{ "schema_version": "pct-article-edition-1.0", "publication_date": "2026-11-06", "export_scope": "articles_only", "source_edition_id": null, "rights": "PCT reserves rights in original editorial content; third-party rights retained. No general source-data reuse license.", "articles": [ { "id": "pct-2026-11-06-company-569566e5b2", "content_type": "article", "desk": "company", "title": "Nebius wants to own the work around the GPU", "publication_date": "2026-11-06", "observation_date": "2026-11-06", "editorial_voice": "AI-assisted Company Desk", "body": "# Nebius wants to own the work around the GPU\n\n*Company desk | Peterson Compute Trading | November 6, 2026*\n*AI-assisted analysis of public sources. No interviews or independently run performance tests.*\n\nA compute provider earns its place in a training run when the job finishes, survives interruption and produces a usable model at an acceptable cost. Nebius is trying to make those operating details its advantage. Its public financing announcements describe a rapidly expanding infrastructure company. Its customer and product material reveals the other ambition: becoming the environment in which a team repeatedly trains, evaluates and serves models.\n\nThat distinction matters commercially. A buyer can compare quoted GPU rates in minutes. Replacing a reliable training and deployment workflow is harder. PCT's thesis is that Nebius is attempting to earn both kinds of revenue: large contracted capacity sales that support infrastructure investment, and a broader cloud business whose usefulness extends beyond access to a scarce chip. Whether the second becomes a durable advantage depends on evidence that product breadth improves customers' completed work.\n\n## Start with the job, not the campus\n\nPrime Intellect provides a concrete example. Nebius's case study describes a customer that aggregates compute and runs distributed pre-training and reinforcement learning. Model instances generate rollouts on multiple clusters, with the results gathered for centralized training. The account reports coordinated use of European and US Nebius locations and identifies SkyPilot and Managed Soperator as tools in the workflow. It also describes a four-node, 16-GPU Blackwell proof of concept. These are vendor-reported operating details, not PCT test results. [s1]\n\nThe customer's own INTELLECT-3 announcement supplies a narrower independent check. Published November 26, 2025, it describes a 106-billion-parameter mixture-of-experts model built on GLM 4.5 Air and names Nebius and Parasail as inference providers. That supports a real service relationship without establishing that Nebius supplied every stage of training or that its infrastructure outperformed alternatives. [s2]\n\nFor a procurement team, this is a more useful starting point than a generic claim about access to thousands of GPUs. The workflow contains separate tasks with different requirements. Generating candidate answers can offer more placement flexibility than a tightly synchronized training step. Moving results, recovering jobs and coordinating schedulers can consume time even when the accelerator itself is fast.\n\nThose observations are PCT's analysis of the architecture. They suggest the right buyer question: which part of the customer's complete learning cycle becomes faster or less labor-intensive on Nebius? A demonstration of an individual chip cannot settle that question. A repeatable trial should measure time from submitted job to accepted output, including queueing, data preparation, failures and reruns.\n\nThe case study also makes reliability central to the offer. It describes automated checks and recovery, and reports a 12-minute average recovery time for multi-host training. The public page does not supply enough methodology for PCT to generalize that figure across workload sizes or regions. It is a diligence prompt rather than a fleet-wide service guarantee. [s1]\n\n## A second customer buys an endpoint\n\nNebius also sells a more managed path through Token Factory. The November 2025 launch announcement identifies Prosus as a dedicated-endpoint customer and Higgsfield as an on-demand and autoscaling inference customer. Its Prosus testimonial describes workloads reaching 200 billion tokens per day. That is an attributed example in a supplier release, not a current platform-wide throughput measurement. The business purpose is clear: customers can purchase a working inference service while delegating more of the serving infrastructure. [s5]\n\nA February 26, 2026 product announcement makes the control surface tangible. Dedicated Endpoints allow a team to select region, GPU type, GPUs per replica and minimum and maximum replica counts. The same announcement describes deploying custom checkpoints and tracking latency percentiles, throughput, replicas and errors. Inference uses an OpenAI-compatible interface and an endpoint-specific routing key. [s3]\n\nConsider an illustrative document-processing service. It might first compare model quality on a fixed set of documents, then run a load test on a dedicated endpoint. The team would measure acceptable outputs per dollar at its latency target, rather than reward the model that generates the most raw tokens. It would test burst traffic and an update to new weights before committing production traffic. This is an example of how to evaluate the product, not a claim about a named customer.\n\nNebius's published configuration example uses two GPUs per replica, a minimum of two replicas and a maximum of eight. The arithmetic implies four baseline GPUs and sixteen at the configured ceiling. It says nothing by itself about the price, achieved throughput, available regional quota or whether a particular model meets the buyer's quality threshold. [s3]\n\nThe strategic appeal is that deployment becomes part of the ongoing model-development relationship. The competing exposure is that every additional managed function creates an obligation to keep the service dependable. A customer who delegates operational work will expect the supplier to absorb more troubleshooting, compatibility and performance responsibility.\n\n## Flexible compute still requires discipline\n\nThe self-managed end of the catalog illustrates a different allocation of risk. Nebius's public preemptible-VM page says prices may change as often as every fifteen minutes. A customer's maximum price caps the GPU-hour rate, not the total bill. On interruption, persistent-volume data is retained but local SSD data is lost; standalone VMs do not restart automatically. Retained resources may continue billing after the VM stops. These are current public product descriptions observed November 6, not an authenticated capacity reservation. [s4]\n\nFor an interruptible evaluation or batch job, PCT would compare the saving with checkpoint overhead, restart labor and the expected cost of losing progress. The result may favor preemptible capacity, but that conclusion must come from the workload. A deadline-sensitive task can be a poor fit even if the advertised rate is attractive.\n\nThis is also a constraint on simplified compute-market comparisons. An hourly price needs the region, configuration, interruption rules and observation time beside it. A list of supported accelerator models does not establish stock. The service boundary matters as much as the hardware label when comparing a VM, a managed endpoint and a multiyear dedicated deployment.\n\n## Revenue is scaling; the capital bill is larger\n\nThe latest quarterly actuals available for this profile are for June, not September. Nebius has scheduled its third-quarter results for November 12. The group also contains Avride and TripleTen and holds investments including ClickHouse and Toloka, so consolidated figures cannot be treated as a pure GPU-cloud income statement. [s12]\n\nThe August 12 release reports second-quarter revenue of $582.3 million, up 454% year over year, and adjusted EBITDA of $236.2 million. Continuing operations nevertheless recorded a $190.4 million net loss. Purchases of property, equipment and intangible assets were $5.657 billion for the quarter, against operating cash inflow of $2.246 billion. Cash and cash equivalents were $8.042 billion at June 30, with restricted cash separately reported. [s6]\n\nThe basic interpretation is that operating expansion and financing requirements must be evaluated together. Positive adjusted EBITDA does not fund the entire investment program. Equally, a large capital outflow during construction does not by itself establish that the underlying service economics are poor. The question is how efficiently that spending becomes reliable, paid-for capacity over its useful economic life.\n\nManagement's second-quarter material gives additional context: it reports a 50% AI-cloud adjusted EBITDA margin, more than threefold quarterly growth in Token Factory production inference workloads, and a year-end contracted-power target of 5 GW. These are different measures with different evidentiary value. The workload growth figure lacks an absolute baseline in the cited summary. Contracted power describes an input to expansion, not a count of commissioned GPUs or billable tokens. [s7]\n\n## Two customer markets, several financing clocks\n\nThe large-contract side can support the broader cloud effort. In July, Nebius announced approximately $775 million of secured financing backed by deployed GPUs and cash flows from an investment-grade customer agreement already in service. The company said those cash flows and the facility together covered more than the underlying deployment's capital expenditure, freeing financing for further cloud expansion. The facility matures in October 2030 and carries SOFR plus 2.50%. This is the company's description of one financing structure, not proof that every new project can reproduce it. [s8]\n\nThe Meta agreement shows a related commercial design. Announced in March, it includes $12 billion of dedicated capacity scheduled to start delivery in early 2027 and up to $15 billion of purchases of additional available capacity over five years. Nebius intends first to sell the latter capacity to other cloud customers, with Meta buying what remains. The stated maximum is therefore not interchangeable with immediately recognized revenue or a single undifferentiated dedicated order. [s10]\n\nPCT's interpretation is that the arrangement may reduce some utilization uncertainty while preserving opportunities to build a wider customer base. It does not eliminate the need to deliver the infrastructure, satisfy contract conditions and maintain a competitive service. The primary announcement does not provide sufficient detail to price all the options embedded in that relationship.\n\nThe financing clocks also differ. The October Citi agreement provides a $3.5 billion senior secured bridge facility maturing October 16, 2027. Its spread starts at 2.75% over Term SOFR and steps up every ninety days to 3.50%, subject to adjustments. The filing describes a minimum-liquidity covenant and security over substantially all borrower assets, with exceptions. It establishes access to the facility, not a verified outstanding draw of the entire amount. [s9]\n\nA separate August transaction closed $5.75 billion of convertible notes: $3.45 billion at 0.50% due 2030 and $2.3 billion at 4.50% due 2034. Concurrent exchanges converted $800 million of older notes into approximately 15.8 million ordinary shares. The low coupon on one tranche should therefore be considered alongside conversion exposure, rather than treated as the complete economic cost of capital. [s11]\n\n## What would establish the advantage\n\nNebius has evidence of real customers, a growing service catalog and substantial financing capacity. Its unresolved challenge is connecting those strengths in a way that withstands less favorable conditions. Hardware generations change, customers can move work, managed services require investment, and construction schedules do not necessarily match funding maturities.\n\nThe most useful next disclosures would separate large dedicated deployments from the broader cloud customer base: revenue mix, retention, concentration and utilization measured consistently. Buyers would benefit from reproducible end-to-end benchmarks, regional availability and service-level terms tied to their actual configurations. Capital providers need the bridge from contracted power to commissioned equipment, customer acceptance and collected cash.\n\nPCT's assessment is that Nebius deserves to be evaluated as an operating platform as well as an infrastructure builder. Its customer workflows provide a plausible reason to choose it beyond chip scarcity. The durability of that reason will be demonstrated in repeated completed work and cash generation, not in the size of the next campus announcement.", "sources": [ { "id": "s1", "title": "Prime Intellect customer story", "url": "https://nebius.com/customer-stories/prime-intellect", "publication_date": null, "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Vendor case study; undated live page; performance claims not independently tested." }, { "id": "s2", "title": "INTELLECT-3 launch", "url": "https://www.primeintellect.ai/blog/intellect-3", "publication_date": "2025-11-26", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Customer primary source confirms Nebius inference relationship." }, { "id": "s3", "title": "Dedicated Endpoints and Custom Weights Hub", "url": "https://nebius.com/blog/posts/dedicated-endpoints-and-custom-weights-hub", "publication_date": "2026-02-26", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Product announcement; example configuration is illustrative." }, { "id": "s4", "title": "Preemptible VMs with spot pricing", "url": "https://nebius.com/preemptible-vms", "publication_date": null, "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Live product description; no authenticated price or stock check." }, { "id": "s5", "title": "Token Factory launch", "url": "https://nebius.com/newsroom/nebius-launches-nebius-token-factory-to-deliver-production-ai-inference-at-scale", "publication_date": "2025-11-05", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Vendor release with attributed customer testimonials." }, { "id": "s6", "title": "Q2 results SEC exhibit", "url": "https://www.sec.gov/Archives/edgar/data/1513845/000110465926094568/tm2622968d1_ex99-1.htm", "publication_date": "2026-08-12", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Unaudited financial statements; quarter ended June30." }, { "id": "s7", "title": "Q2 shareholder material SEC exhibit", "url": "https://www.sec.gov/Archives/edgar/data/1513845/000110465926094568/tm2622968d1_ex99-2.htm", "publication_date": "2026-08-12", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Management metrics and forward capacity guidance." }, { "id": "s8", "title": "July secured financing", "url": "https://nebius.com/newsroom/nebius-raises-775-million-in-first-secured-debt-financing-to-accelerate-global-buildout", "publication_date": "2026-07-17", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Company financing announcement." }, { "id": "s9", "title": "Citi bridge SEC 6-K", "url": "https://www.sec.gov/Archives/edgar/data/1513845/000110465926117978/tm2628078d1_6k.htm", "publication_date": "2026-10-22", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Agreement entered Oct17; filing outlines covenants and collateral." }, { "id": "s10", "title": "Meta infrastructure agreement", "url": "https://nebius.com/newsroom/nebius-signs-new-ai-infrastructure-agreement-with-meta", "publication_date": "2026-03-16", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Contract announcement; deliveries starting early2027 planned." }, { "id": "s11", "title": "Convertible notes closing", "url": "https://nebius.com/newsroom/nebius-group-announces-closing-of-private-offering-of-convertible-senior-notes-with-aggregate-gross-proceeds-of-approximately-5-75-billion", "publication_date": "2026-08-24", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Closed offering and separate exchanges." }, { "id": "s12", "title": "Q3 reporting date", "url": "https://nebius.com/newsroom/nebius-group-announces-date-of-third-quarter-2026-results-and-conference-call", "publication_date": "2026-10-30", "observed_on": "2026-11-06", "access": "Full primary-source HTML successfully opened", "caveat": "Q3 release scheduled November12, after profile cutoff." } ], "facts": [ { "source": "s1", "type": "vendor_case_study", "claim": "Prime Intellect: multi-cluster workflow, Soperator, Blackwell trial; recovery metric is supplier-reported.", "limits": "Undated; no independent benchmark." }, { "source": "s2", "type": "customer_confirmation", "claim": "Customer independently names Nebius as an inference provider.", "limits": "Does not establish exclusive training supply." }, { "source": "s3", "type": "product_capability", "claim": "Endpoint hardware/region/replica controls and custom checkpoints.", "limits": "Configuration example is not a performance or quota guarantee." }, { "source": "s4", "type": "live_product_terms", "claim": "Variable spot price, interruption/storage/restart conditions.", "limits": "No account-level stock or quote obtained." }, { "source": "s5", "type": "attributed_customer_claim", "claim": "Prosus and Higgsfield named inference users.", "limits": "Testimonial scale is not audited platform output." }, { "source": "s6", "type": "reported_financial_actual", "claim": "Q2 revenue $582.3M; adjusted EBITDA $236.2M; continuing-operations loss $190.4M; purchases $5,657.4M.", "limits": "Unaudited group figures, not standalone cloud figures." }, { "source": "s7", "type": "management_metrics_and_guidance", "claim": "Cloud adjusted EBITDA margin 50%; 5GW contracted-power year-end target.", "limits": "Target is not commissioned compute." }, { "source": "s8", "type": "announced_financing", "claim": "$775M secured facility; SOFR+2.50%; October2030 maturity.", "limits": "Company describes one repeatable structure; future repetition not established." }, { "source": "s9", "type": "filed_agreement", "claim": "$3.5B bridge; October16 2027 maturity; stepping spread and collateral.", "limits": "Full draw not verified." }, { "source": "s10", "type": "contract_announcement", "claim": "$12B dedicated and up to $15B additional-capacity arrangement.", "limits": "Delivery scheduled from early2027; maximum value not recognized revenue." }, { "source": "s11", "type": "closed_financing", "claim": "$5.75B new convertible notes plus separate old-note exchanges.", "limits": "Coupon alone omits conversion economics." }, { "source": "s12", "type": "scheduled_disclosure", "claim": "Q3 results scheduled November12.", "limits": "Not available at article cutoff." } ], "inferences": [ "Workflow integration may create differentiation beyond GPU scarcity; durable retention or premium has not been established.", "Large customer contracts may support funding of broader cloud expansion, with execution and counterparty concentration exposures.", "Managed products shift operational responsibility to Nebius; product breadth is not independently proven profitability.", "Spot-price savings require evaluation after restart, checkpoint, storage and deadline costs." ], "unknowns": [ "Current quarter results await November12 release.", "Fleet-wide utilization, regional instant capacity and actual account quota not checked.", "Customer-level revenue concentration and cloud-versus-dedicated profitability not established.", "No independent validation of vendor performance metrics.", "Current bridge draw and full covenant headroom not established.", "Full Meta contract options/conditions unavailable in reviewed announcement." ], "correction_history": [], "license_status": "Original PCT editorial text; public license not selected. Source rights remain separate.", "public_url": null, "desk_extensions": { "title": "Nebius wants to own the work around the GPU", "company": "Nebius", "date": "2026-11-06", "word_count": 1783, "estimated_read_minutes": "8–10", "byline": "PCT Company Desk | AI-assisted public-source analysis", "dedup_note": "Distinct company from SF Compute/CoreWeave/Crusoe. Business/customer-workflow profile rather than repeating Nov4 AIB–Nebius deal narrative.", "scope": "Research and local draft only; no public or site changes." }, "public_projection": { "version": "1.0.0", "date": "2026-11-10", "changes": [ "Internal workspace artifact paths omitted where present. Historical license/status fields retain original provenance; current public reuse policy is stated separately." ], "rights": "PCT reserves rights in original editorial content; third-party rights retained. No general source-data reuse license." } }, { "id": "pct-2026-11-06-deals-56a05d0883", "content_type": "article", "desk": "deals", "title": "Riot’s $28 billion Anthropic lease is a land-and-power deal with a financing bill still ahead", "publication_date": "2026-11-06", "observation_date": "2026-11-06", "editorial_voice": "AI-assisted Deals Desk", "body": "# Riot’s $28 billion Anthropic lease is a land-and-power deal with a financing bill still ahead\n\n*Deals Desk | November 6, 2026 | Analysis*\n\nRiot has secured a consequential commitment at Corsicana, Texas. It has not delivered a new fleet of GPUs. The distinction is the story: an effective lease can move a project out of speculative development while leaving substantial construction, financing and compute-deployment work unfinished.\n\nRiot’s November 4 earnings release describes a roughly $28 billion, 15-year powered-land lease covering 910 MW of critical IT capacity. Its quarterly SEC filing identifies the tenant as Anthropic PBC. Signed October 28, the lease became binding November 4 after board and technical-specification approvals. That is a completed contractual milestone, not a statement that the campus is ready for AI workloads. [Earnings announcement](https://www.riotplatforms.com/riot-platforms-reports-third-quarter-2026-financial-results-and-strategic-highlights/); [SEC filing, Note 3 and Item 5](https://www.sec.gov/Archives/edgar/data/1167419/000110465926122417/riot-20260930x10q.htm).\n\n## What changes hands\n\nRiot will deliver land, power infrastructure and powered pads. The tenant will procure, own and operate the modular compute deployment, with substantially all operating costs on its side. Riot estimates total capital requirements of about $5.5 billion: $2 million per IT MW for site and electrical works, plus $4 million per IT MW reimbursed to the tenant after delivery and rent commencement. [Riot announcement](https://www.riotplatforms.com/riot-platforms-reports-third-quarter-2026-financial-results-and-strategic-highlights/).\n\nThis allocation matters more than the headline’s size. It separates the infrastructure landlord’s task from the customer’s technology choices. A powered pad does not specify how many accelerators will sit on it, which generation they will be, what networking will connect them or how efficiently the resulting system will run. None of those missing quantities should be reverse-engineered from the megawatt number.\n\nFor a compute buyer, the transaction is therefore evidence of a large reservation of future physical infrastructure, not a new offer of immediately rentable GPU-hours. For an infrastructure competitor, it is evidence of an alternative product: providing a site and electrical backbone while the customer takes responsibility for the computing installation.\n\n## A rent schedule, not a spot price\n\nThe SEC filing specifies $130 per kW-month for delivered blocks, with 3% annual escalation. The planned rollout comprises 400 MW, 200 MW and 310 MW; the first block is targeted for June 28, 2027 and the final phase for July 30, 2028. [SEC filing, Note 3](https://www.sec.gov/Archives/edgar/data/1167419/000110465926122417/riot-20260930x10q.htm).\n\nAs a simple illustration, 910,000 kW multiplied by $130 would produce $118.3 million of monthly base rent at the starting rate if all capacity were simultaneously billable at that rate. That is arithmetic, not a forecast of any particular month. Staggered delivery, escalation and contractual adjustments make the actual cash-flow path different. Nor is the result an electricity tariff: the customer’s electricity costs sit outside base rent.\n\nA GPU-hour comparison would be still less defensible. It would mix a real-estate-and-power-infrastructure payment with a service price that normally also embeds accelerator hardware, networking, software, support and utilization assumptions. The useful price unit here is the contracted kilowatt-month, with its scope preserved.\n\n## The financing fallback is economically important\n\nRiot expects non-recourse project debt to fund the tenant reimbursements. That financing is an expectation, not a reported closing. The filing provides a fallback: if acceptable financing cannot be obtained after commercially reasonable efforts, reimbursement becomes rent credits amortizing the amount with 12% annual interest over the remaining initial term. [Riot announcement](https://www.riotplatforms.com/riot-platforms-reports-third-quarter-2026-financial-results-and-strategic-highlights/); [SEC filing, Note 3](https://www.sec.gov/Archives/edgar/data/1167419/000110465926122417/riot-20260930x10q.htm).\n\nThis is a meaningful distinction between financing resilience and financing cost. A fallback can keep the parties from having to abandon the commercial structure merely because a particular debt package is unavailable. It does not make the capital free. In the alternative structure, the landlord collects less cash rent because the customer recovers its funded amount through credits.\n\nThat means the lease’s gross value cannot be treated as cash available to Riot today, or as a substitute for a financing model. A lender or shareholder still needs the timing of expenditures, the final debt terms, the timing of rent and the amount of any credits. The customer’s willingness to fund an alternative also does not establish that it has already done so.\n\n## Delivery risk remains contractual risk\n\nLate delivery can trigger escalating rent credits after 60 days. A phase still undelivered after 270 days can become terminable, subject to notice, cure and extensions. The filing says the full lease is intended for the year-end 10-K, with some material redacted. [SEC filing, Item 5](https://www.sec.gov/Archives/edgar/data/1167419/000110465926122417/riot-20260930x10q.htm).\n\nThese protections explain why “binding” must not become shorthand for “certain revenue.” There is a difference between a tenant retaining the ability to walk away from an unfinished negotiation and having specified remedies under an effective contract. This deal has crossed the first hurdle. The second remains central to its economics.\n\nThe calendar also offers a practical way to judge progress. Construction updates should be tested against actual blocks and the promised delivery sequence. A power announcement alone does not establish acceptance of the pads. Acceptance of pads does not establish installation of compute. Installation does not establish workload performance. Each is a different evidentiary milestone.\n\n## What the record supports today\n\nThe strongest conclusion is that Riot has converted a large future development into a binding customer relationship with disclosed base-rent economics and a defined allocation of responsibilities. It has not publicly demonstrated completion of this 910 MW deployment, closure of the expected project financing or an executable GPU-rental price.\n\nThe next documents worth reading are correspondingly specific: financing terms, block-level delivery and commencement disclosures, and the full lease when filed. Until those arrive, the $28 billion figure describes a long-duration commercial commitment whose value depends on performance. The deal is substantial precisely because the remaining obligations are substantial, too.\n", "sources": [ { "id": "D1", "type": "primary SEC 10-Q", "url": "https://www.sec.gov/Archives/edgar/data/1167419/000110465926122417/riot-20260930x10q.htm", "publication_date": "2026-11-04", "sections": [ "Note 3 Data Center Operations", "Item 5 Other Information" ], "access": "Opened directly; full filing available" }, { "id": "D2", "type": "primary issuer earnings release", "url": "https://www.riotplatforms.com/riot-platforms-reports-third-quarter-2026-financial-results-and-strategic-highlights/", "publication_date": "2026-11-04", "access": "Opened directly" } ], "facts": [ "Counterparty is named Anthropic PBC in primary SEC filing, although release uses Frontier AI Lab.", "Binding effective date is distinct from signature date.", "Base rent is a land/infrastructure price with electricity and compute responsibilities separate." ], "inferences": [ "Financing fallback reduces dependence on a particular external debt closing, but changes landlord cash rent economics.", "A customer commitment reduces speculative leasing risk without removing construction and performance risk." ], "unknowns": [ "Final Corsicana debt size, lenders, terms and closing", "Full lease text and redacted information", "GPU architecture, count and usable workload capacity", "Actual future block delivery, acceptance and rent commencement", "Whether tenant reimbursements ultimately use debt or rent credits" ], "correction_history": [], "license_status": "Original PCT editorial text; public license not selected. Source rights remain separate.", "public_url": null, "desk_extensions": { "edition_date": "2026-11-06", "section": "Deals Desk", "title": "Riot’s $28 billion Anthropic lease is a land-and-power deal with a financing bill still ahead", "word_count": 925, "estimated_read_minutes": 5, "structured_record": { "seller": "Riot Corsicana, LLC (Riot Platforms subsidiary)", "buyer": "Anthropic PBC", "asset": "Powered land, powered pads and related infrastructure, Corsicana Texas", "signed_date": "2026-10-28", "effective_date": "2026-11-04", "effectiveness_conditions": "Board approval and third-party technical-specification approval satisfied, according to 10-Q", "capacity_mw": 910, "capacity_basis": "critical IT load; future delivery, not currently demonstrated operating compute", "headline_contract_value_usd": 28000000000, "headline_value_status": "Issuer approximate initial-term contracted value; not upfront cash or present value", "initial_term": "Starts with first block; ends last day of month containing 15th anniversary of Phase 3 commencement", "options": "Two optional five-year extensions; not exercised commitments", "rent": { "usd_per_kw_month": 130, "annual_escalation_percent": 3, "scope": "Base rent for delivered block capacity; not electricity or GPU-hour price" }, "delivery_targets": [ { "phase": 1, "mw": 400, "first_block": "2027-06-28", "full_phase": "2027-09-11" }, { "phase": 2, "mw": 200, "first_block": "2027-12-24", "full_phase": "2028-01-28" }, { "phase": 3, "mw": 310, "first_block": "2028-07-30", "full_phase": "2028-07-30" } ], "financing_status": "Expected non-recourse project financing; no Corsicana closing established in reviewed sources", "reimbursement_fallback": "Rent credits amortizing qualifying module costs plus 12% annual interest if acceptable financing unavailable after commercially reasonable efforts", "execution_status": "Binding lease; future staged construction/delivery; no complete 910 MW delivery verified", "gpu_model": null, "gpu_count": null, "executable_gpu_hour_price": null }, "calculations": [ { "description": "Illustrative monthly base rent at full capacity and initial rate, not revenue forecast", "formula": "910000 kW * $130/kW-month", "result_usd_month": 118300000 } ], "dedup": { "checked": "Local PCT JSON/Markdown/text artifacts through Nov 5, including Nov 4 trades and Nov 5 Deals Desk; prior task notes.", "result": "No Corsicana or 910 MW matches in checked PCT artifacts. XMax rejected because already in Nov 4 trade inventory; Nscale was Nov 5 lead.", "limit": "No claim that every remote edition or automation watch was exhaustively searched; parent should reconcile live watch index. No watch or site changed." }, "editorial_cautions": [ "This is November 4 disclosure analyzed for November 6, not represented as a November 6 signing.", "SEC identifies Anthropic; no anonymous-source inference needed.", "Do not add extension-option value to binding base term.", "Do not label anticipated project financing closed.", "Do not treat estimated NOI or yield on cost as realized return." ] }, "public_projection": { "version": "1.0.0", "date": "2026-11-10", "changes": [ "Internal workspace artifact paths omitted where present. Historical license/status fields retain original provenance; current public reuse policy is stated separately." ], "rights": "PCT reserves rights in original editorial content; third-party rights retained. No general source-data reuse license." } }, { "id": "pct-2026-11-06-formation-0799b6fe5b", "content_type": "article", "desk": "formation", "title": "Airport slots and the missing right to transfer compute", "publication_date": "2026-11-06", "observation_date": "2026-11-06", "editorial_voice": "AI-assisted Formation Desk", "body": "# Airport slots and the missing right to transfer compute\n\nA reservation is not necessarily an asset that its holder can sell. That distinction matters before a secondary market for unused GPU capacity can operate. A buyer can pay for a precisely specified machine and time window yet lack permission to transfer that booking. The practical implication is to classify contractual rights before comparing prices. Airport slots offer a useful historical analogy: scarcity existed before an organized transfer mechanism, but trading depended on rules defining who could hold a slot and how an assignment became valid.\n\n## What the airport market actually created\n\nThe FAA defines an airport slot as authorization for a planned takeoff or landing at a particular airport, day and time. It is distinct from air traffic control clearance. Holding the scheduling right does not itself authorize the aircraft’s immediate movement. This separation between advance allocation and operational execution is the first useful lesson for compute. A reservation, a successful instance launch and a completed workload are different events. [FAA slot definition, updated April 24, 2020](https://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/systemops/perf_analysis/slot_administration/slot_definition).\n\nThe US Department of Transportation’s December 20, 1985 rule allowed purchases, sales and leases of specified airport slots beginning April 1, 1986, subject to restrictions. The rule required transfer documentation and FAA confirmation before a recipient could use the slot. It also explained that the delayed start allowed a tracking system to be established. These were not peripheral administrative details: a buyer needed confidence that the seller’s right existed and that the authority recognized the new holder. [Original final rule, excerpt](https://archives.federalregister.gov/issue_slice/1985/12/20/52168-52190.pdf).\n\nThere is an important historical qualification. The original rule used a 65 percent utilization threshold over two months. The FAA’s later general overview describes an 80 percent requirement and explicitly allows waivers. Neither figure should be transplanted into a GPU contract, nor should the overview be treated as a complete statement of today’s airport-specific exceptions. The relevant institutional idea is that a transfer right can coexist with conditions governing retention and use. [FAA compliance overview, updated February 1, 2021](https://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/systemops/perf_analysis/slot_administration/compliance_oversight).\n\n## The compute contrast\n\nAWS Capacity Blocks provide a concrete counterexample to the assumption that advance reservations are freely tradable. The documentation describes offerings with a start time, Availability Zone and reservation price. It also distinguishes the reservation’s end from the usable operating window: termination begins during the final 30 minutes for instance types or 60 minutes for UltraServers. A procurement system must preserve that distinction when checking deadlines. [AWS Capacity Blocks documentation, accessed November 6, 2026](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/capacity-blocks-how.html).\n\nMore decisively, section 5.5.2 of the AWS Service Terms says Capacity Blocks are nontransferable, cannot be resold, and are nonrefundable. It also prohibits submitting Capacity Block requests through a third party, including proxy purchasing. These terms do not establish the policy for every compute provider. They do establish that a broker cannot presume permission to buy these blocks for onward resale or execute purchases as a proxy. [AWS Service Terms, page last updated November 5, 2026](https://aws.amazon.com/service-terms/).\n\nThe market-design inference is narrow but consequential. A quotation system should identify separately the customer, the permitted purchaser, the holder of the reservation, and the party operating the workload. A provider-approved arrangement might eventually connect those roles differently. An attractive spread alone cannot establish that arrangement. For this product, recommendations and customer-controlled purchasing are a safer starting point than assuming a universal execution adapter.\n\n## A worked decision about flexibility\n\nConsider an entirely hypothetical procurement decision. An experimental workload has a 70 percent chance of being ready on Monday. If ready, it needs a tested configuration for a fixed window. Two suppliers offer technically equivalent capacity, with identical operating costs and performance. Supplier A charges $10,000 immediately, with no cancellation refund. Supplier B charges $11,000 but expressly permits cancellation before Sunday for a $1,000 fee. These are invented contracts, not AWS quotations.\n\nIf readiness is confirmed before the cancellation deadline, expected reservation expenditure is $10,000 under A. Under B it is 70 percent of $11,000 plus 30 percent of $1,000, or $8,000. B’s higher headline price buys an exit right that reduces expected expenditure by $2,000 in this example. The break-even readiness probability is 90 percent: above that, A becomes cheaper on these assumptions.\n\nThis calculation values cancellation, not speculative resale. Suppose someone instead advertises a $4,000 recovery from selling A’s unused booking. That credit belongs in the calculation only after transfer permission, a feasible process and credible buyer demand are established. A prohibited transfer has no legitimate resale value. Even a permitted transfer may fail to find a buyer before the window expires.\n\nThe recommendation should therefore expose its assumptions: equivalent usable capacity, readiness known by the deadline, the stated cancellation fee and no additional migration cost. It must also show the downside: if the workload is ready, B costs $1,000 more. Expected savings do not remove that possible outcome.\n\n## Why the analogy has limits\n\nAirport slots concern access to congested public infrastructure under regulatory authority. Compute reservations are generally contractual claims on services operated by suppliers. No single compute regulator plays the FAA’s role across clouds. Machines can also differ in memory, networking and software compatibility. Permission to transfer a reservation would not make those differences disappear.\n\nThe analogy supplies a design question, not proof that a secondary exchange will be efficient. Providers may reasonably restrict transfers to preserve customer screening, capacity planning or product segmentation. An intermediary can add fees and operational ambiguity. If providers already reallocate released capacity cheaply, a separate exchange might duplicate an existing function rather than improve it. Whether buyers would pay enough for independent transferability remains an empirical question.\n\nA utilization rule also deserves skepticism. Requiring customers to run work merely to retain future access could reward waste. A compute market should investigate release incentives and explicit cancellation terms before borrowing aviation’s use-or-lose mechanism.\n\n## What PCT should implement first\n\nThe immediate institution is a rights register attached to each quote: who may purchase, whether cancellation or transfer is permitted, the cutoff time, usable service interval, and the evidence needed for execution. Human approval should cover the actual commitment and exit restrictions. The receipt should connect the accepted quote, provider confirmation and service delivered, without implying that a payment receipt proves workload success.\n\nThis approach advances the full requirements-to-receipt workflow while admitting that some products stop at recommendation and customer execution. The open research questions are measurable: how frequently do bookings become unnecessary, how early is that known, what recovery do authorized cancellations achieve, and which suppliers permit reassignment? Until those answers exist, transferability is a contract feature to verify and value, not liquidity to assume.", "sources": [ { "id": "S1", "title": "FAA slots are advance authorizations for planned aircraft operations, distinct from air traffic control clearance.", "url": "https://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/systemops/perf_analysis/slot_administration/slot_definition", "publication_date": "2020-04-24", "observation_date": "2026-11-06" }, { "id": "S2", "title": "1985 final rule enabled specified slot sales and leases beginning April 1 1986, with documentation and FAA confirmation. Original use threshold was 65%, not later 80%.", "url": "https://archives.federalregister.gov/issue_slice/1985/12/20/52168-52190.pdf", "publication_date": "1985-12-20", "observation_date": "2026-11-06" }, { "id": "S3", "title": "FAA overview describes 80% utilization and potential waivers; it is not exhaustive current airport-specific guidance.", "url": "https://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/systemops/perf_analysis/slot_administration/compliance_oversight", "publication_date": "2021-02-01", "observation_date": "2026-11-06" }, { "id": "S4", "title": "Capacity Blocks have specified reservation offerings and a cleanup interval before the end.", "url": "https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/capacity-blocks-how.html", "publication_date": null, "observation_date": "2026-11-06" }, { "id": "S5", "title": "Section 5.5.2 prohibits Capacity Block transfer, resale and third-party proxy purchasing; full price is nonrefundable.", "url": "https://aws.amazon.com/service-terms/", "publication_date": "2026-11-05", "observation_date": "2026-11-06" } ], "facts": [ { "claim": "FAA slots are advance authorizations for planned aircraft operations, distinct from air traffic control clearance.", "source": "https://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/systemops/perf_analysis/slot_administration/slot_definition", "date": "2020-04-24" }, { "claim": "1985 final rule enabled specified slot sales and leases beginning April 1 1986, with documentation and FAA confirmation. Original use threshold was 65%, not later 80%.", "source": "https://archives.federalregister.gov/issue_slice/1985/12/20/52168-52190.pdf", "date": "1985-12-20" }, { "claim": "FAA overview describes 80% utilization and potential waivers; it is not exhaustive current airport-specific guidance.", "source": "https://www.faa.gov/about/office_org/headquarters_offices/ato/service_units/systemops/perf_analysis/slot_administration/compliance_oversight", "date": "2021-02-01" }, { "claim": "Capacity Blocks have specified reservation offerings and a cleanup interval before the end.", "source": "https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/capacity-blocks-how.html", "date": null, "accessed": "2026-11-06" }, { "claim": "Section 5.5.2 prohibits Capacity Block transfer, resale and third-party proxy purchasing; full price is nonrefundable.", "source": "https://aws.amazon.com/service-terms/", "date": "2026-11-05", "date_note": "Page last updated; not evidence that this clause changed on that day." } ], "inferences": [ "Transfer permission should be checked before hypothetical resale proceeds enter procurement economics.", "Provider-specific execution restrictions mean recommendation-to-execution workflows need different permitted paths.", "A rights register can be useful without establishing a secondary exchange." ], "unknowns": [ "No empirical estimates of readiness risk, cancellation demand, recovery value or secondary liquidity.", "No claim that airport slot transfers caused improved efficiency.", "No provider-specific partnership exception or account-specific agreement inspected.", "AWS page update date does not establish novelty of cited clauses." ], "correction_history": [], "license_status": "Original PCT editorial text; public license not selected. Source rights remain separate.", "public_url": null, "desk_extensions": { "date": "2026-11-06", "desk": "Market formation", "title": "Airport slots and the missing right to transfer compute", "word_count": 1105, "estimated_read_minutes": 5.5, "novelty": "Different from Nov 3 freight clearing, Nov 4 electricity zones and Nov 5 grain standards: rights to transfer and administrative recognition of assignment.", "worked_example": { "status": "Hypothetical, not a quote", "currency": "USD", "readiness_probability": 0.7, "A_nonrefundable_price": 10000, "B_price_if_used": 11000, "B_cancel_fee": 1000, "B_expected_cost": 8000, "expected_saving": 2000, "break_even_probability": 0.9, "assumptions": [ "Same technically validated capacity and operating costs", "Readiness known before cancellation deadline", "No migration or other fee differences", "No resale assumed" ] }, "editorial_notes": [ "Public research only. No external changes.", "Source text verified via web reads; archive PDF extracted text inspected.", "Body links directly to primary evidence. Arithmetic independently computed." ] }, "public_projection": { "version": "1.0.0", "date": "2026-11-10", "changes": [ "Internal workspace artifact paths omitted where present. Historical license/status fields retain original provenance; current public reuse policy is stated separately." ], "rights": "PCT reserves rights in original editorial content; third-party rights retained. No general source-data reuse license." } }, { "id": "pct-2026-11-06-spotlight-c405e6f869", "content_type": "article", "desk": "spotlight", "title": "OpenAI Batch API: buying the catalogue update instead of the GPU hour", "publication_date": "2026-11-06", "observation_date": "2026-11-06", "editorial_voice": "AI-assisted Spotlight Desk", "body": "A catalogue has 100,000 descriptions waiting for cleaner categories and short summaries. Nobody needs the answer to the first record in three seconds. They need a complete, usable dataset before the next publishing cycle. That distinction changes the compute purchase.\n\nOpenAI’s Batch API packages inference around deferred delivery. The supplier’s April 23, 2024 enterprise announcement described a 50% discount for asynchronous workloads, including classification and evaluation. This is an established service, not a fresh November launch. The useful question is what a builder must do to turn that discount into a dependable production result. [S1]\n\n## A workload you can picture\n\nOpenAI’s cookbook, dated April 24, 2024, demonstrates extracting movie categories and one-sentence summaries from descriptions. It also includes image captioning. The page is now explicitly marked archived, so treat it as a concrete workflow illustration rather than a current implementation recipe or performance benchmark. [S2]\n\nThe catalogue example here is PCT’s own planning case. Assume the operator has rights to process the text and a fixed category vocabulary. Each record can be processed independently. The output should contain the record identifier, permitted categories and a short description supported by the input. No production catalogue or customer order was inspected for this article.\n\nThis is attractive builder territory because the job is easy to make visible. You can count the records, inspect samples, test the category rules and decide whether the result is good enough before committing the whole collection. A cheaper run that produces inconsistent labels still leaves an expensive cleanup job.\n\n## A batch needs a receipt\n\nThe current guide permits up to 50,000 requests per batch and a maximum 200 MB input file. It also imposes model-specific queued-input-token limits. Our 100,000-record case therefore needs at least two batches by request count alone; file size and account limits may require more. Splitting the work does not create extra quota.\n\nOutputs may arrive in a different order from inputs. The guide tells developers to join them using custom identifiers, and failed requests have a separate error file. A practical receipt should reconcile every submitted identifier with a successful response, an error or unresolved work. [S3]\n\nThat receipt is valuable beyond avoiding a spreadsheet accident. It lets the operator rerun only the records needing attention and preserve accepted results. Replaying an entire collection because one slice failed is an avoidable cost, and can produce inconsistent versions of records that were already usable.\n\nPCT would also keep the prompt version, category vocabulary, chosen model version and acceptance checks beside the output. A clean audit trail makes it possible to explain why a record changed when the next catalogue refresh arrives.\n\n## Put numbers on the choice\n\nHere is illustrative arithmetic, not an OpenAI quote. Assume 100,000 requests, each with 2,000 input tokens and 200 output tokens. That is 200 million input tokens and 20 million output tokens. Suppose the eligible synchronous route costs $1 per million input tokens and $4 per million output tokens. Its token bill would be $280: $200 for input and $80 for output.\n\nApply the documented 50% Batch discount to those assumed rates and the token estimate becomes $140. The gross saving is $140, before taxes, preparation, evaluation, storage, retries or engineering work. The token counts include the prompt, not just the catalogue description, and no cache discount or tool use is assumed. Actual model selection still requires quality and eligibility checks. [S4]\n\nNow suppose this particular implementation requires three additional staff hours at an assumed internal cost of $75 an hour. That adds $225. For one run, the batch workflow costs $365 against the $280 baseline under these assumptions. It has not paid back its incremental setup effort. After two equivalent runs, token savings of $280 exceed that one-time cost by $55, assuming no extra ongoing overhead.\n\nThere is a good commercial lesson here. A 50% token discount does not mean a 50% reduction in project cost. Repeat frequency, existing tooling and the amount of review required can matter more than the posted rate. An established batch pipeline may need little additional setup; a one-off migration may need substantially more than this example.\n\n## The deadline and retention tests\n\nThe current FAQ specifies a 24-hour processing window, but also describes expiration: unfinished work is canceled, completed results are returned and completed work is charged. Treat partial completion as an outcome the application must handle. Submitting at the last possible moment leaves no room for correction or another attempt. [S4]\n\nFor a publishing pipeline, PCT would choose an internal submission cutoff earlier than the actual deadline and define what happens if some records remain missing. The operator might retain the previous accepted description or hold those records out of the release. That is a product decision to make before the run, not an emergency policy to improvise afterward.\n\nData policy can disqualify the route entirely. OpenAI’s FAQ states that zero data retention does not apply to Batch, even for organizations with that setting enabled. Inputs and outputs follow the applicable retention policies. Buyers with strict requirements must verify the actual service terms and configuration before uploading anything. [S4]\n\n## What the market can learn\n\nThis service sells a different delivery condition for model inference. It does not give the buyer a transferable reservation on named GPUs, and the published discount does not reveal a market-clearing price for physical capacity. PCT’s interpretation is narrower: some workloads can trade immediate response for lower token costs without managing their own training checkpoints or GPU fleet.\n\nThe worthwhile experiment is a small, rights-cleared sample with explicit acceptance criteria and a complete result ledger. Measure accepted records, elapsed time, review effort and actual charges. The opportunity is real enough to test. Whether it is economical for a specific catalogue remains unproven until those measurements replace the assumptions.", "sources": [ { "id": "S1", "title": "Introducing more enterprise-grade features for API customers", "publisher": "OpenAI", "published": "2024-04-23", "url": "https://openai.com/index/more-enterprise-grade-features-for-api-customers/", "accessed": "2026-11-06", "access_note": "Opened primary source and checked relevant passages." }, { "id": "S2", "title": "Batch processing with the Batch API", "publisher": "OpenAI Cookbook", "published": "2024-04-24", "url": "https://developers.openai.com/cookbook/examples/batch_processing", "caveat": "Archived recipe; used for workflow illustration only, not current code or performance.", "accessed": "2026-11-06", "access_note": "Opened primary source and checked relevant passages." }, { "id": "S3", "title": "Batch API guide", "publisher": "OpenAI", "published": null, "url": "https://developers.openai.com/api/docs/guides/batch", "accessed": "2026-11-06", "access_note": "Opened primary source and checked relevant passages." }, { "id": "S4", "title": "Batch API FAQ", "publisher": "OpenAI Help Center", "published": null, "url": "https://help.openai.com/en/articles/9197833-batch-api-faq", "accessed": "2026-11-06", "access_note": "Opened primary source and checked relevant passages." } ], "facts": [ { "id": "F1", "claim": "April 23, 2024 announcement describes discounted asynchronous inference.", "source_ids": [ "S1" ], "evidence_type": "dated_primary_publication" }, { "id": "F2", "claim": "The April 24, 2024 cookbook demonstrates category extraction, summaries and image captioning; currently marked archived.", "source_ids": [ "S2" ], "evidence_type": "primary_example_not_independently_reproduced" }, { "id": "F3", "claim": "Current batch limits include 50,000 requests and 200 MB per input file, plus queued token limits per model.", "source_ids": [ "S3" ], "evidence_type": "current_documentation" }, { "id": "F4", "claim": "Output ordering can differ from input; custom identifiers and error files support reconciliation.", "source_ids": [ "S3" ], "evidence_type": "current_documentation" }, { "id": "F5", "claim": "FAQ states 50% discount against synchronous API pricing and a 24-hour processing window with partial expiry behavior.", "source_ids": [ "S4" ], "evidence_type": "current_documentation" }, { "id": "F6", "claim": "Zero data retention does not apply to Batch.", "source_ids": [ "S4" ], "evidence_type": "current_documentation" } ], "inferences": [ { "id": "I1", "claim": "Count accepted records and review effort when evaluating savings.", "label": "PCT analysis" }, { "id": "I2", "claim": "100,000 requests require at least two batches by request count; size and quota can require more.", "basis_fact_ids": [ "F3" ], "label": "derived constraint" }, { "id": "I3", "claim": "Illustrative token costs are $280 synchronous versus $140 Batch; $225 incremental setup gives $365 first-run Batch cost and $55 net savings after two equivalent runs.", "label": "illustrative arithmetic, not provider quote or measured result", "assumptions": [ "100,000 requests", "2,000 input and 200 output tokens per request", "Synchronous rates assumed $1/M input and $4/M output", "50% Batch discount", "Three incremental setup hours at $75/hour", "No cache discount, tools, taxes, retries or other ongoing differences" ] }, { "id": "I4", "claim": "Batch is a model-service delivery condition, not a tradable GPU reservation or physical capacity clearing-price observation.", "label": "PCT market interpretation" } ], "unknowns": [ "Live account quotas and eligible model versions", "Current exact model quote", "Measured input/output token counts", "Accuracy and accepted-record yield", "Actual completion distribution", "Retry rate and review effort", "Applicable retention configuration", "Rights to any proposed dataset" ], "correction_history": [], "license_status": "Original PCT editorial text; public license not selected. Source rights remain separate.", "public_url": null, "desk_extensions": { "id": "pct-2026-11-06-spotlight-batch-api", "title": "OpenAI Batch API: buying the catalogue update instead of the GPU hour", "section": "Product spotlight", "publisher": "Peterson Compute Trading", "editorial_voice": "Spotlight Desk | AI-assisted, builder-focused", "as_of_date": "2026-11-06", "status": "desk_research", "subject": "OpenAI Batch API for catalogue enrichment", "editorial_word_count": 965, "limitations": [ "No paid execution, upload or account-specific quote", "No interviews or independently reproduced benchmark", "Worked economics use explicit assumed rates, not live OpenAI prices", "Historical cookbook is marked archived; current guide controls mechanics" ], "distinct_from_previous_issue": "New tangible inference-enrichment workflow. Previous spotlights covered AWS Capacity Blocks, Mooncake Store and SageMaker Managed Spot Training.", "copyright_licensing_status": "Original PCT draft with factual paraphrase and source links; no copied datasets or third-party media. Public license not selected.", "public_url": null }, "public_projection": { "version": "1.0.0", "date": "2026-11-10", "changes": [ "Internal workspace artifact paths omitted where present. Historical license/status fields retain original provenance; current public reuse policy is stated separately." ], "rights": "PCT reserves rights in original editorial content; third-party rights retained. No general source-data reuse license." } } ], "corrections": [ { "date": "2026-11-05", "scope": "PCT_2026-11-04.json", "description": "H100 spot and on-demand 06.ai records shared an identifier; prior file remains unchanged, this edition uses unique source, product, term and observation-specific IDs. No price correction implied.", "prior_archive_id": "libfile_5a5661b01848819197ddf74ee9b1d16d" } ], "omitted": "Separate benchmark, provider-price and market-signal datasets are not republished in this article archive." }