NVIDIA

nvidia.com · Compute, Inference, Storage · 7 records · JSON

NVIDIA build Inferencetrialemailno cardDrequest-rate
NVIDIA Academic Grant Computefails: identityinstitutionno cardDwrite-gated
NVIDIA DGX Cloud Computefails: identityinstitutionno cardphoneDwrite-gated
NVIDIA DSX Air Computefails: identityinstitutionno cardphoneRretention-clamped
NVIDIA Inception Computefails: identityinstitutionno cardDwrite-gated
NVIDIA LaunchPad Computedeademailno cardDwrite-gated
nvidia ngc catalog Storagedeademail?card?R

NVIDIA build Inference

trialemailno cardDrequest-rate

build.nvidia.com/ · inference endpoint, trial tier · inference/nvidia-build-trial · JSON · compare

Free tier	free trial tier with no credit card required, 'Up to 40 rpm' and '10,000 requests per day', read on 2026-10-07. The same URL re-fetched on 2026-10-22 no longer carries the rate string in the served text, so the figures are quoted at their 2026-10-07 reading unit requests; metered on request-count
Price	none stated
Per-object cap	not applicable: an inference endpoint stores nothing for the caller
Retention	not applicable
Egress	the response is the payload; no separate egress term is published
Identity	email an NVIDIA account; no card required
Limits	request-rate up to 40 requests a minute and 10,000 requests a day per the 2026-10-07 capture of the model page; throttling 'may vary by model and traffic from other users'
Binding	request-rate
Interface	rest
Tooling	any OpenAI-compatible client; curl against POST {base}/chat/completions
Composes with	pairs with an agent only inside a trial
Reliability	no reliability event recorded
Notes	The 10,000 requests a day is the largest published daily request allowance of any free tier found, and it is a trial. The rate was read on 2026-10-07 and the 2026-10-22 re-fetch of the same URL did not carry it in the served text, so the row is documented at 15 days of age and must be re-read before it is planned against. This is the row the topic README's staleness table marks as a trial.
Evidence rung	documented
Checked	2026-10-07 (live); stale on 2026-11-06
Parsed numbers	`$.limits[0].value` limit:request-rate 40.0 requests / minute

Evidence (1)

date	source	what it proved
2026-10-07	build.nvidia.com/deepseek-ai/deepseek-v4.1-flash	'Up to 40 rpm' and '10,000 requests per day' with throttling that 'may vary by model and traffic from other users', captured from the page JSON on 2026-10-07. The 2026-10-22 re-fetch of the same URL did not carry the rate string in the served text

Unknown or withheld values (3)

$.confidence_note withheld
$.free_tier.notes withheld
$.retention.stated not-applicable

NVIDIA Academic Grant Compute

fails: identityinstitutionno cardDwrite-gated

www.nvidia.com/en-us/industries/higher-education-research/ac · GPU cloud · compute/nvidia-academic-grant · JSON · compare

Free tier	'Up to 30,000 NVIDIA H100 80 GB hours or equivalent (max. of eight concurrent GPUs)' plus 'Up to two NVIDIA DGX Spark'. 'GPU hours provided to the PI will expire six months after the award; unused GPU hours will be forfeited.' Eligibility: 'Must be a full-time faculty member at an accredited academic institution that awards research degrees to PhD students.' The vendor page now states 'While our Academic Grant Program is currently not accepting new applications'. unit hours; metered on wall-clock; one-time-credit
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	institution not published
Limits	write-gated a four-page proposal reviewed by an awards panel, gated on full-time faculty status at a PhD-granting institution, so an unaffiliated applicant fails before the proposal is read. source a four-page proposal reviewed by an awards panel, gated on full-time faculty status at a PhD-granting institution, so an unaffiliated applicant fails before the proposal is read.
Binding	write-gated a four-page proposal reviewed by an awards panel, gated on full-time faculty status at a PhD-granting institution, so an unaffiliated applicant fails before the proposal is read.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	The DGX Spark is shipped hardware rather than a hosted slice, so it does not compose with a cloud allowance; the H100 hours are consumed on NVIDIA's side and cannot be redirected to another vendor.
Reliability	no reliability event recorded
Notes	Two failure modes stacked, and the corpus should record both because they decay differently. The permanent one is the faculty gate: postdocs and graduate students must route through a full-time faculty member, and there is no independent-researcher lane. The transient one is the closed portal: the vendor page says not accepting, a third-party portal page says 'Application Halted', and a June 2026 forum thread shows the 30 June round closing early. Worth re-checking each quarter rather than marking dead, because NVIDIA has run this as a rolling quarterly call and an award is USD-equivalent to roughly 30,000 H100-hours. A second variant circulates through university research offices quoting '32,000 A100 40GB hours' and 'one DGX Spark' for other interest areas, which is the same program with a different resource line per call for proposals.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (4)

date	source	what it proved
2026-10-22	dam-cdn.nvd.orangelogic.com/AssetLink/drvk2c5agrv0s0myqncgl6	the AI Inference, Agents, and Systems Software call for proposals: 'Up to 30,000 H100 80 GB hours or equivalent (max. of eight concurrent GPUs)', 'GPU hours provided to the PI will expire six months after the award', 'Full-time faculty at accredited academic institutions that award research degrees to Ph.D. students are eligible. Postdocs and graduate students must work with a full-time faculty member to submit on their behalf', quoted; PDF 4 pages, 200
2026-10-22	www.nvidia.com/en-us/industries/higher-education-research/ac	'While our Academic Grant Program is currently not accepting new applications, we remain committed to supporting the research community', quoted; and the requirements block repeats the full-time faculty condition
2026-10-22	academicgrants.nvidia.com/academicgrantprogram/s/application	the program's own portal page is titled 'Application Halted' and reads 'The NVIDIA Academic Grant Program is not accepting new applications at this time.', per search snippet
2026-06-30	forums.developer.nvidia.com/t/academic-research-grant-portal	an NVIDIA staff reply on the developer forum: 'I just checked with the team, and unfortunately we are currently not accepting new applications', after a submitter found the 30 June 2026 portal closed at 09:50 EDT

More fields (1)

{
 "wave2": {
  "class": "research-credit",
  "free_tier_kind": "one-time-credit",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": "H100 80GB",
   "gpu_count": 8
  }
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

NVIDIA DGX Cloud Compute

fails: identityinstitutionno cardphoneDwrite-gated

docs.nvidia.com/dgx-cloud/lepton/get-started/workspace/index · GPU cloud · compute/nvidia-dgx-cloud · JSON · compare

Free tier	No self-serve free tier. The product's own documentation states 'As DGX Cloud Lepton is still in early access, the workspace needs to be activated after it is created... Contact your TAM to activate your Workspace. TAM will review your request and contact you.' A default Workspace is created only when you create an Organization. NVIDIA separately offers 'up to $100,000 in GPU capacity credits' through DGX Cloud Lepton, restricted to portfolio companies of Accel, Elaia, Partech and Sofinnova Partners. unit none; metered on wall-clock; none-self-serve
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	institution a business relationship is the gate: workspace activation is a manual review by an NVIDIA technical account manager — "Contact your TAM to activate your Workspace. TAM will review your request and contact you." — and the USD 100,000 credit lane is limited to "eligible portfolio companies" of Accel, Elaia, Partech and Sofinnova Partners. No card, phone, hold or identity document is stated, and an aggregator's "no credit card, no approval" claim is contradicted by the vendor documentation
Limits	write-gated workspace activation is a manual review by an NVIDIA technical account manager, and the USD 100,000 credit lane is limited to named VC portfolios. source workspace activation is a manual review by an NVIDIA technical account manager, and the USD 100,000 credit lane is limited to named VC portfolios.
Binding	write-gated workspace activation is a manual review by an NVIDIA technical account manager, and the USD 100,000 credit lane is limited to named VC portfolios.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	The credits compose with nothing for an individual: they are portfolio-gated. A developer who does hold an NVIDIA cloud partner relationship can rent from the Lepton marketplace, which is a normal paid GPU rental.
Reliability	no reliability event recorded
Notes	The aggregator row is the trap was asked to check. It reads as an instant self-serve free tier and it appears to be a restatement of the pre-acquisition Lepton AI Basic plan. The current vendor documentation replaces 'instant' with a TAM review, so the finding is the gate, not the 48 CPU figure.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21
Parsed numbers	`$.limits[0].value` limit:write-gated 100000.0 credit

Evidence (3)

date	source	what it proved
2026-10-22	docs.nvidia.com/dgx-cloud/lepton/get-started/workspace/index	'Contact your TAM to activate your Workspace. TAM will review your request and contact you.', quoted from the workspace activation section
2026-10-22	investor.nvidia.com/news/press-release-details/2025/NVIDIA-D	'up to $100,000 in GPU capacity credits' offered 'to eligible portfolio companies' of Accel, Elaia, Partech and Sofinnova Partners
2026-10-22	aicredits.dev/submissions/45-lepton-ai-nvidia-dgx-cloud-lept	a credit aggregator claims a free Basic plan with 'up to 48 CPUs + 2 GPUs concurrently' and 'Instant Signup and get credits instantly - no credit card, no approval'; the claim is not supported by NVIDIA's documentation, which requires TAM activation

More fields (1)

{
 "wave2": {
  "class": "refutation",
  "free_tier_kind": "none-self-serve",
  "shape": null
 }
}

Unknown or withheld values (9)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

NVIDIA DSX Air Compute

fails: identityinstitutionno cardphoneRretention-clamped

docs.nvidia.com/networking-ethernet-software/nvidia-air-v2/A · cloud VM · compute/nvidia-dsx-air · JSON · compare

Free tier	as indexed from the account-setup page: a free trial with 60 concurrent vCPUs, 60 GiB concurrent memory, 10,000 compute-hour credits and a 1 year duration, restricted to NGC organization owners, and the page states that a business email address is required and that personal addresses such as gmail.com are not accepted. unit hours; metered on wall-clock; trial-credit
Price	none stated
Per-object cap	not applicable
Retention	the trial must be started by an NGC organization owner behind a non-public domain mailbox, and a trial machine sleep date can be pushed at most 72 hours at a time, so 10,000 compute hours are reachable only by an operator who re-arms a calendar every three days.
Egress	not stated
Identity	institution institution, in the form of a business mailbox: the account-setup page states a business email address is required and that personal addresses such as gmail.com are not accepted, and the trial must be started by an NGC organization owner. A forum report adds that public-suffix addresses such as QQ, Gmail and Outlook are rejected and that a domain mailbox was accepted at signup. No card, hold, phone number or ID document is stated in any evidence row: unknown. The page was read through the search index, not first-party.
Limits	retention-clamped the trial must be started by an NGC organization owner behind a non-public domain mailbox, and a trial machine sleep date can be pushed at most 72 hours at a time, so 10,000 compute hours are reachable only by an operator who re-arms a calendar every three days. source the trial must be started by an NGC organization owner behind a non-public domain mailbox, and a trial machine sleep date can be pushed at most 72 hours at a time, so 10,000 compute hours are reachable only by an operator who re-arms a calendar every three days.
Binding	retention-clamped the trial must be started by an NGC organization owner behind a non-public domain mailbox, and a trial machine sleep date can be pushed at most 72 hours at a time, so 10,000 compute hours are reachable only by an operator who re-arms a calendar every three days.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	60 vCPU and 60 GiB concurrent with 10,000 compute hours is the largest single free CPU allowance found, and it composes with a domain: a custom-domain mailbox is the price of entry, which the domains topic already prices.
Reliability	no reliability event recorded
Notes	Two independent forum reports disagree on whether the sleep window is 72 hours or 24. DSX Air is a data-centre simulation product, so the machines are network simulators and not a general build host; that is a tooling note and not a rejection. The business-email gate is the finding: this is another institution-shaped free allowance with a five-figure compute-hour number beside it.
Evidence rung	reported
Checked	2026-10-22 (live); stale on 2026-11-05
Parsed numbers	`$.limits[0].value` limit:retention-clamped 72.0 hours

Evidence (4)

date	source	what it proved
2026-10-22	docs.nvidia.com/networking-ethernet-software/nvidia-air-v2/A	the search index of this page carries the trial table of 60 vCPU, 60 GiB, 10,000 compute hours, 1 year, and the sentence that personal email addresses such as gmail.com are not accepted; a direct broker read of the URL returned the Air user-guide shell instead of the account page, so the table itself was not read first-party
2026-05-21	lewuxian.com/17288/	reported, translated: a domain mailbox is required and public-suffix addresses such as QQ, Gmail and Outlook are rejected; the machine needs re-arming every 3 days; 10,000 hours is about 416 days of one vCPU
2026-05-19	www.nodeloc.com/t/topic/85975	forum thread titled around using Nvidia's 60c60g machine for free for a year confirms the shape and that a domain mailbox was accepted at signup
2026-05-20	www.nodeseek.com/post-738507-1	counter-report, translated: no stock was available; the token expires immediately; login and machines must be raced for; and a later reply says the machine now sleeps automatically every 24 hours, which contradicts the 72 hour figure

More fields (1)

{
 "wave2": {
  "class": "vendor-trial",
  "free_tier_kind": "trial-credit",
  "shape": {
   "vcpu": 60,
   "ram_gib": 60,
   "gpu_class": null,
   "gpu_count": 0
  }
 }
}

Unknown or withheld values (8)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.tooling not-stated

NVIDIA Inception Compute

fails: identityinstitutionno cardDwrite-gated

programs.nvidia.com/phoenix/application · cloud VM · compute/nvidia-inception · JSON · compare

Free tier	A free program with no application fees, deadlines or cohorts. Members get developer tools and training, preferred pricing on NVIDIA hardware and software, and 'access to free cloud credits from NVIDIA and partners'. Minimum requirements state the company must be officially incorporated and less than 10 years old. Solo founders with no other employees may apply. unit none; metered on wall-clock; membership-perks
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	institution Explicitly requires the company to be 'officially incorporated', a working website, a business email on a non-generic domain (gmail.com and info@ aliases refused), and two named contacts, with a carve-out that solo founders may apply. The incorporation requirement is the binding term and is not waived for solo founders.
Limits	write-gated The incorporation requirement fails first, so the member-only cloud credits and preferred hardware pricing are unreachable without a legal entity; the credit amount itself is not published. source The incorporation requirement fails first, so the member-only cloud credits and preferred hardware pricing are unreachable without a legal entity; the credit amount itself is not published.
Binding	write-gated The incorporation requirement fails first, so the member-only cloud credits and preferred hardware pricing are unreachable without a legal entity; the credit amount itself is not published.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	Inception is the front door to the partner credit offers rather than a credit in itself, and the published partner credit found here, Lambda, is USD 1,000 and also demands incorporation. The chain is short and every link needs a company.
Reliability	no reliability event recorded
Notes	Membership is free and the partner credits are the actual value, but no dollar figure is published for the NVIDIA-side cloud credits at all, so the amount is null rather than guessed. This is the refutation of the GPU-vendor startup path: the program that gates every partner credit requires a legal entity.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (3)

date	source	what it proved
2026-10-22	programs.nvidia.com/phoenix/application	qualification requires the company to be 'officially incorporated', less than 10 years old, with a working website and a non-generic business email, quoted
2026-10-22	programs.nvidia.com/phoenix/application	'Solo founders with no other employees may also apply' alongside the incorporation requirement, quoted
2026-10-22	www.nvidia.com/en-us/startups/	benefits include 'Access free cloud credits from NVIDIA and partners', with no amount, tier or expiry published, quoted

More fields (1)

{
 "wave2": {
  "class": "startup-credit",
  "free_tier_kind": "membership-perks",
  "shape": null
 }
}

Unknown or withheld values (9)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

NVIDIA LaunchPad Compute

deademailno cardDwrite-gated

www.nvidia.com/en-us/launchpad/ · cloud VM · compute/nvidia-launchpad · JSON · compare

Free tier	Withdrawn. The 2024-02-29 capture states 'NVIDIA LaunchPad provides free access to enterprise NVIDIA hardware and software through an internet browser' and 'Try up to $20,000 (cloud-equivalent) of accelerated compute infrastructure for free'. On 2026-10-22 the URL redirects to build.nvidia.com/explore/discover, an NVIDIA moderator wrote on 2025-11-24 that 'the lab has been discontinued and not just redirect folks over to build.nvidia.com', and the lab host dx.nvidialaunchpad.com now serves 'Your environment is no longer active'. unit none; metered on wall-clock; guided-lab-access
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	email not published
Limits	write-gated access was never automatic: each lab needed a submitted application and a manual approval before the environment was provisioned. source access was never automatic: each lab needed a submitted application and a manual approval before the environment was provisioned.
Binding	write-gated access was never automatic: each lab needed a submitted application and a manual approval before the environment was provisioned.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	Nothing chains now. The successor is build.nvidia.com, whose free tier is a request-rate allowance on hosted inference and carries no accelerator of its own.
Reliability	no reliability event recorded
Notes	This is the refutation exists to record: the largest published number in the accelerator-programme market, USD 20,000 of cloud-equivalent capacity, belonged to a programme NVIDIA has closed. Two candidates were considered before the verdict: (a) the labs moved to a new URL and are still open, refuted by the staff reply naming discontinuation; (b) the page is A/B tested and reachable on another path, refuted because dx.nvidialaunchpad.com itself reports the environment inactive. What would reopen it: a live lab catalog under a new NVIDIA URL with an application form.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (4)

date	source	what it proved
2026-10-22	web.archive.org/web/20240229143632/https://www.nvidia.com/en	archived 2024-02-29 capture of the LaunchPad page carrying 'Try up to $20,000 (cloud-equivalent) of accelerated compute infrastructure for free' and the four-step 'Choose a Lab / Fill in the Application Form / Wait for Approval / Take Additional Labs' workflow
2026-10-22	forums.developer.nvidia.com/t/morpheus-ai-and-launchpad/3524	NVIDIA staff reply dated Nov 24 2025: 'The Morpheus product page needs to be updated to reflect that the lab has been discontinued and not just redirect folks over to build.nvidia.com.'
2026-10-22	www.nvidia.com/en-us/launchpad/	broker fetch returned 200 with x-web-final-url https://build.nvidia.com/explore/discover after 2 redirects, so the LaunchPad URL no longer serves a LaunchPad page
2026-10-22	dx.nvidialaunchpad.com/	the lab host serves 'NVIDIA Developer Experience / Your environment is no longer active'

More fields (1)

{
 "wave2": {
  "class": "refutation",
  "free_tier_kind": "guided-lab-access",
  "shape": null
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

nvidia ngc catalog Storage

deademail?card?R

catalog.ngc.nvidia.com/ · rejected-provider · storage/nvidia-ngc-catalog · JSON · compare

Free tier	Curated by NVIDIA, not user-writable: containers, models and Helm charts published by NVIDIA and partners. api.ngc.nvidia.com is an authenticated pull path (404/401 unauthenticated), and the unit none; metered on none
Price	none stated
Per-object cap	not recorded
Retention	not recorded
Egress	not recorded
Identity	email (placeholder) not published
Binding	none no constraint of a named kind was recorded; the verdict rests on the quota sentence rather than on a named limit
Interface	unknown
Tooling	not recorded
Composes with	not recorded
Reliability	Curated by NVIDIA, not user-writable: containers, models and Helm charts published by NVIDIA and partners. api.ngc.nvidia.com is an authenticated pull path (404/401 unauthenticated), and the
Notes	Curated by NVIDIA, not user-writable: containers, models and Helm charts published by NVIDIA and partners. api.ngc.nvidia.com is an authenticated pull path (404/401 unauthenticated), and the
Evidence rung	reported
Checked	2026-10-21 (inherited); stale on 2026-11-04

Evidence (2)

date	source	what it proved
2026-10-21	catalog.ngc.nvidia.com/	Curated by NVIDIA, not user-writable: containers, models and Helm charts published by NVIDIA and partners. api.ngc.nvidia.com is an authenticated pull path (404/401 unauthenticated), and the
2026-10-21	api.ngc.nvidia.com/v2/models/nvidia/clip/versions	cited as the basis of the rejection: Curated by NVIDIA, not user-writable: containers, models and Helm charts published by NVIDIA and partners. api.ngc.nvidia.com is an authenticated pull path (404/401 unauthenticated), and the

More fields (1)

{
 "dead_end": {
  "failure_class": "dead-service",
  "why": "Curated by NVIDIA, not user-writable: containers, models and Helm charts published by NVIDIA and partners. api.ngc.nvidia.com is an authenticated pull path (404/401 unauthenticated), and the",
  "retry_when": null,
  "first_recorded": "2026-10-21"
 }
}

Unknown or withheld values (15)

$.chaining not-recorded
$.confidence_note withheld
$.egress.terms not-recorded
$.extras.dead_end.retry_when process
$.free_tier.notes process
$.identity_cost.detail defaulted
$.identity_cost.kind identity-defaulted
$.interface interface-unknown
$.interface_detail withheld
$.limits_note process
$.per_object_cap.notes not-recorded
$.per_object_cap.stated not-recorded
$.retention.notes process
$.retention.stated not-recorded
$.tooling not-recorded