Skip to main content
STATUS: PRE-CONSTRUCTION · SITE A UNDER EXCLUSIVITYNODE: VASILIKOS-01 — 34.7246°N, 33.2247°ECAMPUS: RISC-V PHASE 1 · MULTI-SILICON EVAL · PLANNEDPOWER: 42MW ON-SITE GENERATION · DESIGN TARGETSTATUS: PRE-CONSTRUCTION · SITE A UNDER EXCLUSIVITYNODE: VASILIKOS-01 — 34.7246°N, 33.2247°ECAMPUS: RISC-V PHASE 1 · MULTI-SILICON EVAL · PLANNEDPOWER: 42MW ON-SITE GENERATION · DESIGN TARGET
AGICY.AI
StackTechnology OverviewRISC-V sovereign stackComputeBare-metal EU compute
FacilitiesData CentersVasilikos campusSustainability100% renewable mission
HardwareTenstorrent GalaxyPhase 1 Blackhole fleet (pre-COD)AESOLAR AlpineEnergy stack · hail-class PV + BESSAMD HeliosPhase 2 open rack-scale (eval · pre-COD)Taalas HC1Phase 2+ · model-hardwired ASIC (watchlist · pre-COD)CerebrasPhase 2 · wafer-scale eval (pre-COD)IBM z17 / LinuxONE 5Phase 2+ · trusted AI next to data (pre-COD)d-Matrix CorsairPhase 2–3 · on Watchlist (pre-COD)AWS Trainium 4Phase 2+ · custom XPU watchlist (pre-COD)
Submit your Hardware for reviewPropose accelerators for the Vasilikos fleet
GatewayCopperwayEU-sovereign OpenAI-compatible gatewayTry PlaygroundNewLive Copperway demo · PII vaultSovereign Exchange5-year cross-org sovereign planFor BuildersWhat developers can run today
ProductsReserve CapacityPre-construction LOI tiersMarketplaceCompute marketplaceGPUs Rent LiveLiveEU partner GPU now · until CODCompute VouchersSovereign compute creditsModel LeaderboardFrontier model rankingsPricingSubscription tiers
SolutionsInferenceProduction inferenceTrainingtt-train · planned campusFine-TuningLoRA, PEFT, domain modelsSovereign CloudEU-jurisdiction cloudAgentic PlatformAutonomous AI agentsReadinessPublic-data AI readiness scan
IndustriesFinancial ServicesRegulated finance AIHealthcareClinical & life sciencesGovernmentSovereign govt workloadsSemiconductorsFab & EDA data, EU-residentEnergyGrid & generation forecastingUtilitiesNIS2 essential entitiesPublic SectorCivil & public benefitProcurementLawful tender criteria
PricingTiers
Capital & EducationInvestInstitutional data room & deal flowAcademyAI training programs
Individuals & Family OfficesLiving in EUNewClass B capital allocation · no visa framingInternationalNewPlan B · equity alternative to propertyGreece Golden Visa€250k / €400k / €800k bands · vs Class BCyprus Permanent ResidenceReg. 6(2) parallel counsel · vs Class B
IntelligenceResearchPublications & portalsPublic-Record DeskGEMI registry · filings · courtsData CentersVasilikos campus briefing
CompanyAboutBrand · HoldCo targetMissionCharter & sovereigntyTrust CenterSecurity portal · docs · status
Schedule Briefing
Sign In
RESEARCH & ANALYSIS

Cerebras CS-3: Wafer-Scale Training (July 2026)

Inside the WSE-3 wafer-scale engine — 4 trillion transistors, 900,000 AI cores, and why AGICY evaluates CS-3 as Phase 2 optionality (not an owned fleet).

Published: July 5, 2026Updated: 19 Aug 2026 · CS-4Read Time: 10 min
cerebras cs3 sovereign training

Updated 19 August 2026. Cerebras announced CS-4 on 18 August 2026 (Nexus rack, three WSE-3 Turbo wafers — not a new WSE-4 die). This article is the 5 July 2026 CS-3 deep-dive. Living product page: /hardware/cerebras. AGICY has not ordered CS-3 or CS-4.

The Cerebras CS-3, powered by the WSE-3 wafer-scale engine, was Cerebras' shipping wafer-scale system when this article was published (5 July 2026). With 4 trillion transistors, 900,000 cores, and 44 GB of on-chip SRAM, Cerebras claims ~1,800 tokens/sec on Llama 3.1 70B. Unlike inference-only accelerators, the CS-3 supports training up to 24 trillion parameters via weight streaming (vendor-claimed). AGICY evaluates CS-3 / CS-4 as Phase 2 optionality — not an owned training hall.

WSE-3: The World's Largest AI Chip

Cerebras' third-generation Wafer Scale Engine occupies an entire 300 mm silicon wafer — 56× larger than the biggest discrete GPU. By keeping the die intact, data never leaves the chip boundary, eliminating the inter-chip communication overhead that plagues multi-GPU clusters.

  • 4 trillion transistors— fabricated on TSMC 5 nm
  • 900,000 AI-optimised cores — sparse-linear-algebra engines
  • 44 GB on-chip SRAM — no external HBM required
  • 21 PB/s memory bandwidth — 1,000× more than an H100
  • Weight Streaming architecture — decouples compute from model storage, enabling linear scaling across 64+ CS-3 systems for models up to 24T parameters
~1,800 tok/s
Llama 3.1 70B inference throughput

Measured on a single CS-3 system via the Cerebras Inference API. This is approximately 7× the throughput of Groq LPU and 20× faster than NVIDIA H100 tensor-core serving for the same model at comparable batch sizes.

Cerebras vs Groq LPU: Speed Comparison

DimensionCerebras CS-3 (WSE-3)Groq LPU
ArchitectureWafer-scale, weight streamingLPU (Language Processing Unit), SRAM-only
Llama 3.1 70B speed~1,800 tok/s~250 tok/s
Llama 3.1 8B speed~4,000 tok/s~800 tok/s
Training✓ Up to 24T parameters✗ Inference only
On-chip memory44 GB SRAM230 MB SRAM per chip
Max model size (single system)24T params (weight streaming)~70B params
Sovereign deploymentOn-prem & private cloudGroqCloud only (US-hosted)

Business Milestones

Cerebras' transition from stealth hardware lab to public company has been one of the fastest in semiconductor history:

  • NASDAQ IPO (CBRS) — May 2026:Listed at a $12 B valuation after clearing CFIUS national-security review, making it the first wafer-scale chip company on a public exchange.
  • $20B OpenAI training contract: Multi-year deal to supply CS-3 clusters for frontier model pre-training, displacing a portion of NVIDIA H100/B200 capacity.
  • AWS Marketplace partnership: CS-3 on-demand instances available in us-east-1 and eu-west-1, lowering the barrier for enterprises to benchmark wafer-scale performance.
  • 92% YoY revenue growth (Q1 2026): Driven primarily by sovereign-cloud contracts outside the US, signalling strong international demand for export-unrestricted AI training hardware.

Why AGICY Deploys Cerebras for Phase 2

AGICY's infrastructure roadmap is deliberately multi-silicon:

  • Phase 1 — Tenstorrent RISC-V (inference): 1,801 Galaxy servers targeting 17T tok/year for low-latency serving of open-weight models. Campus is pre-construction (COD TARGET H2 2027) — not operational.
  • Phase 2 — Cerebras CS-3 (training):Adds wafer-scale training capacity so EU customers can fine-tune and pre-train models up to 24T parameters without data leaving EU jurisdiction. Deployment target: Q1 2027.
  • Phase 3 — Hybrid orchestration: Unified scheduler routes workloads to the optimal silicon — Tenstorrent for inference cost-efficiency, Cerebras for training throughput — across the Vasilikos campus.

The combination of RISC-V inference and wafer-scale training creates a vertically sovereign stack: no CUDA lock-in, no US-CLOUD-Act exposure, and no dependency on a single silicon vendor.

“The future of sovereign AI requires both inference and training capability on EU soil. Inference without training is renting intelligence — training on sovereign hardware is owning it.”

GPU leasing, rent, buy, or colocation

Search intent we actually fulfil on /leasing: GPU leasing, rent GPU server, buy AI accelerator (client-site finance under T&Cs), or AI colocation / GPU colo — also typed as collocation. Planned Cyprus campus; not a live hall.

This page is Cerebras CS-3 — Phase 2 quote-only evaluation. GPU leasing / rent / buy / colocation via /leasing. Pre-construction.

  • Tenstorrent Galaxy — Phase 1 · list bands L1–L3
  • Cerebras CS-3 — Phase 2 · quote only
  • IBM z17 — Phase 2+ · quote only
  • AMD Helios — Phase 2 · quote only
  • d-Matrix — Phase 2–3 · quote only

GPU leasing estimate · Hardware roadmap · Reserve SRA

Reserve Sovereign Compute Capacity

Secure priority access to AGICY's Phase 2 Cerebras CS-3 training clusters. Fine-tune and pre-train frontier models entirely within EU jurisdiction.

Secure Your SRA Allocation →

References & Primary Sources

  1. Cerebras Systems — WSE-3 Chip Architecture and CS-3 Specifications
  2. Cerebras Inference Benchmarks — Llama 3.1 70B & 8B Throughput Results
  3. Groq LPU Technology — Architecture and Performance Claims
  4. SEC EDGAR — Cerebras Systems S-1 Filing and IPO Documentation (CBRS)
  5. AGICY Holdings Internal Business Plan, Sections 4.3 (Phase 2 Training) and 5.1 (Multi-Silicon Strategy), 2026.

§ FIN — Close of Document

Ready to build on sovereign infrastructure?

Schedule a confidential briefing with our team. NDA-protected, no commitment.

Schedule a briefing →
EU JURISDICTION · CYPRUSGDPR ART. 28 DPA-READY · BY DESIGNNIS2-ALIGNED · BY DESIGNEU AI ACT ART. 12 LOGGING SUPPORT · BY DESIGNRISC-V NATIVE · OPEN ISA

Design-alignment statements for a pre-construction facility — not certifications or attestations. Basis: compliance FAQ, § 08. Careers: we aim for 50-50 gender balance across hiring cohorts.

AGICY.AI

Advanced Governance & Intelligence Cyprus

The sovereign architecture for the Cyprus mind.
Humanitarian mandate: civilian public benefit only — healthcare, education, civil resilience. Civilian / humanitarian mandate only.
VASILIKOS ENERGY CENTRE, LIMASSOL DISTRICT · PRE-CONSTRUCTION
34.7246°N · 33.2247°E
Principal campus: Cyprus Vasilikos (Phase 1). Parallel HoldCo path: sovereign compute project in Greece (TARGET / planning) — ~20 MW-class Tenstorrent / air-cooled inference positioning for EU diversification; separate CapEx, no offtake claimed.

The Ledger — monthly briefing
  • CopperwayEU-sovereign OpenAI-compatible gateway
  • Try PlaygroundNewLive Copperway demo · PII vault
  • Compression & PII vaultNewSovereign path controls in Playground
  • Sovereign Exchange5-year cross-org sovereign plan
  • For BuildersWhat developers can run today
  • Reserve CapacityPre-construction LOI tiers
  • MarketplaceCompute marketplace
  • GPUs Rent LiveLiveEU partner GPU now · until COD
  • Compute VouchersSovereign compute credits
  • Model LeaderboardFrontier model rankings
  • PricingSubscription tiers
  • Research HubPublications & portals
  • Public-Record DeskGEMI registry, filings, court decisions
  • Data CentersVasilikos sovereign campus briefing
  • Frontier AI EthicsSovereign alternative to frontier dominance
  • Cyprus Court DecisionsBilingual legal research feed
  • Intelligence DrainEU talent & compute outflow analysis
  • National Equity in AISovereign AI for smaller EU nations
  • Compiler-First ChallengeJim Keller thesis vs GPU orthodoxy
  • GDDR6 & Open InfrastructureGalaxy economics & sovereign TCO
  • Blackhole Performance RisksVendor benchmarks & due diligence
  • Security Audit ResearchCybersecurity methodology & findings
  • AI Readiness AuditExecutive readiness assessment
  • Readiness HubPublic-data scans & named assessments
  • ProcurementLawful sovereignty criteria for tenders
  • AboutProject identity & status
  • MissionCharter & sovereignty
  • CareersCulture, benefits & hiring ethos
  • Open PositionsEngineering, research & operations roles
  • Ethics & CharterAnti-misconduct & responsible AI
  • Editorial & MethodologySources, claims, corrections
  • Trust CenterSecurity portal · docs · status
  • InvestInstitutional data room & deal flow
  • Living in EUIndividuals & FOs · Class B allocation
  • International investorsPlan B · Greece / Cyprus rails · Class B
  • Equity participation (legacy)CY & GR individual interest · counsel-gated
  • AcademyAI training programs
  • ContactBriefings & inquiries
  • Privacy PolicyGDPR · data processing
  • Terms of ServicePlatform usage terms
  • Cookie PolicyTracking & consent
  • SRA TermsReserve capacity agreement
  • Gateway Pricing DisclaimerCopperway pricing basis
© 2026 AGICY· PROJECT / BRAND OPERATOR · PLANNED CYPRUS ENTITIES NOT YET INCORPORATEDDOC: AGICY.AI · REV 2.0 · SOVEREIGN LEDGER
AGICY