Press & media
VoltageGPU press kit Confidential GPUs you verify yourself.
Boilerplate, a dated fact sheet, the founder, logos, and every public page that mentions us, each one opened before it was listed. Interview or fact check: press@voltagegpu.com.
01Boilerplate
About VoltageGPU, ready to paste
In one sentence
VoltageGPU rents NVIDIA GPUs inside Intel TDX confidential VMs, where the tenant generates and checks the Intel and NVIDIA attestations, from Solaize, France.
In one paragraph
VoltageGPU is the trading name of VOLTAGE EI, a French company founded in 2025 by Julien Aubry in Solaize, France. It rents NVIDIA GPUs inside Intel TDX confidential virtual machines, billed per second, and lets the tenant check the machine without trusting VoltageGPU: from inside their own VM, on a nonce they choose, the tenant generates the Intel TDX quote and the NVIDIA GPU attestation and verifies both against Intel and NVIDIA. NVIDIA GPU attestation has been verified this way on single-GPU H200, H100 and RTX PRO 6000 Blackwell VMs and on an 8x H100 node, where the NVSwitch fabric is not verified. The evidence is published with dates, and the verifier, voltage-verify, is open source under the MIT licence. VoltageGPU also runs an OpenAI-compatible inference API limited to models served in trusted execution environments.
02Fact sheet
Dated facts, with where to check them
- Company
VOLTAGE EI, a French sole proprietorship (entreprise individuelle), SIREN 943 808 824. Legal notice
- Based in
Solaize, France
- Founded
2025, by Julien Aubry. Bootstrapped, no outside investors.
- What it sells
Confidential VMs: NVIDIA GPUs inside Intel TDX trust domains, where the tenant generates the Intel TDX quote and the NVIDIA GPU attestation from inside the VM, on a nonce they choose, and verifies both against Intel and NVIDIA. Also an OpenAI-compatible inference API limited to TEE models (14 in the catalogue at the time of the snapshot).
- NVIDIA GPU attestation verified from inside the VM
- H200 141 GB, single GPUevidence for H200 141 GB, single GPU
- 8x H100 80 GB nodeevidence for 8x H100 80 GB nodeAll 8 GPUs attested in NVIDIA Protected PCIe mode; the NVSwitch fabric is not verified.
- H100 80 GB, single GPUcapture not published
- RTX PRO 6000 Blackwell 96 GB, single GPUevidence for RTX PRO 6000 Blackwell 96 GB, single GPU
- Not verified
The NVSwitch fabric of the 8x H100 node. The 8x H200 node, which reads the same mode but has not been attested. B200 and B300.
- Live prices
- RTX PRO 6000 Blackwell, 1 GPU$3.80 per hour8 available now
- H100, 1 GPU$6.95 per hour8 available now
- H200, 1 GPU$8.08 per hournone available right now
- 8x H100 node$55.62 per hour1 available now
Attested Confidential VM tiers only, whole VM, read from our pricing snapshot at . Every tier is on the pricing page.
- Billing
Per second, no subscription. The first hour is charged at launch and the unused part is refunded when the VM stops.
- Data and hosting
French data controller; customer database in the EU (Frankfurt). GPU capacity comes from sub-processors listed publicly, in data centres that are not limited to the EU.
- Open source
voltage-verify, the verifier, MIT licence, Python, first released on 12 September 2026.
- Press contact
03Founder
Julien Aubry
Founder, VOLTAGE EI (VoltageGPU) · Solaize, France
Julien Aubry founded VoltageGPU in 2025 and runs it as a bootstrapped, solo-founded company, with no outside investors.
Questions in English or French, answered in writing.
“Hardware enclaves solve the trust problem, if you can prove the machine can't see the data.”
04Press kit
Logos and colours
The mark is silver and drawn for dark backgrounds. Need a larger file or a photo? Ask at press@voltagegpu.com.
Mark, transparent background
PNG, 500 × 500
Download Mark, transparent background, PNG, 500 × 500
Mark, centred, transparent background
PNG, 500 × 500
Download Mark, centred, transparent background, PNG, 500 × 500
App icon
PNG, 256 × 256
Download App icon, PNG, 256 × 256
Touch icon
PNG, 180 × 180
Download Touch icon, PNG, 180 × 180
- Background#050508
- Verified green#4ade80
- Ink#f2f2f6
Type: Inter for text, JetBrains Mono for figures. Keep the mark on a dark background, do not recolour it, and do not place it next to NVIDIA, Intel or Confidential Computing Consortium logos: none of them is a partner.
05Mentions
Where VoltageGPU has been mentioned
Public pages only, each one opened before it went on this list. Private conversations are not listed.
Coverage and articles
Articles written about VoltageGPU by someone else.
None yet. An article is listed here once it is published and still online, never before.
Directories and listings
GPU and AI directories that list us today. Pending requests are not listed.
ComputePrices
VoltageGPU provider page on ComputePrices (opens in a new tab)
An independent GPU price comparator lists VoltageGPU with its confidential and standard GPU prices. Their page says it was last reviewed on 18 September 2026.
GitHub, awesome-confidential-computing
voltage-verify in the awesome-confidential-computing list (opens in a new tab)
A curated list of confidential computing resources lists voltage-verify in its attestation section. Added through our pull request, merged on 17 September 2026.
Technical discussions
Public threads we opened or took part in. Taking part is not an endorsement by anyone.
GitHub, NVIDIA's nvtrust repository
Issue 153, filed from our 8x H100 node. Replies pointed to the driver-matched NSCQ library and explained that switch evidence needs configuration on the host side. Still open; the NVSwitch fabric remains unverified on our side.
Confidential Computing Consortium, attestation mailing list
NVSwitch attestation from inside a TDX guest, in the public list archive (opens in a new tab)
Julien Aubry's post with the logs and an open question. A list member replied on 21 September 2026.
Confidential Computing Consortium, attestation mailing list
Our correction of that post, in the public list archive (opens in a new tab)
Julien Aubry retracts his first explanation: the NSCQ session does open from the guest once the matching library is installed.
NVIDIA Developer Forums
The same question, asked with the logs. No reply as of 9 October 2026.
Independent reproduction
Other people who ran the attestation on our machines with their own tools, and wrote it down.
Custodian (custodylabs.co)
Custodian's home page: both proofs reproduced on a VoltageGPU Confidential VM (opens in a new tab)
Mohammed Zoheb Shaik, founder of Custodian, a key broker for model weights, repeated the Intel TDX and NVIDIA H200 attestation from his own tenant VM on 6 September 2026, with his own verifier and a nonce he generated. Cited with his permission; his report is not redistributed.
Custodian (custodylabs.co)
Custodian's hardware validation inside VoltageGPU Intel TDX guests (opens in a new tab)
Custody tests with vLLM on an RTX PRO 6000 Blackwell (19 September 2026) and an H100 (23 September 2026), both in confidential mode, written up by Custodian.
Serial Alice, our write-up
Serial Alice's attestation test on our H100 Confidential VM, and the vLLM fault it found
Serial Alice, in Portugal, verified the attestation chain of our single-GPU H100 Confidential VM with their own tools and found a vLLM fault under confidential computing. Published on our blog after they reviewed the text.
From us: open source and publications
Not mentions, but what the entries above point to.
GitHub
voltage-verify source code on GitHub (MIT) (opens in a new tab)
Binds an Intel TDX quote and an NVIDIA GPU attestation to a workload manifest, then verifies the bundle offline without trusting the provider.
PyPI
voltage-verify on PyPI (opens in a new tab)
pip install voltage-verify. Release 0.2.0 was current when this list was checked.
GitHub
Dated attestation evidence bundles on GitHub (CC BY 4.0) (opens in a new tab)
Every published bundle, re-verifiable, with the documented limits such as the NVSwitch fabric.
GitHub
tdx-guest-probe on GitHub: what a TDX guest can and cannot observe (opens in a new tab)
One script and its real output from a VoltageGPU guest, with the limit it cannot see past.
DEV Community
Our DEV article: getting both attestation proofs with nothing but an API key (opens in a new tab)
What was missing from our own API on 17 September 2026, and what we fixed.
Each address was opened on ; dead links and pages that do not name us are left out. NVIDIA, Intel and the Confidential Computing Consortium are not partners, customers, sponsors or endorsers of VoltageGPU: the technical threads above are public reports we opened or took part in. Seen us somewhere we missed? Write to press@voltagegpu.com.
06Go further
The pages behind the facts
- Confidential VMsThe product: tiers, limits and how to deploy.
- The two proofs, generated by the tenantIntel TDX quote and NVIDIA attestation, with the raw files.
- Attested SKUs and their evidence (JSON)What is verified, on which machine, since when.
- Why the NVSwitch fabric is not verifiedThree runs on the 8x H100 node, one correction, full logs.
- voltage-verify: files, checksums and replay commandsThe open-source verifier, mirrored on our site.
- Live pricingEvery tier, billed per second.
- Confidential VM documentationThe attestation steps, command by command. (opens in a new tab)
- About VoltageGPU and its founderWho runs the company, and how.
07Contact
Press contact
Company
VOLTAGE EI
SIREN 943 808 824 · Solaize, France