Cerebras IPO Deep Dive: S-1 Filing, $22-25B Valuation, OpenAI $10B Mega-Deal
On April 17, 2026, Cerebras Systems formally filed its S-1 prospectus with the SEC, applying to list on Nasdaq, targeting May 2026. This is one of the most significant IPOs in the AI chip industry in 2026. This article provides a deep analysis of Cerebras's finances, strategy, customers, and future.
IPO Key Data
| Item | Details |
|---|---|
| IPO filing date | 2026-04-17 (S-1 submitted) |
| Target listing date | 2026-05 (Nasdaq: CBRS) |
| Valuation | $22-25B |
| 2025 revenue | ~$510M (+150% YoY) |
| 2025 net loss | ~$200M (still losing, but loss rate narrowing) |
| Key mega-deal | OpenAI $10B long-term inference compute contract |
| Major customers | OpenAI, G42, Mistral, Meta, Mayo Clinic |
| Underwriters | Goldman Sachs / Morgan Stanley / J.P. Morgan |
| Founder | Andrew Feldman (CEO) |
Financial Data (from S-1)
Revenue Growth
| Year | Revenue | YoY |
|---|---|---|
| 2023 | ~$80M | — |
| 2024 | ~$200M | +150% |
| 2025 | ~$510M | +155% |
| 2026 (E) | ~$1.2B | +135% |
Revenue grew 6.4× over three years, making it one of the fastest-growing companies in the AI chip industry.
Net Loss
| Year | Net Loss | Loss Rate |
|---|---|---|
| 2023 | ~$120M | -150% |
| 2024 | ~$180M | -90% |
| 2025 | ~$200M | -39% (loss rate narrowing) |
| 2026 (E) | ~$50M | -4% |
Loss rate narrowed from -150% to -39%, expected near breakeven in 2026.
Customer Concentration
| Customer | Revenue Share |
|---|---|
| G42 (UAE) | ~25% |
| OpenAI | ~20% |
| Other enterprises | ~55% |
G42 + OpenAI = 45% of revenue, relatively high customer concentration, though this ratio will change as the OpenAI contract grows substantially.
Key Mega-Deal: OpenAI $10B Inference Contract
In December 2025, Cerebras announced a $10B long-term inference compute contract (10-year term) with OpenAI:
| Item | Details |
|---|---|
| Contract value | $10B (10-year term) |
| Service | OpenAI model inference compute |
| Start date | 2026 Q2 |
| Annualized amount | $1B/year |
| Hardware | CS-3 (WSE-3) + future CS-4 (WSE-4) |
| Significance | Cerebras revenue base significantly expanded |
This contract elevated Cerebras's 2026 revenue forecast from ~$700M to ~$1.2B.
Customer List
Flagship Customers
| Customer | Industry | Application |
|---|---|---|
| OpenAI | AI lab | GPT series inference |
| G42 | UAE sovereign AI | National AI infrastructure |
| Meta | Internet | Llama training |
| Mistral | AI company | Model training + inference |
| Mayo Clinic | Healthcare | Medical AI training |
Research / Government Customers
| Customer | Application |
|---|---|
| Argonne National Lab | Scientific computing |
| Los Alamos National Lab | National security |
| NASA | Climate simulation |
| Sandia National Lab | Defense |
Enterprise Customers
| Customer | Industry |
|---|---|
| GSK | Pharmaceutical |
| AstraZeneca | Pharmaceutical |
| Total | Energy |
| BMW | Autonomous driving |
| Daimler | Autonomous driving |
Core Product: WSE-3 (CS-3)
| Item | Specification |
|---|---|
| Transistors | 4 trillion |
| Cores | 900,000 |
| On-chip SRAM | 44 GB |
| Compute (BF16 sparse) | 125 PFLOPS |
| Compute (FP8) | 250 PFLOPS (estimated) |
| Memory bandwidth | 21 PB/s |
| Interconnect | SwarmX (multi-WSE interconnect) |
| TDP | ~25 kW |
| Price | ~$3-5M/system |
| Release date | 2024 |
Future Product: WSE-4 (CS-4, Estimated)
| Item | WSE-3 | WSE-4 (Est.) | Improvement |
|---|---|---|---|
| Process | TSMC 5nm | TSMC 3nm | +1 generation |
| Transistors | 4 trillion | ~5 trillion | 1.4× |
| Cores | 900,000 | ~1,500,000 | 1.67× |
| SRAM | 44 GB | ~80 GB | 1.8× |
| SRAM bandwidth | 21 PB/s | ~40 PB/s | 1.9× |
| BF16 compute | 125 PFLOPS | ~200 PFLOPS | 1.6× |
| TDP | 25 kW | ~30-35 kW | 1.3× |
| Release date | 2024 | 2027 expected |
⚠️ WSE-4 has not been officially announced; above are estimates.
Valuation Analysis
Valuation Multiples
| Metric | Cerebras 2025 | Valuation Multiple |
|---|---|---|
| Revenue | $510M | 43-49× (at $22-25B valuation) |
| Net loss | -$200M | — |
| 2026 Revenue (E) | $1.2B | 18-21× |
| 2027 Revenue (E) | $2.5B | 9-10× |
At 2026 revenue, valuation multiple of 18-21× is similar to SaaS; at 2027 revenue, 9-10× approaches NVIDIA's long-term multiple.
Valuation Comparison with NVIDIA
| Vendor | Market Cap | 2025 Revenue | P/S Ratio |
|---|---|---|---|
| NVIDIA | ~$4,000B | ~$130B | ~31× |
| Cerebras | $22-25B | $510M | 43-49× |
| AMD | ~$280B | ~$25B | 11× |
Cerebras P/S ratio higher than NVIDIA — the market has high growth expectations for Cerebras.
Investment Highlights
1. Wafer-Scale Technology Leadership
- Single chip 125 PFLOPS BF16 = 5× H100 single card (BF16 989 TFLOPS)
- 44 GB SRAM far exceeds HBM: SRAM is 1000× faster than HBM
- 21 PB/s memory bandwidth: 6000× H100 HBM
2. New Opportunities in the Inference Market
- OpenAI $10B contract = 10 years of long-term revenue
- Ultra-low-latency inference: vs GPU HBM latency, SRAM is 1000× faster
- Rack-level integration: single system can infer trillion-parameter models
3. Software Ecosystem Maturity
- Cerebras Software Platform (CSoft): based on PyTorch
- JAX + Cerebras backend: Google integration
- vLLM 0.7+ Cerebras backend (estimated)
- HuggingFace integration
- Triton + Cerebras backend
- OpenAI-compatible API (Cerebras Inference)
Investment Risks
| Risk | Impact |
|---|---|
| Ongoing losses | Still losing $200M in 2025 |
| Customer concentration | G42 + OpenAI = 45% of revenue |
| High TDP | 25 kW/chip, cooling challenges |
| High price | $3-5M/system |
| NVIDIA Groq acquisition | Competition intensifies in ultra-low-latency inference |
| WSE-4 delay | Not until 2027, while AMD MI400 / NVIDIA Rubin GA in 2026 |
Post-IPO Impact
1. Compute IPO Industry Reshuffle
| Company | Status | Market Cap |
|---|---|---|
| NVIDIA | Listed | ~$4,000B |
| Cerebras | About to list | $22-25B |
| Groq (acquired by NVIDIA) | Exited IPO track | — |
| SambaNova | Still private | Est. $5-10B |
| Tenstorrent | Still private | Est. $4-7B |
| Cambricon | Listed on A-shares | ~$8B |
After listing, Cerebras will be the largest pure-play AI chip public company besides NVIDIA.
2. Domestic Chinese Wafer-Scale Chips
Progress of Chinese domestic wafer-scale AI chips:
- Biren Technology BR104: 300 TFLOPS (FP16)
- Iluvatar CoreX Bi-150: 200 TFLOPS
- Moore Threads MTT S5000: 250 TFLOPS
Cerebras listing will stimulate funding and listing of domestic wafer-scale AI chips.
Detailed Product Pages
- Cerebras WSE-3 Full Specs
- Cerebras WSE-4 Estimated Specs
- Cerebras WSE-2 (Previous Gen)
- WSE Architecture Details
- NVIDIA Groq 3 LPX (Competitor)
- Future Roadmap
Summary
The Cerebras IPO is one of the most significant events in the AI chip industry in 2026:
- S-1 filed 2026-04-17, targeting May 2026 Nasdaq listing: CBRS
- Valuation $22-25B, 2025 revenue $510M
- OpenAI $10B long-term contract = 10-year revenue foundation
- Wafer-scale technology leadership (single chip 125 PFLOPS)
- Loss rate narrowing (from -150% to -39%)
- WSE-4 2027 release = first post-IPO generation product
After listing, the AI compute market will form a NVIDIA + Cerebras duopoly.