Training Example: Phala – Review the Data, Give Your Score & Compare to the Real AI Evaluation

Secured HardwareDNA visual fallback
[H1] Trusted AI
Private execution. Verifiable results.Run agents, private LLM models, and GPU jobs inside hardware-backed TEEs. Keep secrets private, and prove what ran.npm install -g phalaStart buildingTalk to sales
[IMG: Phala]
[IMG: Google]
AWS$ phala deploy -c docker-compose.yml -n myapp building compose plan selecting CPU TEE: Intel TDX selecting GPU TEE: NVIDIA CC injecting encrypted env cvm ready: https://myapp.phala.network CPU machineGPU machineConfidential VM
[H2] Confidential AI cloud
Move existing Docker Compose workloads into CPU or GPU confidential machines. Keep the deploy path familiar; make the runtime verifiable.$phala deploy -c docker-compose.yml -n myappCheck detail pagehardware quoteruntime measurementverifier reportterminal · attestation.json$ curl https://api.phala.network/attest/<cvm-id> {"quote": "AgABAL8LAAAMAAsA...", "mrenclave": "a7f2c8d9e1b4...", "tcb_level": "UpToDate", "signature": "Intel/NVIDIA signed", }✓ TEE signature valid · code integrity verified Attestation
[H2] Every result can carry proof
Instead of asking users to trust a cloud claim, Phala emits runtime measurements that software can verify.$phala cvms attestation <cvm-id>Read attestation docs
[H2] Trusted by 5,000+ users
Trusted by industry leaders and developers worldwide.
[IMG: Nvidia logo]
[IMG: OpenRouter logo]
[IMG: Z.AI logo]
[IMG: OPPO logo]
[IMG: Venice AI logo]
[IMG: Intel logo]
[IMG: Near logo]
[IMG: Uniswap logo]
[IMG: Flashbots logo]
[IMG: Crossmint logo]
Products
[H2] Full service for AI privacy: agent sandbox, LLM, and GPU.
phala deploy -c docker-compose.yml -n myapp proof pathcloudRuntime sandbox
[H3] Agent sandbox
Run agent backends in a confidential VM with sealed keys, private memory, and verifiable execution.$phala deployDeploy sandboxH200US · 24 vCPU141GB VRAMIntel TDX + NVIDIA CCfrom $3.20/GPU/hrB300US · 16 vCPU288GB VRAMIntel TDX + NVIDIA CCfrom $5.60/GPU/hrproof pathgpuConfidential GPU
[H3] GPU marketplace
Launch H100, H200, and B300 GPU capacity with TEE-backed runtime proof and public attestations.$nvidia-smi -qOpen GPU marketplaceConfidential models
[H3] Private LLM models with real model choice.
OpenAI-compatible LLM endpoints, private prompts, and verifiable runtime state.Check allQwenQwenGoogleQwenDeepSeekPhalaPhalaMoonshotAIZ.aiQwenQwenMiniMaxQwenQwenGoogleQwenDeepSeekPhalaPhalaMoonshotAIZ.aiQwenQwenMiniMaxencryptedQwen: Qwen3.5-122B-A10B262K context$0.46/M inputCheck detail$Copy APIencryptedQwen: Qwen3 32B41K context$0.12/M inputCheck detail$Copy APIencryptedGoogle: Gemma 4 31B262K context$0.15/M inputCheck detail$Copy APIencryptedQwen: Qwen3.6 35B A3B262K context$0.20/M inputCheck detail$Copy APIencryptedDeepSeek: DeepSeek V4 Pro800K context$1.50/M inputCheck detail$Copy APIencryptedPhala: Gemma-4 26B-A4B Uncensored (Heretic)66K context$0.15/M inputCheck detail$Copy APIencryptedPhala: Qwen3.6 35B-A3B Uncensored (Aggressive)131K context$0.30/M inputCheck detail$Copy APIencryptedMoonshotAI: Kimi K2.6262K context$1.09/M inputCheck detail$Copy APIencryptedZ.ai: GLM 5.1203K context$1.21/M inputCheck detail$Copy APIencryptedQwen: Qwen3.5-27B262K context$0.30/M inputCheck detail$Copy APIencryptedQwen: Qwen3.5 397B A17B262K context$0.55/M inputCheck detail$Copy APIencryptedMiniMax: MiniMax M2.5197K context$0.20/M inputCheck detail$Copy APICheck all
[H2] All-in-one confidential compute platform for AI workloads.
Platform
[H3] Built for private AI work
Write code, dockerize, and deploy it as trustless TEE apps.EasyBuild in minutesOpenAudit everythingPrivateHardware-level securityVibe codingmarvin@Mac ~/ai-agent % claude codeClaude Codebun ‹ claudedeploy this agent privately on Phala and prove what ran $ phala deploy -c docker-compose.yml -n ai-agent ✓ CVM created · encrypted env loaded · app booting $ phala cvms get ai-agent --json · public_urls[0].app → https://ai-agent-7bd8.dstack-prod.phala.network $ phala cvms attestation <cvm-id> ✓ Attestation valid. Image digest matches. Safe to use.
[H2] Proven at Scale
Built for enterprise security and regulatory requirements.Building with confidential AI0+UsersRuntime proofs generated and checked0+Daily AttestationsNear-native confidential GPU execution0%TEE PerformanceTotal VMsLive network source from DuneOpen sourceConfidential model tokens/day2026-05-293.6BCrawled from Phala's OpenRouter provider chart during server render.daily tokens10 active modelsView token source
[H2] Real-World Success Stories
Discover how leading companies are leveraging Phala's confidential AI to build exceptional digital experiences, while maintaining complete data privacy and regulatory compliance.Financial
Services
[H3] Private Financial AI
Phala enabled us to process sensitive trading data with AI while maintaining complete regulatory compliance. We've reduced compliance costs by 40% while improving model accuracy.Healthcare
Research
[H3] Medical AI with Sealed PHI
Multi-party collaboration on patient data without privacy compromise. Accelerated drug discovery by 60% while maintaining HIPAA compliance.AI SaaS
Platform
[H3] Enterprise AI SaaS
Phala's confidential AI helped us land Fortune 500 clients who required verifiable data protection. Increased enterprise sales by 300%.Decentralized
AI
[H3] Decentralized GPU and AI Economy
Built autonomous trading agents with verifiable execution. Users trust our AI because they can verify every decision on-chain.
[H2] Enterprise-Grade Compliance & Security
Deploy confidential AI with confidence. Phala is SOC 2 Type I certified and HIPAA compliant, with ISO 27001 certification in progress and privacy-by-design controls aligned with GDPR.Visit Trust CenterSOC 2 Type I CertifiedHIPAA CompliantISO 27001 In Progress99.9% Uptime SLAGDPR Compliant Processing24/7 Enterprise SupportFAQ
[H2] Common Questions & Answers
Find out all the essential details about our platform and how it can serve your needs.1
[H3] What is Trusted Execution Environment (TEE)?
TEE is a secure area inside a processor that protects code and data from the operating system, hypervisor, and other applications.2
[H3] How does confidential AI protect sensitive data?
Sensitive data and AI models remain private during processing by running inside hardware-backed secure environments.3
[H3] Is Phala compatible with existing AI frameworks?
Yes. Phala supports existing Docker services and popular AI frameworks including TensorFlow, PyTorch, and Hugging Face.4
[H3] What are the performance implications?
Confidential GPU workloads typically target near-native performance, with roughly 5-10% overhead depending on workload and hardware.5
[H3] How can I verify the security of my AI workloads?
Phala exposes cryptographic attestations so users and systems can verify the workload and runtime state.6
[H3] How do I get started?
Install the Phala CLI, deploy a Docker workload, then inspect status, logs, and attestation from the command line.Start building
[H2] Build AI you can prove.
Deploy private workloads, verify execution, and scale from models to GPU jobs.Start buildingOpen dashboardVibe codingTalk to salesinstallcopy$ npm install -g phala Install the CLI, then deploy private workloads with verifiable runtime state.

Contact
[H1] Talk to Phala.
Founder welcome video16 secBy submitting, you agree to our Terms of Service and Privacy Policy.
[IMG: Phala]
[IMG: Phala]
Private execution. Verifiable results.Newsletter© 2026 Hashforest Technology. All rights reserved. Privacy • Terms

Confidential VM
[H1] Better Security.Less complexity.
Run Docker in a confidential VM.Deploy existing containers into hardware-backed TEEs. Keep AI secrets private, and prove what ran.
[IMG: Phala Cloud]
Phala Cloudmanaged network
[IMG: dstack-cloud]
dstack-clouddeploy on AWS, GCP, or BYOH
[IMG: AWS]
AWS
[IMG: Google Cloud]
GCPBYOHBYOHVibe codingRead CVM docsContact salesWhat is TEE hardware creates an isolated runtime for your workload. the CPU or GPU runs code inside a measured boundary. that measurement covers firmware, runtime, image, and configuration. Private memory keeps operators, hosts, and cloud admins out. secrets stay in encrypted memory while apps keep running. operators can schedule the workload without reading it. attestation proof Phala turns TEE hardware into a developer primitive. proof can leave the cloud and travel with the output. verify hardware, firmware, image, and runtime before trusting the result.01
[H3] privacy
TEE memory keeps secrets away from hosts and operators.02
[H3] verified runtime
Attestation binds hardware, OS, VM, image, and app config.03
[H3] public proof
Agents can attach an inspectable proof to the output.Secure hardware paths
[H2] One runtime for CPU and GPU TEEs.
Use the same Docker deployment path for agent services, private APIs, inference workers, and GPU jobs. Phala handles the secure hardware layer and exposes proof for verification.Vibe coding · verifyRead attestation docsView Trust Centerhardware proof rail
[H3] CPU, GPU, and runtime state stay on one verifiable path.
Intel TDXConfidential VM memory and CPU execution.NVIDIA GPU TEEPrivate inference and GPU memory isolation.Phala attestationProof binds hardware, image, compose, and result.CloudApps CVMs 3Attestations Deployments Application
[H3] ai-agent
verified endpointhttps://agent.phala.network app id0x530f...7bd8 compose hash8df2...ad91 agent-api 0.36 vCPU running AMS worker-gpu H200 TEE verified SFO attestation quote.json ready global custom-domain https online edge self hosteddstack-cloudmanaged cloudPhala CloudGlobal infrastructure
[H2] Phala Cloud Worldwide Network
Distributed CVM and GPU capacity across multiple continents for low-latency confidential compute.vCPUs268017 allowed nodesGPU devices648 teepodsnamed regions◎522 node rows
[H3] Nodes (17)
Filter??gpu-in1IN-DL-1GPUIntel Xeon Platinum 8580 / 240 vCPUs / H200 x8??gpu-in2IN-DL-1GPUIntel Xeon Platinum 8580 / 240 vCPUs / H200 x8??gpu-use1US-EAST-1GPUIntel Xeon Platinum 8592+ / 256 vCPUs / H200 x8??gpu-use2US-EAST-1GPUIntel Xeon Platinum 8592+ / 256 vCPUs / H200 x8??gpu-use3US-EAST-1GPUIntel Xeon Platinum 8592+ / 256 vCPUs / H200 x8??prod1US-EAST-1CPUIntel Xeon Gold 6554S / 144 vCPUs??prod10FR-PARIS-1CPUIntel Xeon Gold 5515+ / 32 vCPUs??prod11US-WEST-1CPUIntel Xeon Gold 6530 / 128 vCPUs??prod12US-WEST-1CPUIntel Xeon Gold 6530 / 128 vCPUs??prod2US-WEST-1CPUIntel Xeon Gold 5512U / 56 vCPUs??prod3US-WEST-1CPUIntel Xeon 6780E / 288 vCPUs??prod4US-WEST-1CPUIntel Xeon Gold 6530 / 128 vCPUs??prod5US-WEST-1CPUIntel Xeon 6710E / 128 vCPUs??prod6EU-WEST-1CPUIntel Xeon Gold 6554S / 144 vCPUs??prod7US-WEST-1CPUIntel Xeon 6710E / 128 vCPUs??prod9US-WEST-1CPUIntel Xeon 6710E / 128 vCPUs??tdxlab-testnetUS-WEST-1CPUUnknown CPU modelEasy to use
[H2] Vibe-code a CVM
A developer or coding agent can generate a compose file, deploy it to a CVM, seal secrets, and fetch attestation without learning TEE provisioning.Vibe codingmarvin@Mac ~/ai-agent % claude codeClaude CodeOpus 4.7 · xhigh effort · Phala CLI installedDeploy this agent as a confidential VM. Keep OPENAI_API_KEY private and return proof. ⏺ I found docker-compose.yml, added encrypted env handling, and will deploy with Phala CLI. $ phala deploy -c docker-compose.yml -n ai-agent ✓ sealed OPENAI_API_KEY locally ✓ CVM scheduled on Intel TDX node ams-tdx-02 ✓ service endpoint ready: https://ai-agent.dstack.phala.network $ phala cvms attestation ai-agent --json ✓ quote verified · compose hash matches · runtime measurement ready ⏺ The VM is live. I attached attestation.json so users can verify what ran. docker-compose.ymlgeneratedservices:
agent:
image: ghcr.io/team/agent:latest
ports:
- "8080:8080"
environment:
- OPENAI_API_KEY=${OPENAI_API_KEY}
- DATABASE_URL=${DATABASE_URL} 01
[H3] Bring your compose file
Use the Docker Compose file you already ship with: images, ports, volumes, private registries, and multi-service apps.Read docs02
[H3] Seal secrets before upload
Encrypted secrets are encrypted locally before they are sent to Phala Cloud. Only the CVM TEE can decrypt them at boot.Read docs03
[H3] Return proof for users and agents
Attestation proves genuine TEE hardware, expected runtime state, and the exact app configuration that ran.Read docs
[H2] What Phala handles for the CVM path.
The point is not to make teams assemble TEE primitives by hand. Phala turns private runtime, proof, deployment, and operations into one product surface.
[H3] Private runtime memory
TEE hardware protects workload memory from the host, operator, and cloud layer.
[H3] Existing Docker workflow
Bring Docker Compose, ports, volumes, environment variables, and private registries.
[H3] Encrypted secrets
Secrets are encrypted client-side and decrypted only by the CVM TEE at boot.
[H3] Attestation proof
Verify the hardware quote, runtime measurements, and compose hash.
[H3] Cloud operations
Public HTTPS endpoints, logs, updates, lifecycle controls, SDKs, and Terraform.
[H3] CPU and GPU TEEs
One confidentiality story for agent services, APIs, inference workers, and GPU jobs.Works with the apps and infra you already run24 integrations
[IMG: GitHub]
GitHub
[IMG: Docker]
Docker
[IMG: Kubernetes]
Kubernetes
[IMG: OpenAI]
OpenAI
[IMG: Anthropic]
Anthropic
[IMG: Hugging Face]
Hugging Face
[IMG: TensorFlow]
TensorFlow
[IMG: PyTorch]
PyTorch
[IMG: LangChain]
LangChain
[IMG: Jupyter]
Jupyter
[IMG: Databricks]
Databricks
[IMG: Supabase]
Supabase
[IMG: MongoDB]
MongoDB
[IMG: AWS]
AWS
[IMG: GCP]
GCP
[IMG: NVIDIA]
NVIDIA
[IMG: Intel]
Intel
[IMG: Ethereum]
Ethereum
[IMG: Solana]
Solana
[IMG: Coinbase]
Coinbase
[IMG: Vercel]
Vercel
[IMG: Next.js]
Next.js
[IMG: Phala Cloud]
Phala Cloud
[IMG: dstack]
dstackQuick Win Confidential VM
[H3] Phala Cloud
Managed CVM capacity, deployment workflow, endpoints, logs, updates, and attestation surfaced through one cloud dashboard.$phala deployCloud CLI docsPhala managedCVMControl Plane
[H3] Self-hosted dstack-cloud
Run the control plane on AWS, GCP, or bring-your-own hardware when the boundary, rack, or region must stay under your operation.self hostedphala cloudBYOH / AWS / GCPCVMFAQ
[H2] Common questions
Everything you need to know about Confidential VM1
[H3] What's the difference between Intel TDX and NVIDIA GPU TEE?
Intel TDX provides VM-level CPU isolation while NVIDIA GPU TEE offers hardware-secured GPU memory and compute for AI workloads.2
[H3] Do I need to modify my Docker containers for TEE deployment?
No modifications needed. Your existing containers work as-is with automatic TEE security applied at infrastructure level.3
[H3] How do I verify my application is running in genuine TEE?
Use attestation endpoints to get cryptographic reports signed by Intel/NVIDIA hardware proving TEE genuineness and code integrity.4
[H3] What's the performance impact of memory encryption?
CPU workloads see 2-5% overhead, GPU AI/ML workloads see 5-7% overhead while maintaining hardware acceleration.5
[H3] Can Phala operators access my running containers?
No. TEE hardware prevents any external access including from operators, cloud providers, or system administrators.6
[H3] How does environment variable encryption work?
Variables are encrypted with TEE public keys during deployment. Only your running TEE instance can decrypt them.7
[H3] What compliance standards does this meet?
SOC 2 Type I certified and HIPAA compliant, with ISO 27001 in progress. Supports GDPR and FedRAMP requirements with hardware-backed security guarantees and audit trails.8
[H3] How do I debug applications running in TEE?
Standard debugging tools work through encrypted channels. Remote debugging, logging, and profiling maintain security.

GPU TEE Marketplace
[H1] TEE-ready GPUs for AI builders.
H100, H200, and B300 capacity with CVMs, dual attestation, and TEE-aware operations.Trial a machine for 24 hours, reserve a slot, or quote dedicated clusters. Phala handles the hard part: confidential GPUs, Intel TDX runtime, NVIDIA attestation, and the DevOps required to keep it working.Trial nowVibe codingQuote priceConfidential GPU cloud
[H2] Capacity first. Proof after the workload runs.
H100 / H200 / B300
[IMG: NVIDIA H100 chip]
[IMG: NVIDIA H200 chip]
[IMG: NVIDIA B300 chip]
hardware proof rail
[H3] H100, H200, and B300 move through one verifiable GPU path.
GPU TEEH100GPU TEEH200GPU TEEB300Trial24h minimumReserveslots and clustersVerifyCVM + GPU evidenceMarketplace inventory
[H2] Capacity with proof built in.
Pick a GPU for a 24-hour trial, reserve a slot for sustained jobs, or quote a dedicated cluster. Every path starts from TEE-ready infrastructure instead of a raw GPU box.Trial ready
[H3] NVIDIA H100
[IMG: NVIDIA H100 chip]
Proven confidential inference and fine-tuning capacity.Memory80GB HBM3Bandwidth3.35 TB/sRegionUS-WestScale1-2 GPUsOn-demand$3.08/GPU/hr24h minimumSlot$2.38/GPU/hrreservedTrial nowDetailsSlot ready
[H3] NVIDIA H200
[IMG: NVIDIA H200 chip]
High-memory runtime for larger private model jobs.Memory141GB HBM3eBandwidth4.8 TB/sRegionUS-West / IndiaScale1-8 GPUsOn-demand$4.80/GPU/hr24h minimumSlot$3.20/GPU/hrreservedTrial nowDetailsQuote now
[H3] NVIDIA B300
[IMG: NVIDIA B300 chip]
Blackwell Ultra confidential capacity for frontier inference.Memory288GB HBM3eBandwidth8 TB/sRegionUS-East / US-WestScale1-8 GPUs1-month$6.50/GPU/hr30d minimumSlot$5.60/GPU/hrreservedTrial nowDetailsPrices include Intel TDX + NVIDIA confidential computing readiness. Volume and enterprise pricing are quoted by workload.Quote price
[H3] Performance metrics for private AI GPU planning
H100H200B300relative index1xH1001.9xH2003.2xB300LLM inferencemodel + KV cache80GBH100141GBH200288GBB300GPU memoryfeed batches3.35TB/sH1004.8TB/sH2008TB/sB300Memory bandwidth
[IMG: NVIDIA H100 chip]
NVIDIA H10080GB HBM3
[IMG: NVIDIA H200 chip]
NVIDIA H200141GB HBM3e
[IMG: NVIDIA B300 chip]
NVIDIA B300288GB HBM3eGPU comparison
[H2] H100 vs H200 vs B300
Compare the capacity shape before the quote. H100 is the fast trial path, H200 adds memory headroom, and B300 is the Blackwell Ultra path for frontier inference and dedicated clusters.Exact throughput depends on model, batch size, precision, and runtime. Phala quotes the GPU together with the confidential VM path, GPU CC readiness, and attestation operations.GPU cloud mockup
[H3] Capacity lanes with proof state.
The marketplace view should make the buying motion obvious: trial, reserve, then scale into a dedicated cluster with TEE readiness attached.
[IMG: NVIDIA H100 chip]
H10080GB HBM3from$3.08/hr
[IMG: NVIDIA H200 chip]
H200141GB HBM3efrom$4.80/hr
[IMG: NVIDIA B300 chip]
B300288GB HBM3efrom$6.50/hrverifiedCVM runtimeverifiedGPU CC modeverifiedDual attestationGPU TEE proof path
[H2] What Phala handles for the CVM path.
GPU isolation is only useful when the entire path — runtime, GPU mode, and evidence collection — is verifiable end-to-end. Phala delivers all three together.

++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++

cvm-enclave · 80×24 · 24fpsdensity: .:-=+*#%@01
[H3] CVM runtime
Docker workloads run inside an Intel TDX confidential VM with GPU passthrough. The runtime is sealed against the operator and measured by firmware before the workload starts.
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
@ @
@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @@@@ @ @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ ...: :::: :... .::: :::: ...: :::: ::.. .::: :::: ...: :::: : @
@ @
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
gpu-cc · 80×22 · 24fpsdensity: .:-=+*#%@02
[H3] GPU CC mode
NVIDIA Confidential Computing seals model weights, activations, and KV cache inside protected GPU memory. The GPU enforces compute isolation alongside the CPU TEE.

@@@@@@@@@@@@
@=--::--=++@+-:.
@=--::--=++@#*+-::....
@@@@@@@@@@@@ ...........
.......... @@@@@@@@@@@@@
........... @======++***@
..........@=====++****@
..........@====++*****@
.::::........... @===++******@
.:-*%@@%*-:. @@@@@@@@@@@@@
@@@@@@@@@@@@ ...........::::.
@:::-==++++@..........
@:::-==++++@
@@@@@@@@@@@@

dual-attestation · 80×20 · 24fpsdensity: .:-=+*#%@03
[H3] Dual attestation
Intel TDX and NVIDIA each emit a signed quote. Phala collects both and exposes them through one verifier so the CVM and the GPU prove themselves together.Buying paths
[H2] Start small. Reserve when it works.
The marketplace is structured around how AI builders actually buy GPUs: test quickly, reserve capacity when a workload proves out, then move to enterprise deals when the cluster becomes production-critical.01 / On-demand
[H3] Trial a confidential GPU in 24 hours.
Short test windows for builders validating private inference, model serving, or proof generation.Trial now02 / Slot
[H3] Reserve capacity before the next run.
Predictable GPU access for sustained training, fine-tuning, and benchmark windows.Quote price03 / Enterprise
[H3] Dedicated clusters with TEE operations.
Custom H100, H200, or B300 deals with TEE-aware infrastructure support and deployment planning.Talk to salesAI solution paths
[H2] Use private models where AI touches secrets.
The private model endpoint is the first entry point. The same privacy primitive extends to agents, data workflows, and training.LLM API
[H3] Private AI inference
Serve OpenAI-compatible model calls where prompts, outputs, and customer context need encrypted-in-use protection.Open solutionencrypted
[H4] DeepSeek V3.1
128K$0.27/M inputencrypted
[H4] Qwen3 Coder
256K$0.40/M inputencrypted
[H4] Llama 3.3 70B
128K$0.15/M inputencrypted
[H4] GPT OSS 120B
128K$0.10/M inputencrypted
[H4] Claude Sonnet 4.5
200K$3.00/M inputencrypted
[H4] Gemini 2.5 Pro
1M$1.25/M inputAgents
[H3] Private AI agents
Run agents with keys, tools, memory, and actions inside a verified runtime instead of a visible automation cloud.Open solutionTraining
[H3] Private model training
Adapt models on proprietary data while keeping datasets, gradients, checkpoints, and evaluation traces inside the boundary.Open solutionprivate training run
[H4] Observe without exposing weights.
H100 CC01
[H5] dataset
sealed02
[H5] fine-tune
running03
[H5] eval
private04
[H5] checkpoint
verifiedloss curveproof attachedattestation.jsonData
[H3] Private AI data
Move models to sensitive records and return approved outputs without exposing raw data to the model operator.Open solutionsourceEHR datasourceCustomer recordssourceInternal docsTEE clean room
[H4] query without raw access
approved output
[H5] aggregate only
no row exportproof linked
[IMG: Phala]
[IMG: Phala]
Private execution. Verifiable results.Newsletter© 2026 Hashforest Technology. All rights reserved. Privacy • Terms