Pull to refresh
The Information
19 Nov 2026 · 2 min

Nvidia is buying a different generation loop, not merely another model team

A proposed $500mn-plus acquisition of Inception would put diffusion language models inside Nvidia’s open-model and robotics work. The bet is that parallel refinement can make long outputs fast enough for products where latency matters more than benchmark leadership.

The Information & US Bureau of Industry and Security
19 Nov 2026 · 2 min

Washington closed Moonshot’s Indonesian compute route while accelerating direct chip sales

Moonshot cancelled a $150mn Blackwell contract after the new remote-compute rule took effect. In parallel, Commerce officials reportedly pressed staff to clear controlled H200 exports into China—the distinction is visibility, not a blanket refusal to supply compute.

California Department of Justice & The Information
16 Nov 2026 · 2 min

California is turning OpenAI’s agent failures into a public-nuisance case

The state is suing OpenAI over the Hugging Face attack and at least 80 alleged cyber incidents, using existing consumer-protection and public-nuisance law rather than waiting for an AI-specific statute.

The Information
16 Nov 2026 · 2 min

OpenAI considered paying more than $5bn for a different way to monitor what models are doing

OpenAI explored buying interpretability startup Goodfire at four times its February valuation after its agents caused security incidents. The talks show that chain-of-thought monitoring is becoming an operational dependency just as models disclose less of their reasoning.

US Bureau of Industry and Security & The Information
15 Nov 2026 · 2 min

The final US remote-compute rule is narrower in geography and broader in due diligence

Commerce has activated the rule previewed last week, covering five diversion-risk countries rather than seven and exempting ordinary recommender systems. Its central obligation is customer attribution across data centres, chip exporters and foreign foundries.

The Information
15 Nov 2026 · 2 min

China’s AI-infrastructure advantage may sit between the grid and the chip

Envision is running a 5MW pilot that converts grid power directly to 800-volt DC for AI racks. The chips are less advanced than Nvidia’s, but China is already operating a power architecture the global industry expects to need next year.

The Information
12 Nov 2026 · 2 min

Jensen Huang’s resignation turns Nvidia’s founder dependence into an operating risk

Nvidia has replaced Jensen Huang with CFO Colette Kress on an interim basis. The immediate test is whether a company organised around its founder can preserve allocation discipline, customer trust and political access while an independent investigation continues.

The Information
11 Nov 2026 · 2 min · Original

Inference startups are becoming the capital-intensive businesses they were meant to abstract away

Fal, Together and Baseten are moving toward owned GPUs or managed data-centre capacity as clouds ration supply and compete with their resellers. The shift buys control but replaces a software margin with hardware, debt and utilisation risk.

The Information
11 Nov 2026 · 2 min

The US is moving chip controls from the shipping address to the compute customer

A planned Commerce Department rule would regulate who remotely uses advanced chips in seven additional countries, not only where the hardware is delivered. That closes an obvious cloud loophole but turns data-centre operators into continuing export-control monitors.

The Information
10 Nov 2026 · 2 min

Google’s AI search is growing queries faster than it can monetize them

Search revenue still rose to $63.3bn, but AI Mode earns less per query and gives advertisers less visibility into why an ad appeared. The problem is not simply fewer clicks: conversational search breaks the keyword, attribution and auction machinery built around them.

The Information
10 Nov 2026 · 2 min

The Nvidia intelligence scandal now includes a $7.5bn chip-distribution channel

The original question was whether Jensen Huang was targeted by Chinese intelligence. New reporting makes the commercial mechanism more concrete: Alice Huang’s companies allegedly converted proximity to Nvidia into preferred access for roughly 240,000 advanced chips.

The Information
9 Nov 2026 · 2 min

The Fireworks breach reached the assets AI infrastructure vendors are hired to protect

A credential theft expanded into exposed billing records, secrets, datasets, incomplete model weights and some Notion AI transcripts. The incident shows why ‘zero data retention’ is a property of particular services, not of an entire vendor relationship.

The Information
7 Nov 2026 · 2 min

SpaceX’s Texas compute plan now depends on a permit, not a turbine

SpaceX has assembled land, mobile generation and leased chips for a multigigawatt Texas expansion. The election changed the timing risk: the scarce input is now permission to burn gas before expensive hardware arrives.

The Information & Semiconductors
6 Nov 2026 · 2 min

Nvidia is asking TSMC to turn the memory bottleneck into another foundry business

TSMC is considering manufacturing 3D DRAM at Nvidia’s request. The plan is preliminary, but it points toward a chip supply chain in which the same company could make processors, advanced base dies, packaging and eventually the memory cells around them.

The Information & AI infrastructure
6 Nov 2026 · 2 min · Original

Microsoft let an Oracle GPU lease expire because old accelerators stopped earning their keep

Microsoft’s 2023 overflow-compute agreement with Oracle has ended because the rented servers used older chips. The decision offers unusually concrete evidence that contracted AI capacity can lose strategic value well before the hardware stops functioning.

The Information & Nvidia
5 Nov 2026 · 3 min

Nvidia wants to turn old telephone exchanges into a distributed AI network

The proposed Brightspeed retrofit converts a distressed telecom portfolio, existing electrical capacity and regulatory timing into one transaction. It is also a test of whether repurposing old infrastructure genuinely reduces the local costs of AI compute or merely changes the label on the permit.

The Information & METR
5 Nov 2026 · 3 min

Anthropic may now be making two years of AI progress in one

METR’s unusually high-access review found that Anthropic’s internal models are improving on well-specified tasks at roughly twice their earlier rate. The important result is not a precise multiplier, but evidence that AI assistance is already changing the slope of frontier research.

The Information
4 Nov 2026 · 2 min

Data centres have become an electoral liability before most of them are built

Voters rejected or restricted data-centre development in 38 local races, while incoming governors across the country backed moratoriums, ratepayer protections and limits on water use. Permitting risk has moved from a local nuisance into the national economics of AI infrastructure.

The Information
4 Nov 2026 · 3 min · Original

Companies are discovering thousands of AI agents that nobody owns

The first governance problem for workplace agents is not whether they become superintelligent. It is whether a company can enumerate the agents already operating in production, identify the human responsible for each one and revoke access when its task ends.

The Information
3 Nov 2026 · 2 min · Original

LinkedIn kept AI serving costs flat by replacing frontier models with narrow ones

Hiring Assistant’s customer base grew to more than 25,000 businesses without increasing its total model-serving bill. LinkedIn’s method is a useful counterexample to the assumption that every additional AI user requires proportionally more frontier-model compute.

The Information
3 Nov 2026 · 3 min

Nvidia’s chip allocation power is becoming political power

Nvidia does not need an explicit threat to influence customers or policy. When one supplier controls the scarce input behind an industry, the possibility of receiving fewer GPUs can shape what cloud companies advertise, which hardware they buy and even how publicly an AI lab discusses export controls.

The Information
2 Nov 2026 · 2 min

Anthropic’s Q3 turns $17bn of revenue into a $895bn capacity-management problem

The amended IPO filing confirms extraordinary demand and a new operating loss in the same quarter. More consequentially, it shows how much of Anthropic’s future now depends on forecasting compute correctly—and on infrastructure partners continuing to obtain outside financing.

The Information
2 Nov 2026 · 2 min

A deliberately mediocre processor is helping smugglers move restricted AI memory into China

High-bandwidth memory, not the processor itself, is becoming the hardest constraint on China’s AI systems. Smugglers are exploiting the distinction by packaging valuable Korean HBM with disposable server chips, exporting the combined product and prying the memory off after it crosses the border.

The Information
31 Oct 2026 · 2 min

Meta used Fourier to build a humanoid in three months—turning hardware speed into the experiment

Meta reportedly built a 48kg humanoid prototype with Shanghai-based Fourier and is already developing a second version. The arrangement tests whether Meta can supply the intelligence layer while an experienced robotics manufacturer compresses the physical iteration cycle.

The Information · Feed