Tech Current Daily Brief — August 25, 2026

The strongest technology signal from the last 24 hours is that AI is spreading outward from model training into the infrastructure, capital, regulation and real-world systems around it. Nvidia moved its Groq 3 LPX inference accelerator into full production as agentic workloads make decode speed a new competitive layer. The chipmaker is also reportedly discussing another investment in Perplexity at a valuation above $30 billion. New reporting from Ukraine showed the darker edge of deployment: officials say a Russian drone that killed three civilians in July operated autonomously using an Nvidia Jetson Orin module. At the policy layer, New Zealand’s prime minister proposed an under-16 social-media ban with penalties tied to global revenue. And investors are placing another aggressive bet on physical AI, with General Intuition in talks to raise at a $6 billion pre-money valuation.

Today’s 5 Top Stories

Nvidia puts Groq 3 LPX into full production as inference becomes the next AI battleground

Nvidia announced on August 24 that Groq 3 LPX, its dedicated interactive-inference accelerator for the Vera Rubin platform, has entered full production. In testing published with Artificial Analysis, Nvidia says the system generated roughly 3,400 output tokens per second on Gemma 4 31B with a 100,000-token input context, about four times the fastest public endpoint in that benchmark. Nebius is the first cloud provider announced for Groq 3 LPX through its Token Factory, while SpaceXAI plans to use Nvidia Vera CPUs for the CPU-heavy orchestration, tool use, code execution and simulation work behind agentic AI.

Why it matters: The performance bottleneck is shifting from how fast models can be trained to how cheaply and responsively they can reason in production. Agentic workloads can require hundreds or thousands of sequential inference steps, making decode latency and tokens-per-dollar strategically important. Nvidia is responding by turning inference into a specialized hardware layer rather than treating GPUs as the only accelerator that matters.

Nvidia is reportedly discussing a Perplexity investment at a $30 billion-plus valuation

Reuters, citing The Information, reported that Nvidia is in talks to invest in Perplexity as part of an equity round that would value the AI search and agent company at more than $30 billion. The proposed valuation would be more than 50% above Perplexity’s previous financing. The report also puts Perplexity’s annualized revenue above $750 million, up from less than $250 million at the start of 2026, with some of that growth attributed to its cloud-based Perplexity Computer agent. Neither company confirmed the discussions.

Why it matters: Nvidia is increasingly investing not only in the infrastructure layer that consumes its chips but also in the applications and agent platforms that can create future inference demand. Perplexity’s reported revenue trajectory also suggests that AI search is evolving into a broader paid-agent business rather than remaining a thin search interface.

Fresh evidence from Ukraine shows lethal autonomous AI moving from theory into deployment

A New York Times investigation published August 24 reported that Ukrainian officials and forensic investigators believe a Russian Molniya drone that struck a gas station in Zaporizhzhia on July 6, killing three civilians, was guided entirely by an experimental AI system rather than a human operator. Investigators said the drone carried an Nvidia Jetson Orin computing module and lacked the communications antennas normally needed for remote piloting. Nvidia said it does not sell the modules in Russia and noted that Jetson devices are widely available through resale markets.

Why it matters: The critical shift is not a new model benchmark but the use of commodity edge-AI hardware inside autonomous weapons. If the Ukrainian assessment is correct, it demonstrates how quickly civilian AI compute can migrate into lethal systems and how difficult export controls become once widely distributed modules are available through secondary markets.

New Zealand proposes an under-16 social-media ban with fines up to 10% of global revenue

New Zealand Prime Minister Christopher Luxon unveiled a proposal that would require major social-media platforms to take reasonable steps to verify that users are at least 16. Possible methods include account information, facial age estimation, digital-ID services and formal identification. Companies that fail to comply could face fines of up to 10% of global revenue. Messaging services, online games and productivity-focused AI apps such as ChatGPT, Gemini and Copilot are excluded. The proposal is politically uncertain: two coalition partners oppose it, and Parliament will dissolve before a vote can occur ahead of the November election.

Why it matters: Australia’s age ban is becoming a policy template rather than an isolated experiment. The technical burden is shifting toward platform-level age assurance, while the size of the proposed penalty shows regulators are increasingly willing to tie compliance risk to global revenue rather than local turnover.

General Intuition is in talks to raise at a $6 billion valuation as physical AI attracts another premium

TechCrunch reported on August 24 that New York-based General Intuition is in talks to raise funding at a $6 billion pre-money valuation from investors including Valor Equity Partners, Point72 Ventures and Seven Seven Six, with existing investors Khosla Ventures and General Catalyst also participating. The reported financing comes only weeks after the company raised $320 million at a $2.3 billion valuation. General Intuition is building foundation models for agents that move through space and time, using hundreds of millions of hours of gameplay and associated action labels as a training signal for generalized behavior.

Opportunity signal: Investors are increasingly treating action data as a possible equivalent of the text corpus that unlocked language models. The bet is that large action models trained on games and simulations can transfer into robotics, where high-quality data about what an agent did—and what happened next—is much scarcer than internet text.

Data & Market Pulse

The numbers today point to an AI market expanding simultaneously across hardware, applications and physical systems. Nvidia says Groq 3 LPX reached roughly 3,400 output tokens per second at 100K context in its published benchmark. Perplexity’s reported annualized revenue has climbed above $750 million from below $250 million at the start of the year. General Intuition’s proposed $6 billion pre-money valuation is roughly 2.6 times the $2.3 billion valuation attached to its round only weeks ago. Meanwhile, New Zealand’s proposed social-media rules would expose noncompliant platforms to penalties as high as 10% of global revenue. The common thread is that AI value—and AI risk—is now being priced outside the foundation-model layer.

Trend Watch

1. Inference is becoming a distinct architecture market. Training created the GPU boom; agentic software is creating demand for systems optimized around decode latency, orchestration and high token throughput.

2. Capital is moving both up and down the AI stack. Nvidia is reported to be backing an application-layer agent company while physical-AI startups are receiving frontier-lab-style valuations before their technology is widely deployed.

3. AI governance is becoming deployment-specific. Social platforms are being pushed toward age verification, while autonomous weapons are exposing a different problem entirely: how to control the downstream use of general-purpose edge compute once it has entered global resale markets.

What to Watch Next

For Nvidia, the next signals are commercial availability and pricing for Groq 3 LPX, Nebius deployment timing, and whether the company gives more detail on inference strategy when it reports results on August 26. For Perplexity, watch whether the reported round closes and whether Nvidia receives commercial terms beyond a conventional equity stake.

For autonomous systems, watch for independent verification of the Ukrainian forensic findings and any export-control response focused on edge-AI modules. In New Zealand, the policy question moves to the election: Luxon will need new parliamentary support before the age-ban proposal can advance. For General Intuition, the key test is whether the reported financing closes and whether gameplay-trained action models can demonstrate repeatable transfer into real robotic systems.