Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Nvidia is extending its Blackwell-based Jetson Thor range with smaller 16GB and 32GB modules due in early 2027, but undisclosed pricing, preliminary specifications and unverified workload claims leave their deployment economics unresolved.
Nvidia is extending Jetson Thor downward rather than replacing its flagship robotics computer. The T3000 and T2000 reduce compute, CPU cores and memory to put the Blackwell-based architecture into more robots and edge systems. Whether that produces a commercially useful middle tier cannot yet be judged: the hardware is months away, the specifications remain partly preliminary and Nvidia has not disclosed prices.
The Jetson T3000 combines a Blackwell GPU, eight Arm Neoverse CPU cores, 32GB of LPDDR5X memory, 273GB/s of memory bandwidth and 25GbE connectivity. Nvidia says in its announcement that the module delivers 865 FP4 teraflops in roughly half the size and power of the T5000.
Those reductions are not proportional. The T3000 has about 42% of the T5000's quoted FP4 throughput and one-quarter of its 128GB memory capacity, while retaining the same stated memory bandwidth. Nvidia's narrower claim is that the T3000 can deliver similar inference performance to the T5000 on multimodal workloads including large language, vision-language, vision-language-action and world foundation models.
That does not establish general performance parity. FP4 is a very-low-precision format used for AI inference, and its throughput cannot be compared directly with conventional FP32 compute or with a competitor's differently defined TOPS figure, as a technical assessment notes. The retained sources include no independent benchmark results for the promoted workloads. Applications constrained by model size or working memory may also experience the cut from 128GB to 32GB differently from workloads limited mainly by bandwidth.
The T2000 makes a deeper trade. Nvidia gives only two headline specifications: 400 FP4 teraflops and 16GB of memory, with visual AI agents, autonomous mobile robots and industrial manipulators among the intended uses. A preliminary comparison, assembled from the launch announcement and a system partner's specifications, lists six CPU cores, 137GB/s of memory bandwidth and 10GbE networking. Its author cautions that Nvidia has not released the full T2000 and T3000 specifications.
| Module | Quoted AI compute | Memory | CPU cores | Memory bandwidth |
|---|---|---|---|---|
| Jetson T2000 | 400 FP4 TFLOPS | 16GB | 6 | 137GB/s |
| Jetson T3000 | 865 FP4 TFLOPS | 32GB | 8 | 273GB/s |
| Jetson T4000 | 1,200 FP4 TFLOPS | 64GB | 12 | 273GB/s |
| Jetson T5000 | 2,070 FP4 TFLOPS | 128GB | 14 | 273GB/s |
Even basic power figures are unsettled. One technical report assigns 70 watts to the T3000 and 40 watts to the T2000, while another analysis estimates about 65 watts for the T3000 and leaves the T2000 unspecified. Nvidia's announcement states only that the T3000 uses roughly half the T5000's power. A separate launch report lists 237GB/s of T3000 memory bandwidth, conflicting with Nvidia's 273GB/s figure.
Nvidia also says in the announcement that it is preparing an IGX T3000 with the same stated performance, integrated functional safety and support for its Halos robotics safety stack. That version broadens the addressable machines operating near people, but it makes the surrounding safety software part of the product decision rather than proving an advantage from compute alone.
Deepu Talla, Nvidia's vice president for robotics and edge AI, said customers wanted a more compact, lower-power Thor option without the roughly 2,000-teraflop class of compute in the T5000, according to a launch interview. The new modules fill that product gap: the T3000 carries half the memory of the T4000 and the T2000 halves it again.
Nvidia explicitly links the T3000 migration pitch to high memory prices. Yet less memory is only a proxy for a lower bill of materials. The company has not said what either module will cost, and it has not quantified system-level savings after other deployment requirements. The existing lineup provides only a rough boundary: one analysis cites a $3,000 T4000 and expects the previous-generation Orin family to remain below the new Thor modules.
The launch also uses software optimization to press the same cost argument. Nvidia says in its account that new Jetson agent skills automate memory optimization, system configuration and deployment across Thor and Orin. It says UBTech, Agile Robots and Connect Tech reduced memory consumption by as much as 15GB and moved from 64GB Jetson AGX Orin modules to 32GB versions. Nvidia also says SandStar cut as much as 4GB and moved from a 16GB Orin NX configuration to 8GB, while NoTraffic reduced memory use by 30% on TX2 NX.
Those are company and partner claims, not independent benchmarks. They also describe different economic outcomes. A smaller memory configuration may lower hardware cost; freed capacity may instead stay in the system as headroom for added features. The examples show that optimization can change a device choice, but they do not establish what a T2000 or T3000 deployment will cost.
Nvidia is packaging the modules with a wider robotics workflow. Cosmos 3 Edge is a 4-billion-parameter world foundation model intended to reason and generate actions through on-device inference. Nvidia says in the product announcement that developers can post-train it for a robot's embodiment and sensors in about a day, then deploy it on Thor for real-time vision analysis and robot policy. JetPack, Isaac simulation and perception tools, Nemotron and GR00T models, Halos safety software and the new agent skills extend that path from development into deployment.
The potential leverage is software continuity: a customer can emulate a smaller module on the current Jetson AGX Thor developer kit and keep the same architecture and software stack when production hardware arrives. If Nvidia prices the new modules aggressively, that continuity could bring more edge systems into its toolchain. Until prices and switching costs are known, however, that remains a strategic possibility rather than a measured customer benefit.
Nor is Nvidia the only vendor selling integration. Qualcomm says its Dragonwing IQ10 reference design combines compute, AI acceleration, camera and sensor interfaces, motion control, networking and a layered robotics software stack. The company advertises up to 700 TOPS, 18 Oryon CPU cores, support for as many as 12 GMSL2 cameras and global availability beginning in September 2026.
That reference design is not a like-for-like substitute for a Jetson module, and Qualcomm's TOPS number cannot rank it against Nvidia's sparse FP4 teraflops. It nevertheless redistributes the competitive question: robot makers are choosing among integration paths, sensor and control interfaces, software ecosystems, availability schedules and costs—not merely peak AI arithmetic.
Developers will see the product in stages. Nvidia says T3000 emulation will arrive with JetPack 7.2.1 later in July 2026 on the existing Jetson AGX Thor developer kit. T2000 emulation will follow in a future release. The physical modules are scheduled for the first quarter of 2027, a sequence also reflected in the launch roadmap and contemporaneous coverage.
Emulation can expose software compatibility and approximate performance targets; it cannot establish the economics or behavior of a finished robot. Three pieces of evidence will determine whether the new modules materially expand Thor's market:
Until then, Nvidia has built a broader product ladder and an earlier software migration path. It has not yet shown that the lower-memory rungs deliver lower deployment costs.
Anthropic’s 20-year lease gives TeraWulf a customer for 401 megawatts of future AI infrastructure, but rent starts only after delivery and the proposed utility deal passes power and grid costs to TeraWulf, leaving financing and project margin unresolved.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Twenty-nine countries signed an agreement creating WAICO as an independent intergovernmental organization, while China paired the launch with capacity-building offers that are not yet confirmed as WAICO programs.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
OpenAI says a handful of GPT-5.6 Sol file-deletion reports most commonly involved Full Access without sandboxing or Auto-review. Its own evaluations show a more complicated risk picture, and the company has not yet published an incident rate or evidence that its promised safeguards stop the failure.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
A rogue browser extension can trigger Claude for Chrome’s fixed Gmail, Docs, Calendar and business workflows; the attack is constrained by default approvals but can run silently for users who enabled unattended action.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Gold Eagle has begun collecting and prioritizing AI-discovered software vulnerabilities, but the voluntary federal clearinghouse has not disclosed results, operating rules or the resources that would turn findings into deployed fixes.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.