Strategic PartnersIssue 486 records
NVIDIA
Cal Poly's November 2025 plan for its $3 million NVIDIA AI Factory (four DGX B200 systems, full operation targeted for January 2026, energy tracking through the campus solar farm) is newly relevant after the system's Oct 22 unveiling, which the research stream has already covered; no metered power or energy figures have been published. In San Jose, the council voted 8-0 on Oct 20 to uphold a 20 MW data-center permit inside NVIDIA's leased chip-testing building, whose data hall is supplied with 8.5 MW, rejecting a labor-led appeal that said the in-fill environmental exemption ignored power and cooling needs. The city's draft standards for 20 MW+ data centers and R&D facilities, with water, clean-power, generator and audit rules, go to the Planning Commission Nov 18, as other California cities choose moratoria. NVIDIA also released Nemotron 3.5 Bolt E4B, a small local agent model; an independent test found it about as accurate as its Gemma base but much faster.
- Evidence records
- 6
- Cross-source patterns
- 2
- Evidence classes
- 2 vendor claim2 independent reporting1 standards or public-body guidance1 independent research
- Outcomes
- 3 emerging2 mixed1 cautionary
- Source freshness
- 4 new this fortnight1 older, newly relevant1 recent
- Research completed
- 2026-10-24
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
NVIDIA compute placed in existing buildings is justified by capacity claims that no one has yet metered or verified
Cal Poly's November 2025 plan put a four-node DGX B200 AI Factory on campus, promised tools to track its energy use through the university's solar farm and targeted full operation by January 2026; no measured power or energy figures have been published. In San Jose, a 20 MW data-center permit inside NVIDIA's leased chip-testing building (existing data hall supplied with 8.5 MW) was upheld under an in-fill exemption, while appellants argued the review ignored the power, UPS and cooling needed. San José's draft standards would require third-party verification of actual energy and water use against environmental assumptions.
Operating questionWhen GPU capacity is added inside an existing campus or commercial building, what metered before-and-after power, cooling and water data support the claim of little or no new impact, and who verifies it after opening?
Supporting evidenceCalifornia Polytechnic State University, San Luis Obispo; NVIDIA; Mark III SystemsCity of San José (City Council, Planning); NVIDIA (tenant); Menlo Equities (owner); Citizens for Responsible Industry (appellant)City of San José (City Manager's Office, Mayor's Office)
Local approvals of AI compute proceed case by case while the rules that would govern them are still drafts
San Jose upheld the NVIDIA-building data-center permit on Oct 20 while its own 20 MW+ standards, which explicitly cover R&D facilities and move such appeals to the council, await Planning Commission review on Nov 18 and council readings in December. Elsewhere in California, Oakland, San Francisco and Gilroy chose moratoria, Los Angeles County postponed a ban vote, and an advocacy tally counts at least 26 restrictions proposed or passed since June, while new state laws focus on disclosure.
Operating questionFor a planned AI compute site, which local definitions, moratoria and draft standards could apply before it opens, and does the project plan hold if pending rules are applied to applications still under review?
Supporting evidenceCity of San José (City Council, Planning); NVIDIA (tenant); Menlo Equities (owner); Citizens for Responsible Industry (appellant)City of San José (City Manager's Office, Mayor's Office)City of San José; City of Santa Clara; City of Oakland; City and County of San Francisco; City of Gilroy; Los Angeles County; Food & Water Watch
Evidence ledger
Full record · every source keeps its link and limitations
Cal Poly (Nov 2025): AI Factory to use four NVIDIA DGX B200 systems with Mark III Systems, fully operational by January 2026, with ownership and cost-avoidance claims
Cal Poly's announcement says the AI Factory is powered by an NVIDIA DGX B200 BasePOD, funded by a $3 million Noyce School investment, built with Mark III Systems, and comprises four DGX B200 systems with high-performance storage, networking and the NVIDIA AI software stack. It projected full operation by January 2026. It claims ownership of data, models and applications, says comparable cloud GPU capacity can cost individual users thousands of dollars, cites up to three times training and 15 times inference performance versus previous-generation systems, and mentions tools to track energy use through the campus solar farm. It envisages regional businesses working with faculty and students on the system.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant because the system was unveiled on Oct 22, 2026: comparing this plan with the launch shows a roughly nine-month gap between the projected January 2026 full operation and broad campus access, and gives the configuration and claims a U.S. public university used to justify an owned cluster.
- Evidence and measured results
- University press release (operator claim, classed as vendor-claim because it is self-reported and repeats NVIDIA performance figures). Performance multipliers are generation-over-generation figures, not Cal Poly measurements. Cloud cost comparison is qualitative. No utilisation targets, staffing plan or energy figures.
- Limitations and uncertainty
- Pre-deployment announcement; the January 2026 target was not met for broad access; performance figures are vendor-sourced.
San José Spotlight: San Jose council rejects appeal, upholding a 20 MW data-center permit inside NVIDIA's leased chip-testing building at 300 Holger Way
On Tuesday Oct 20 the San Jose City Council voted unanimously, with one member absent and the vice mayor recused over a financial conflict, to reject a labor-backed appeal of planners' July 8 approval of a special use permit for a data center using up to 20 megawatts inside an existing R&D building at 300 Holger Way. The owner, Menlo Equities, says NVIDIA leased the building about two years ago on a 10-year lease, has spent $100 million remodelling it for testing and research, and already runs a chip-testing lab with a data hall supplied with 8.5 MW. The appellants' attorney argued the city ignored power distribution, uninterruptible power and cooling infrastructure a 20 MW data center needs. A councilmember said the city's data-center definition needs updating.
Why it matters, evidence and limitations
- Why it matters
- Shows how a city handles an NVIDIA facility at the boundary between R&D lab and data center: an in-fill environmental exemption for interior changes, contested by labor and environmental groups, decided while the city drafts standards for 20 MW+ projects. Relevant to local-government permitting staff and to universities or agencies hosting GPU capacity in existing buildings.
- Evidence and measured results
- Local news report of a council vote with quotes from the owner's representative, the appellants' attorney and residents. Power figures (8.5 MW existing, 20 MW permitted) come from the owner and the permit. No measured emissions, noise or utility data. City staff materials (opened separately, not cited) describe a Class 32 in-fill exemption and no exterior changes.
- Limitations and uncertainty
- Reporting relies on parties to the dispute; the article carries no NVIDIA statement; the actual future power draw is not known.
City of San José: draft Uniform Standards for 20 MW+ data centers and large-load R&D facilities—recycled or low-water cooling, carbon-neutral power, 25-hour generator testing cap, third-party verification
San José's draft standards define Large Load Data Centers and Large Load Research and Development Facilities as 20 MW or greater. Proposals include City Council hearing of land-use and CEQA appeals above 20 MW; cooling with 100% recycled water or low-water closed-loop or air-cooled systems, banning potable open-loop cooling; 100% carbon-neutral electricity; EPA Tier 4 generators with testing limited to 25 hours a year; annual sewer-discharge and noise monitoring; third-party verification of actual energy and water use against environmental-clearance assumptions; a 1,500-foot notification radius; and an anticipated $4-5 per square foot housing linkage fee and about $50,000 per MW reliability fee, both under study. The Planning Commission is due to review them Nov 18, with council readings Dec 8 and 15; they would apply to applications still under review when they take effect.
Why it matters, evidence and limitations
- Why it matters
- Newly relevant because the council's Oct 20 Holger Way decision concerned an NVIDIA-leased R&D building permitted at 20 MW, the threshold the draft uses, and because the standards cover R&D facilities, not only data centers. A detailed local template for any city, county or university district reviewing GPU facilities.
- Evidence and measured results
- Official city release describing draft standards subject to change; no adopted text, no enforcement record, fees still under study. Retrieved via a fetch tool after curl received HTTP 403.
- Limitations and uncertainty
- Draft only; a critic quoted in separate local coverage says the standards rely on words like 'encourage' and 'monitor'; generator limits do not apply during outages.
San Jose Inside/CalMatters: as Oakland, San Francisco and Gilroy pause data centers, San Jose and Santa Clara choose standards over moratoria; at least 26 California restrictions proposed or passed since June
CalMatters reporting republished by San Jose Inside says Oakland unanimously approved a 45-day moratorium on new data-center approvals in early October, San Francisco approved a similar moratorium, Gilroy extended its September moratorium, and Los Angeles County postponed a vote on banning hyperscale facilities on unincorporated land. San Jose rejected a temporary moratorium in favour of permanent standards and this week voted 8-0 to uphold the data center inside the NVIDIA-used building at 300 Holger Way; Santa Clara is pursuing stricter water and air rules. Food & Water Watch counts at least 26 resolutions, moratoriums or bans passed or proposed in California since June, while the seven state laws Newsom signed last month focus mainly on disclosure, environmental studies and grid protection.
Why it matters, evidence and limitations
- Why it matters
- Gives the regional policy setting for any California public body hosting or approving GPU capacity: local governments are moving faster than the state and are split between pauses and standards. Agencies and universities planning on-site AI compute should expect scrutiny of water, power and rate impacts.
- Evidence and measured results
- Regional news report summarising local actions with quotes from Food & Water Watch and an industry coalition; the count of 26 is an advocacy group's tally. No impact data on the facilities themselves.
- Limitations and uncertainty
- Advocacy-group tally; article focuses on commercial data centers, not public-sector or campus GPU clusters; contains a typo repeating 'San Jose'.
NVIDIA releases Nemotron 3.5 Bolt E4B 4GB, an open small model post-trained from Gemma 4 E4B for local conversational agents on RTX, Jetson, RTX Spark and DGX Spark
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model built on Google's Gemma 4 E4B with Nemotron post-training, aimed at voice and device assistants, interactive characters and robotic or IoT interfaces running on GeForce RTX, Jetson, RTX Spark and DGX Spark. NVIDIA claims about 40% higher average quality than its prior model on conversational-agent benchmarks, up to about 80% fewer output tokens than peers, operation in as little as about 4 GB in target workloads, multi-turn tool use, configurable reasoning, 64k context and support for 35 languages. It cites memory, latency, cost and privacy constraints as reasons developers choose local models.
Why it matters, evidence and limitations
- Why it matters
- Limited and indirect: small local models could support on-device assistants for public-facing kiosks, accessibility tools or field devices where data cannot leave the device, and teaching on campus RTX or Jetson hardware. No public-sector deployment is described.
- Evidence and measured results
- Vendor release with vendor-selected benchmarks measured on an RTX 5090; no independent evaluation on the page. See the Atomic Agent test for a small independent comparison.
- Limitations and uncertainty
- Vendor claims; benchmarks chosen by NVIDIA; no SLED deployment evidence.
Atomic Agent test: Nemotron 3.5 Bolt E4B nearly ties its Gemma 4 E4B base on agent tasks (12/18 vs 11/18) but finishes far faster; the '4GB' GGUF file is 6.18 GB
Hours after release, Atomic Agent ran Bolt E4B and its base Gemma 4 E4B on six offline agent tasks, three runs each, on one RTX 5090. Bolt passed 12 of 18 runs and Gemma 11 of 18, which the authors call close to a tie; Bolt finished all runs in 113 seconds versus 662, mostly because it generated about a tenth of the tokens. VRAM at 64K context was about 4.9 GB for Bolt and 4.3 GB for Gemma; NVIDIA's Q4_K_M file was 6.18 GB. Bolt ships Gemma 4's chat template under a Nemotron name, which caused their agent to fall back to a plain mode. Licences differ: OpenMDW-1.1 for Bolt, Apache 2.0 for Gemma.
Why it matters, evidence and limitations
- Why it matters
- Useful corrective for public-sector teams evaluating small local models: the independent test shows accuracy roughly equal to the free base model, with speed gains, licence differences and integration quirks that vendor pages omit.
- Evidence and measured results
- Small independent test (n=3 per task, six tasks, one GPU) by a developer with its own agent product; authors note one-run differences may be noise and that they used one sampling setting for all models rather than NVIDIA's recommended temperature.
- Limitations and uncertainty
- Very small sample; single hardware setup; tester's own product; reasoning cap of ~500 tokens per step may affect reasoning-mode results.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Independent research
- Research conducted outside the implementing organization.