Together AI Secures $800M for GPU Infrastructure, Sticky Eurozone Inflation Pauses ECB Cuts, and Copper Drops to Multi-Month Low

Three stories from the first week of July 2026 encapsulate the contradictions in the global macro-investment landscape: Together AI's $800M Series C at $11B (GPU cloud infrastructure for open-source AI workloads) confirms the AI infrastructure "stack" is receiving the most concentrated capital in venture history — in 2026, the top 10 infrastructure deals represent 84% of all AI VC by dollar volume. Eurozone core CPI stuck at 2.5% (services inflation at 4.1%) forces ECB President Lagarde to abandon September cut expectations at Sintra — the exact moment German Bund yields jump 12bps to 2.45% — amplifying the transatlantic monetary policy divergence and hitting European banks (who hold €2.3T in sovereign bonds) with MTM losses. And copper's 3.2% single-session drop to $8,950/tonne (LME warehouse inventories +15% in one week to 210,000 tonnes = 2-year high) reveals a structural split between copper's near-term demand weakness (China property sector, Eurozone manufacturing PMI at 45.5%) and its 5-year structural bull case (EV fleet, data centre power distribution, offshore wind cables, AI cooling systems).

---

📈 Together AI — The Open-Source GPU Cloud Thesis

$11B Valuation and What It Reflects About AI Infrastructure

Together AI profile:

What Together AI does — the open-source LLM infrastructure layer: The landscape of AI models has bifurcated into:

1. Proprietary models (OpenAI GPT-4o, Anthropic Claude, Google Gemini): API access only, fixed pricing, no model customisation
2. Open-source models (Meta Llama 3, Mistral Large, Qwen-2, Falcon): Full weights publicly available, require compute infrastructure to run

Together AI operates in the open-source layer — providing the compute infrastructure and optimised inference engines that allow developers to run open-source LLMs at scale:

Why enterprises prefer open-source infrastructure (Together AI's core thesis):

The cost differential is the most compelling driver: for enterprises processing 10B tokens/month (large enterprise use case), the annual cost difference is $174M/year (OpenAI GPT-4o) vs $6.5M/year (Llama 3 via Together).

The Series C mechanics — who and why:

Nvidia as investor is the key signal: Nvidia investing in Together AI means Together will receive priority allocation of H100/H200/B200 GPUs — critical in a market where GPU wait times are 4–6 months. This gives Together a structural supply advantage over competing GPU cloud providers.

The broader AI infrastructure VC landscape — 2026:

---

💶 ECB and Eurozone Inflation — The Sintra Inflection Point

Services Inflation at 4.1% and the Pause Mechanics

Eurozone June 2026 CPI — full breakdown:

The services inflation "last mile" problem: Services inflation stuck at 4.1% is the ECB's "last mile" problem — the same phenomenon the Fed experienced in 2025. Services inflation is sticky because:

1. Wage-price spiral in services: Services are 70–80% labour costs → wages rising +4.2% YoY in Eurozone → service prices rising at similar pace
2. Non-tradable prices: Services (haircuts, restaurants, housing) can't be "imported" at lower prices — not subject to global deflation pressures
3. Contract indexation: Many European service contracts have annual CPI escalation clauses — automatically renewing at CPI + fixed premium

The Lagarde "back to basics" framing at Sintra: Lagarde's Sintra intervention had two distinct messages:

• Message 1 (to markets): "We are not in a hurry to ease further" → explicitly removes September cut expectation
• Message 2 (to economists): "We have moved away from complex forward guidance and asset purchases" → signals the ECB is using rates as the sole tool, not QE or tiering

The bond market reaction — Bund yield to 2.45%:

The European bank MTM problem: European banks hold ~€2.3T in sovereign bonds (primarily Bunds, BTPs, OATs). When Bund yields rise 12bps:

• 10-year bond price falls ~1.1% (duration × yield change = 8.3 × 0.12%)
• On €2.3T portfolio → MTM loss of ~€25B across European banking system
• This doesn't hit P&L (bonds are held in "hold to maturity" category), but it erodes unrealised gains buffers

The more significant impact: higher Bund yields → higher sovereign borrowing costs for EU governments → fiscal constraint → less stimulus → amplifies economic contraction.

---

📉 LME Copper at $8,950 — The Near-Term Demand Gap vs Long-Term Bull Case

Inventory Surge and the Two-Horizon Copper Story

Copper price trajectory — 2026:

The LME warehouse inventory surge:

The geography of the inventory build is revealing: 61% of inventory is in Asian LME warehouses (Singapore, South Korea, Taiwan). This directly implicates China as the demand-side weakness driver — if Chinese buyers were actively purchasing, they would be drawing from Asian warehouses, not allowing inventory to accumulate.

Why China demand is weak despite government stimulus: China's government has announced multiple rounds of infrastructure and property sector stimulus in 2025–2026, but copper demand remains depressed because:

1. Property sector structural decline: China's residential construction is 20% below 2021 peak — each new home uses ~200kg of copper (wiring, plumbing, HVAC). With 2M fewer homes started/year, that's 400,000 tonnes of annual demand lost permanently
2. Credit transmission failure: Local governments are debt-constrained (land revenue collapsed with property market) → infrastructure stimulus isn't translating into construction starts
3. Export sector weakness: China's export factories are operating at lower capacity due to US/EU tariffs → less copper demand from manufacturing

The near-term vs long-term copper split:

The AI data centre copper demand — a new structural driver: A 1 GW AI data centre requires approximately:

• 8,000 tonnes of copper (power distribution, busbars, transformers, cooling systems)
• Global AI data centre buildout target: 100+ GW by 2030 (from current ~15 GW for AI workloads)
• Implied copper demand: 680,000 tonnes of new AI data centre copper demand by 2030

This is equivalent to adding ~3% of annual global copper demand (25M tonnes/year) from a single new use case that barely existed in 2023.

New mine supply — the Peru and DRC expansion: The near-term copper price pressure is being amplified by new mine supply coming online:

• Peru: Quellaveco (Anglo American) ramping to 330,000 t/year; Tía María (Southern Copper) expected 2026 production start at 120,000 t/year
• DRC: Ivanhoe Mines' Kamoa-Kakula Phase 3 expansion targeting 600,000 t/year by H1 2027

Combined: ~550,000 tonnes of new annual supply coming online in 2026–2027, against demand growth of only ~200,000 tonnes/year in the current weak environment. This supply-demand imbalance drives the short-term bearish outlook.

---

📌 The Bottom Line

• together-ai-800m-series-c-11b-gpu-cloud-open-source-inference: $800M/Salesforce+Nvidia+a16z+Sequoia/$11B; Nvidia as investor = priority GPU allocation (H100/H200 wait times 4-6 months = massive competitive moat); open-source cost advantage: GPT-4o $5+$15/1M tokens vs Llama 3 via Together $0.54+$0.54 — 10B tokens/month enterprise: $174M/year vs $6.5M/year ($167M annual savings); 4 enterprise drivers: data privacy + zero lock-in + regulatory compliance + fine-tuning; FlashAttention custom CUDA: 30-40% faster; Model Router: 20-35% cost reduction; competitors: CoreWeave ($35B), Lambda Labs ($5B), Baseten ($15B); AI infra 2026: top-10 deals = 84% of all AI VC by dollar volume.
• ecb-sintra-lagarde-services-inflation-4-1pct-bund-yield: Headline CPI 2.3% but services stuck at 4.1% (0.0pp improvement); "last mile" services sticky: 70-80% labour costs + wages +4.2% YoY + non-tradable (can't import deflation) + CPI-indexed contracts; Lagarde: "not in hurry to ease" = September cut off table; "back to basics" = rates only, no QE signals; Bund 10Y +12bps to 2.45%; European bank MTM: €2.3T sovereign bond portfolio → +12bps = ~€25B unrealised MTM erosion; Bund → sovereign borrowing costs → fiscal constraint → amplifies Eurozone contraction; services inflation removal condition: wage growth must fall to <3% (currently 4.2%).
• copper-lme-8950-210k-tonne-inventory-china-demand-ev-transition: LME copper −9.1% YTD to $8,950; inventory +15.4% in 1 week to 210,000t (2-year high); 61% in Asian warehouses = China demand weakness; cancelled warrants 28%→11% (demand interest collapsed); China property −20% from 2021 peak = 400,000t/year demand lost permanently; credit transmission failure (local government debt); near-term bearish: Peru/DRC +550,000t/year new supply vs +200,000t/year demand growth; long-term bull case: AI data centres 680,000t new demand by 2030 + EV fleet + offshore wind + green hydrogen electrolysers.

---

📬 Stay Updated

Get global market analysis delivered to your inbox every week. Subscribe to our free newsletter →

---

Disclosure: This post contains affiliate links. If you purchase through our links, we earn a small commission at no extra cost to you. We only recommend products we believe in.