive regulation.

Alphabet Absorbs Intrinsic. Alphabet's robotics software company Intrinsic is consolidating into Google proper, reflecting deeper integration of robotics tooling, AI, and developer ecosystems under one platform. The move signals Google's commitment to AI beyond cloud compute — into physical-world automation.

LLM Deanonymization Research. Researchers demonstrated that large language models can infer authorship and potentially deanonymize text at scale, raising concerns for whistleblowers, dissidents, and anonymous communities. The finding highlights an underappreciated security dimension of increasingly capable language models.

What to Watch

The Meta-Google deal opens a window onto how AI compute procurement will evolve through 2026 and 2027. Key indicators to monitor include whether other hyperscalers follow Meta's multi-vendor template, how quickly Google's TPU joint venture signs additional customers, whether ASML's 1000-watt EUV timeline holds, and how rare earth supply diversification efforts progress in the United States and allied nations.

The AI chip market is transitioning from a monopoly to a market — and the speed of that transition will determine the economics of artificial intelligence for the rest of the decade.

Sources

• SiliconANGLE: Google and Meta Strike New Multibillion-Dollar AI Chip Deal
• Benzinga: Google Scores Major Win as Meta Agrees to Multi-Year AI Chip Rental Deal
• Tom's Hardware: ASML Makes Breakthrough in EUV Chipmaking Tech
• TechStartups: Top Tech News February 27, 2026
• TrendForce: ASML Demonstrates EUV Light Source That Could Lift Chip Output 50%