NVIDIA

Samsung Electronics' preliminary guidance puts July–September 2026 operating profit at about ₩107.4 trillion on sales of about ₩195 trillion, nearly nine times a year earlier. The Korea Times says it is the first time a Korean company has topped ₩100 trillion in a quarter. Samsung gave no division split; analysts estimate memory, including HBM4 for Nvidia's Vera Rubin, earned around ₩110 trillion while the devices unit lost money. Final results are due Oct. 29.

Four months after Nvidia and Microsoft announced RTX Spark PCs for local AI agents, Microsoft opened pre-orders for the first one: the Surface Laptop Ultra from $2,599, available Oct. 16, with a $5,999 US-only Surface Dev Box in November. Nvidia named eight PC makers. Microsoft also made Microsoft Execution Containers, a policy-driven containment layer for AI agents on Windows, generally available. Specs are vendor figures, and there are no independent reviews yet.

SpaceX is in early talks to borrow about $40 billion for Nvidia AI chips, the Financial Times and Bloomberg report. The FT, via Reuters, puts the structure at roughly $10B bank loans plus $30B investment-grade debt, with Apollo expected to lead and Pimco among lenders, targeting a 2027 close. Bloomberg independently confirms talks and Pimco interest, and stresses the discussions are early-stage and may not happen. The borrower is SpaceX — not xAI. No party has confirmed a deal.

Nvidia-backed neocloud Lambda is raising up to $4 billion at a $14.5 billion pre-money valuation in what would be its final private round before a planned 2027 IPO, the Wall Street Journal reports, citing people familiar and an LP letter. Blackstone and Coatue are leading. The letter puts Lambda's backlog at $50 billion in September, up from $15 billion in June. TechCrunch infers much of that jump from Anthropic's reported $35 billion commitment. Single-origin WSJ reporting; figures attributed accordingly — the round is not confirmed closed.

Second Vice Minister of Science and ICT Ryu Je-myung said at a Friday (Oct. 2) briefing that Korea will start a state-backed frontier AI project next year, aimed at competing with leading Chinese open-source models and funded by government equity rather than grants. The Korea Times (Oct. 5) reports a 4.7 trillion won ($3.48 billion) earmark — 3.9 trillion for 10,000 Nvidia Vera Rubin GPUs and 800 billion for training data — with participating companies required to match the state’s stake. Structure, anchor firms and governance are still open; the project needs next year’s budget.

Updated Oct. 8, afternoon: Hours after its book-build closed, Firmus’s prospectus and investor presentation were withdrawn from the virtual data room used by fund managers, AFR’s Street Talk reported, and in a later report the AFR said bankers had moved to considering pulling the deal altogether. The ABC said Goldman Sachs was reportedly brought in to try to rescue the float. Backer Maas Group fell as much as 30%, its biggest drop on record per Bloomberg. All of this is reporting, much of it from unnamed sources; Firmus has made no announcement, and there is no final price or withdrawal notice. Earlier on Oct. 8: the book reportedly closed short, with pricing talks near A$8.25 against A$11.

Greenko founders’ AM Intelligence announced two further firm, binding orders for 20,000 Nvidia Rubin GPUs as Vera Rubin NVL72 systems across about 70 MW in India and Malaysia, delivery Q2 2027. With its Aug. 25 Hyderabad order for 9,000 GPUs, committed capacity is about 29,000 Rubin GPUs and close to 100 MW. Mint, citing two people, puts the new orders at about $4 billion — a figure not in the company release. A further 300 MW / $20B+ plan is an intention, not financing in hand.

Reflection published Beam on Oct. 5: a 501-billion-parameter sparse MoE (23B active) open-weight text model it says approaches GLM-5.2 on coding and reasoning with 3–4× less estimated inference compute. Weights, a technical report and a model card are promised later this month under Apache 2.0. Benchmarks are Reflection’s own; Kimi K3 and GLM-5.3 still lead on most shared tasks. Beam is not on Artificial Analysis yet.

Nvidia’s October 2 blog says a 64GB unified-memory DGX Spark configuration — same GB10 Grace Blackwell Superchip, DGX OS, and Nvidia AI stack as the higher-memory SKU — goes on sale Friday, Oct. 23 from Acer, ASUS, Dell, Gigabyte, HP, and MSI, starting at $4,999. Nvidia says it supports up to 100-billion-parameter models on device; two units can pool to 128GB. The up-to-1.7x cluster figure is Nvidia’s own Qwen 3.8 27B test.

After a White House AI luncheon, Trump posted a Joint Commitment on Frontier Responsibilities with Anthropic, OpenAI, Google, Meta, Nvidia, and xAI. The text pledges four layers of self-policing — including an independent external auditor and a board committee — with no enforcement, company-chosen auditors, and no duty to publish findings. It is Washington’s political product after the Astra shelf and rogue-agent week, not a statute and not the earlier UK-tester hold.

At Dreamforce, Salesforce and NVIDIA launched Koa, Salesforce’s first CRM reasoning model for Agentforce, post-trained from Nemotron 3 Super on a synthetic 27-year CRM corpus. Salesforce keeps the weights. Vendor CRM Benchmark claims 3× fewer errors; pilot now, GA winter 2026 US.

OpenAI’s Chris Lehane told Washington reporters that safety talks with Anthropic and Google DeepMind have run for weeks and do not need an antitrust waiver. Same day, Jensen Huang and Mark Zuckerberg rejected a coordinated slowdown.

Tata Consultancy Services said subsidiary HyperVault and partners will invest up to ₹70,000 crore (~$7.4 billion) in a phased 1 GW liquid-cooled AI data-center campus on 264 acres in Hyderabad. Watch whether 1 GW is phase-1 power or a 2030s roadmap number.

Nvidia and Hugging Face signed a definitive agreement: $11.9 billion to shareholders plus up to $1 billion in retention equity, close targeted for the first half of 2027. Huang promises the Hub stays multi-hardware. Signed deal, not the August rumor.

Bloomberg and DCD report Anthropic signed about $35 billion of cloud with Nvidia-backed Lambda for roughly 350 MW at Hut 8’s Beacon Point campus in Nueces County, Texas, with Nvidia holding the lease. Anthropic declined comment.

Nvidia buys $3.5 billion of MediaTek convertible bonds and MediaTek adopts NVLink Fusion — so hyperscalers can drop custom XPUs into Nvidia rack-scale factories. Edge (RTX Spark / DGX Spark) and automotive co-design ride along. How you tax Big Tech silicon without blocking it.

Nvidia reported fiscal Q2 2027 revenue of $96.221 billion, up 106% year over year, guided Q3 to $108 billion ±2% with no China data-center compute, and said Vera Rubin was already about one-fifth of data-center revenue this quarter. Colette Kress pointed to ~70% FY28 growth versus ~44% Street.

Amazon and Nvidia said AWS will add two million more Nvidia GPUs in 2027–2028 on top of the GTC 2026 million-plus commitment, bring Vera CPUs to AWS, and put 100,000 GPUs on IL6+ AWS for U.S. federal AI factories. No dollar terms. The interesting clause is Trainium on NVLink Fusion.

Bloomberg and Reuters reported Anthropic will pay Nscale about $45 billion over six years for roughly 460 megawatts of Nvidia Vera Rubin compute at Nscale’s Monarch campus in Mason County, West Virginia, from late 2027. Anthropic declined comment. Talks/lease reported as agreed — not an Anthropic newsroom ship.

OpenAI’s Engineering post reported the first measured InferenceX numbers for its Broadcom-co-developed inference ASIC: 1.5–1.9× more AI work per watt at peak throughput and 1.7–3.6× lower end-to-end latency versus GB200/GB300. Lab-reported vs commercially shipping Blackwell; first-party deployment planned by year-end. SemiAnalysis verified some runs on-site.

Taiwan’s Keelung District Prosecutors Office indicted nine people — including an employee of Nvidia’s Taiwan unit and Super Micro staff — over the alleged illegal export of Nvidia B300 AI servers to China. Reuters: the scheme dressed 130 B300 systems as staying in Taiwan; later coverage cites 74 shipped via Japan and Indonesia and 56 seized. Allegations, not convictions. First known criminal indictment of an Nvidia employee in a Taiwan AI-chip smuggling case.

Nvidia said xAI will deploy Vera CPUs — the 88-core Olympus CPU Nvidia bills as built for agents — to orchestrate Grok’s next agentic workloads, scale Grok infrastructure on Vera Rubin toward gigawatts, and base its first-generation Starmind AI satellite on an optimized Vera Rubin NVL72 rack. No matching x.ai/news post — Nvidia primary is the source. Starmind is planned, not a launch date.

Nvidia put Groq 3 LPX — the interactive inference accelerator that extends Vera Rubin — into full production. Artificial Analysis measured 3,431 output tokens per second on Gemma 4 31B at 100K context on a private Nvidia-hosted endpoint. Nebius is the first named cloud; Groq will follow with Dell. Report 3,431 as AA-on-Nvidia-gear, not a worldwide production-API record.

The Information reported Nvidia is discussing an investment in Perplexity as part of an equity round that would value the AI search and agent startup at more than $30 billion. Reuters relayed the exclusive. Talks, not a signed deal; Nvidia had also considered a licensing-plus-hire structure. Perplexity declined to comment; Nvidia did not immediately respond.