From the NVIDIA edition of October 24, 2026
Firebird (Oct 23): US export licence for 45,000 NVIDIA Grace Blackwell GPUs in Kazakhstan; at least 10% of capacity pledged to Kazakh companies and government agencies
Firebird Inc.; Kazakhtelecom; Government of Kazakhstan (Ministry of Artificial Intelligence and Digital Development); NVIDIA · National government sovereign AI infrastructure (public-private AI factory) · Kazakhstan (Data Center Valley, Ekibastuz region; Astana)
- Publisher
- Firebird newsroom
- Original publication
- 2026-10-23 (release dateline; the page header shows October 24, 2026)
- Source retrieved
- 2026-10-25
- Event date
- 2026-10-23
What happened
Firebird says it has secured a U.S. export licence for 45,000 NVIDIA Grace Blackwell GPUs for Phase 1 of its Kazakhstan AI Factory, under construction with Kazakhtelecom at Data Center Valley and targeting commercial launch in Q3 2027; a planned Phase 2 would take total potential capacity to about 500 MW. The release also describes an estimated $5 billion investment and a cluster of up to 109 MW with more than 45,000 GB300 GPUs and over 130 permanent jobs. Firebird commits at least 10% of total GPU capacity to Kazakh companies and government agencies, plans Firebird Labs Kazakhstan (equity, compute access, scholarships, research partnerships with alem.ai) and Kazakh-language AI work. The expansion builds on cooperation agreements signed with the government in June 2026.
Why it matters
International public-sector signal, not a U.S. SLED case: it is a concrete example of a government securing a reserved share of a privately operated NVIDIA AI factory in return for state cooperation. The 10% public/local allocation is the transferable element for U.S. states weighing shared or reserved capacity in partner-built facilities; export-licensing, energy and governance context does not transfer directly.
Evidence and measured results
Company press release (classed vendor-claim) quoting Kazakhstan's Deputy Prime Minister and Minister of AI and Digital Development, NVIDIA's CEO and Firebird's CEO. No contract text, allocation mechanism, pricing or delivery schedule beyond the Q3 2027 target is published.
Limitations and uncertainty
Self-reported announcement before any Phase 1 operation. Internal figures differ (45,000 GPUs and up to 109 MW and $5 billion for Phase 1; ~500 MW total potential with Phase 2); the June 2026 package has been reported elsewhere as a $10 billion, 100,000-GPU plan. 'Total potential' capacity is not committed capacity.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-25; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Governments want sovereign or local access to large NVIDIA capacity without financing and operating an AI factory themselves.
- Stakeholders
- National or state AI/digital ministry
- State CIO or research-computing consortium
- Telecom or data-center partner
- Public universities and agencies as tenants
- Finance and procurement officials
- Discovery questions
- How would a reserved public share be allocated, priced and prioritised?
- What audit rights would the public body have over use of its share?
- Which phase and hardware generation is being committed, and by when?
- What happens to the reserved share if the provider is sold or exits?
- Value hypothesis
- Interpretation: a reserved share of partner-built capacity could give public users access to frontier GPUs sooner than a public build, if allocation and pricing terms are enforceable.
- Potential engagement
- Advisory review of reserved-capacity and shared-tenancy terms for a state or university system considering a partner-operated AI factory.
- Unsupported claims
- That Phase 1 capacity exists or will launch on schedule
- That the 10% share is free or discounted to government users
- That this model produces measurable public benefit
Pre-sales engineering
Role takeaway
- Fit
- Public bodies needing occasional access to large-scale training/inference capacity they cannot host.
- Architecture and integration
- Multi-tenant access to a provider-run GB300-class cluster at a national data-center zone, with public tenants connecting through provider interfaces.
- Prerequisites
- Defined public tenant identity and access model
- Data-residency and classification rules for workloads
- Network connectivity to the provider site
- Constraints
- Export-licence conditions on GPU use are not described
- Phase 1 launch targeted Q3 2027
- Hardware generation stated inconsistently in the release
- Security
- Require tenant isolation, logging and data-residency commitments for government workloads; none are stated in the release.
- Proof of value
- Confirm delivered GPU count and generation against the 45,000 figure at launch
- Measure the public share's actual availability and queue times
- Test isolation between government and commercial tenants
Delivery
Role takeaway
- Work
- Set up governance to allocate and monitor a reserved public share of a partner-operated cluster.
- Dependencies
- Phase 1 construction and commissioning
- Final allocation agreement with the provider
- Power and connectivity at Data Center Valley
- Ownership
- The government AI ministry or a designated allocation body, with the provider as operator.
- Skills and adoption
- Public researchers and agencies would need onboarding to use large shared GPU clusters.
- Governance checkpoints
- Publish allocation rules for the reserved share
- Agree audit and reporting rights
- Review export-licence obligations before onboarding tenants
- Acceptance criteria
- Proposed: public share measurable and reported monthly as at least 10% of delivered GPU hours
- Proposed: documented isolation tests passed before government data is processed
- Risks
- Schedule slip beyond Q3 2027
- Allocation captured by a few users
- Conflicting capacity and investment figures obscure commitments
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Phase 1 is a large Grace Blackwell (GB300) deployment co-built with the national telecom operator at a designated data-center zone; a planned Phase 2 would bring total potential capacity to about 500 MW. Hardware naming is inconsistent within the release (Grace Blackwell/GB300 in the body, Vera Rubin in the NVIDIA CEO quote), so the generation and phasing should be confirmed.
Governance
Who approves, reviews and stays accountable for outcomes?
A reserved-capacity pledge (at least 10%) gives the state a stake but no stated mechanism for allocation, pricing, priority or audit; export-licence conditions are not described.
Security and privacy
What data, permissions and controls need testing?
The release states no data-residency, access-control or security commitments for government tenants; the NVIDIA CEO is quoted on sovereign computing, not on controls.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Plans scholarships, specialist training and joint research and 130+ permanent jobs; no targets, budgets or eligibility rules are given.
Procurement
What should contracts, pricing and exit terms secure?
Shows a government-to-provider agreement model (cooperation agreements, then provider investment and reserved capacity) rather than a public purchase; terms that a public buyer would need (service levels, price, exit) are not disclosed.
Operating model
Which teams own the service once it runs?
The state relies on a foreign-headquartered provider to operate the capacity and allocate the public share, which places allocation and service governance outside government operations.
Publication history
- 2026-10-24NVIDIA · Issue 495 resources
Stable resource ID: firebird-kazakhstan-export-license-45000-grace-blackwell-gpus-20261023