NVIDIA compute partnerships, open AI security and Amazon satellites shape TMT premarket

Key Developments

NVIDIA and SSI put Vera Rubin capacity behind frontier-model scaling

NVIDIA and Safe Superintelligence Inc. said July 27 that they entered a multiyear partnership, with NVIDIA also making an investment in SSI (NVIDIA). NVIDIA said SSI will receive access to the next-generation Vera Rubin platform and that the investment plus platform access will allow SSI to increase its compute by an order of magnitude (NVIDIA). TechCrunch corroborated the compute-access structure and reported, citing a person familiar with the deal, that NVIDIA’s investment stretches into multiple billions, while PitchBook data put SSI’s total capital raised at $7 billion and post-money valuation at $32 billion (TechCrunch).

The read-through is that frontier-model partnerships are becoming a direct systems-design channel for NVIDIA. SSI gets scarce Vera Rubin capacity without disclosing a product roadmap, while NVIDIA gets technical feedback from a lab explicitly focused on scaled alignment research. The more consequential angle is customer selection: if NVIDIA can reserve next-generation platforms for research labs whose workloads stress future architectures early, platform roadmaps and model-lab economics become more closely linked.

What to watch: Track whether SSI discloses Vera Rubin deployment scale or timing, whether NVIDIA names more frontier labs with Vera Rubin access, and whether SSI’s Google Cloud relationship remains complementary to the new NVIDIA platform path (TechCrunch).

Open Secure AI Alliance turns model openness into a cybersecurity policy argument

NVIDIA said July 27 that the Open Secure AI Alliance will work to remediate and disclose vulnerabilities using open technologies, with inaugural partners including NVIDIA, Microsoft, Hugging Face, IBM, Red Hat, Salesforce, SAP, Siemens, Snowflake, SpaceXAI, Palantir and the Linux Foundation (NVIDIA). NVIDIA said the recent Hugging Face security incident showed that defenders need open frontier agentic systems, because Hugging Face ran the open-weight GLM 5.2 model on its own infrastructure to analyze more than 17,000 actions and contain the intrusion (NVIDIA). CNBC corroborated the policy backdrop, reporting that NVIDIA, Microsoft, SpaceX, Palantir and other companies launched the initiative as Washington weighs restrictions on Chinese open-weight models (CNBC).

The competitive implication is that open-weight strategy is moving from developer distribution into security governance. NVIDIA and Microsoft are framing openness as a defensive-control issue: enterprises and governments can inspect, adapt and run models locally when incident response requires speed and data control. That argument also narrows the regulatory debate, because the policy target becomes malicious conduct and cross-border model access rather than open tooling itself.

What to watch: Watch whether U.S. agencies separate Chinese-model restrictions from open security tooling, whether Microsoft’s MDASH and NVIDIA’s NOOA are adopted beyond launch partners, and whether Hugging Face publishes more technical detail on the 17,000-action incident response (NVIDIA).

NVIDIA pushes agentic engineering from software toolkit to chip-design loop

NVIDIA said July 26 that it expanded NVIDIA Agent Toolkit for engineering by adding PhysicsNeMo libraries and updated CUDA-X libraries, giving developers agent-ready AI physics skills, accelerated solvers and quantum-chemistry capabilities for engineering workflows (NVIDIA). The company said Cadence is using NVIDIA Nemotron, accelerated computing and CUDA-X libraries with Cadence AuraStack AI Super Agent and Millennium M2000, delivering up to 20x faster multiphysics performance (NVIDIA). In a companion post, NVIDIA said early testing of Cadence Jasper and Synopsys VCS on Vera CPU showed up to 1.5x higher performance on selected workloads, and that Vera combines 88 custom NVIDIA Olympus CPU cores with LPDDR5X memory and second-generation Scalable Coherent Fabric (NVIDIA).

The operational read-through is that NVIDIA is trying to compress the semiconductor design cycle with both model-level agents and CPU-level throughput. PhysicsNeMo and CUDA-X make engineering tools callable by agents, while Vera targets the CPU-bound verification work that still gates tapeout. If Cadence, Synopsys and Siemens workflows standardize around NVIDIA agents and NVIDIA CPUs, the platform value extends from selling accelerators to shaping how future chips are designed.

What to watch: Monitor whether Cadence, Synopsys or Siemens publish customer production metrics beyond selected workload tests, and whether Rosa CPU or Rigel-core roadmap disclosures show the Vera feedback loop converting into measurable chip-design cycle gains (NVIDIA).

Amazon adds a direct-to-device layer to the Kuiper satellite strategy

CNBC reported July 27 that Amazon asked the Federal Communications Commission to approve up to 5,105 internet satellites for a direct-to-device constellation (CNBC). CNBC said the proposed network would use Amazon infrastructure plus Globalstar satellites and spectrum, after Amazon announced in April that it planned to acquire Globalstar for roughly $11.6 billion (CNBC). CNBC also reported that Amazon expects the Globalstar deal to close in 2027, D2D deployment to begin in 2028, Kuiper now has more than 390 satellites in orbit, and Amazon remains required to launch all 3,232 planned first-generation satellites by July 2029 (CNBC).

The strategic read-through is that Amazon is treating satellite connectivity as a layered service rather than a single broadband bet. Kuiper’s initial internet-from-space network addresses fixed and mobility broadband, while Globalstar spectrum can add phone-level reach for users beyond terrestrial coverage. The governing variable is execution timing: a 2028 D2D start and a 2029 first-generation deployment obligation leave little room for launch, regulatory or integration slippage.

What to watch: Track FCC treatment of the 5,105-satellite request, Globalstar deal timing, and whether Amazon discloses carrier or device partners before the planned 2028 D2D deployment (CNBC).

Google AI Overviews move search behavior further inside the platform

TechCrunch reported July 27, citing Similarweb analysis, that Google AI Overviews appeared in 43% of searches, up from 15% a year earlier (TechCrunch). The same report said AI Mode visits rose from 126 million in June 2025 to 279 million by May 2026, and that AI responses with citations increased more than fivefold during the past year (TechCrunch). TechCrunch also said only 6.8% of U.S. ChatGPT desktop queries included citations as of May 2026, while U.S. desktop referrals from ChatGPT more than doubled from 25% in March 2026 to nearly 60% by May 30 after a search update (TechCrunch).

Figure 1 — Google AI Overviews appeared in 43% of searches by mid-2026, up from 15% a year earlier, per TechCrunch citing Similarweb. Source: TechCrunch.

The platform implication is that search distribution is becoming an answer-layer contest. For Google, more AI Overview coverage can keep users inside Google properties longer and shift monetization toward conversational sessions. For publishers and competing AI interfaces, the more important metric is not only whether citations exist, but whether cited answers actually produce outbound visits.

What to watch: Watch whether Google reports AI Mode engagement separately, whether publisher referral data improves or weakens as citation volume rises, and whether ChatGPT’s higher post-update referral share changes Google’s answer-page design choices (TechCrunch).

This is an AI Briefing — AI-generated analysis published under TLCapital.AI. It is not personal research or positions, and it is not investment advice. Figures are sourced to primary filings with dates noted throughout. Do your own diligence.