BIG TECH

4 big AI trends that emerged from Nvidia GTC

Nvidia revealed a torrent of new products at GTC 2026. But beyond the cores, exaflops, and tokens, Nvidia wants to play a larger leadership role in the AI industry, and there's some urgency around why.

So rather than focusing on the chips and hardware announced (spoiler: all are faster and more efficient than previous models), we're going to focus on the downstream effects of these chips and their larger impact on the tech ecosystem.

These were the trends that dominated the show:

1. Robots and Physical AI: As expected, robots were wandering all around the San Jose Convention Center this week. Some gave out candy, some greeted you at the information desk, and others showed off their dexterity at tasks such as picking up Beanie Babies or opening boxes. Nvidia, of course, unveiled new offerings for the entire physical AI realm, including new models for autonomous vehicles and robots, solutions to the data bottleneck, and partnerships with companies ranging from Uber to Figure and World Labs. Though the chip giant isn’t making any robots itself, Nvidia’s strategy is clear: To provide the foundation for anyone that wants to build AI that moves through the real world. 

2. Agents: Nvidia unleashed a lot of hyperbole about OpenClaw at GTC, including calling it the biggest open-source project of all time (it's not, it's simply the one with the most stars on GitHub). Still, there's no denying that personal AI agents are the ChatGPT-level trend of 2026. And beyond hyping it up, Nvidia also delivered NemoClaw, a secure, private framework that makes OpenClaw safer and easier to use — especially for professionals.

3. Open models: Another area where Nvidia doubled down hard at this year's GTC was its open models. Just before GTC, Nvidia unveiled Nemotron 3 Super, its 120B parameter model that is now one of the most open models in the world and is performing surprisingly well in benchmarks. But the big news at GTC was the announcement of the Nemotron Coalition, which brings together Nvidia's six open model families and partner companies investing in open models. Perplexity, Cursor, Thinking Machines Lab, Mistral AI, Black Forest Labs, LangChain, Reflection AI, and Sarvam are inaugural members.

4. AI and jobs: Coding tools were the talk of the town at GTC. But even though these tools are making the future of software engineering look unclear, some of AI’s biggest names are looking at the tech’s impacts beyond code. In a panel, executives of major AI startups largely agreed that coding is simply the first generation of automation, representing a far larger shift that could impact all knowledge work. As Nvidia CEO Jensen Huang said to the panel in front of a packed theater, “almost all work can be specified as code.”

GTC COVERAGE BROUGHT TO YOU BY IREN

Unleashing NVIDIA Blackwell Performance on IREN Cloud™

IREN’s Prince George data center campus is capable of housing over 20k NVIDIA Blackwell and Blackwell Ultra GPUs, and they are constructing a new 10 MW (IT load) liquid-cooled data center to support NVIDIA GB300  NVL72  deployments, capable of handling more than 4.5k GPUs. Supported by IREN’s vertically integrated platform, this installation is part of a broader 4.5 GW power portfolio, offering customers a scalable environment, optimized to run the most complex AI models with confidence. Learn more.

If one thing is clear from the past week, it’s that Nvidia is hustling to keep itself at the center of the AI universe. But the secret sauce that has put it center stage is not just its GPUs but its CUDA software platform for working with those GPUs and other chips. And that factor is now in question. Companies like Modular (which was also at GTC) are working on making a better CUDA that's more efficient and works across more platforms, including Nvidia's biggest rivals. And whether it's Modular or someone else, it feels inevitable that the CUDA advantage is destined to come to an end for Nvidia. So how will it keep itself centered? It will do so by spreading its influence far beyond chips and CUDA, extending its roots into open models, agents, and physical AI. The more it has to offer, the more it can reel in loyal customers to a sticky ecosystem of interconnected AI platforms. To put it plainly, Nvidia doesn’t just want to be a strong competitor in the AI space, it wants to be the game maker.

TOGETHER WITH DELVE

Migrate to Delve and get a $2,000 VISA card in your inbox

Delve is the AI-native compliance platform that actually does the work for you, auto-collecting evidence from AWS, GitHub, and your stack so you don’t have to chase screenshots or babysit integrations. Use AI security questionnaire tooling, AI copilot, and everywhere else to make compliance feel less, dreadful. Welcome to the new age.

The proof is in the pudding:

• Bland → Switched, got compliant, and unlocked $500k ARR in 7 days

• 11x → Streamlined audits and moved faster on enterprise deals

• micro1 → Scaled compliance without adding headcount.

Bonus: Delve will handle your migration for free. Zero-touch. No disruption. No starting over.
If you’re dreading opening your current SOC 2 tool, that’s your sign.

Book a demo here and trigger a migration - get $2000 sent straight to your inbox as soon as you’re onboarded.

PRODUCTS

Cursor’s low-cost model takes on Claude Code

As AI coding tools like Anthropic's Claude Code and OpenAI's Codex soar in popularity, Cursor is raising its game.

On Thursday, Cursor launched Composer 2, the AI startup’s most capable coding model yet. While there are many AI coding models on the market from leading firms, Composer 2’s appeal lies in its ability to perform comparably to those models at a much lower cost. 

“Developers need models good enough to trust and cheap enough to use constantly,” Cursor told The Deep View. “At enterprise scale, cost per token directly affects deployment economics.”

For instance, on the Terminal-Bench 2.0, a benchmark made to evaluate real AI agent workflows, Composer 2 performed comparably to Anthropic's Opus 4.6 and trailed OpenAI's GPT-5.4, while remaining the lowest-priced option by a significant margin.

Quick look below: 

• Composer 2: Score: 59.8%, Price: $0.50/M input and $2.50/M output 

• Opus 4.6: Score: 58.0%, Price: $5/M input and $25/M

• GPT-5.4: Score: 75.1%, Price: $2.50/M input and Output: $15.00/M output 

Composer 2 also outperforms all preceding models on the CursorBench, Terminal Bench 2.0 and SWE-Bench Multilingual. 

Composer was trained for long-horizon tasks using reinforcement learning, making it capable of solving challenging coding workloads. Composer 2 achieved improved model quality and lower cost through continued pretraining, enabling Cursor to scale reinforcement learning, according to the company's blog post. 

However, Cursor notes that beyond the model's prowess, what sets the platform apart from competitors is its full development environment. 

“Our users can choose from Claude, GPT, Gemini, and our own Composer models in one environment,” said Cursor. “The key is product scope: we're a full development environment with long-running agents, code review, a plugin marketplace, automations, and deep enterprise security."

As the AI coding tool race heats up, developers gain access to increasingly capable development tools. This is significant because it not only enables experienced developers to produce what they need faster, but also allows people with ideas and no technical background to throw their hat in the ring. The outcome in the near future should be a new wave of software and application innovation that solves real-world problems and needs. And it's not just coming from Cursor. Google’s latest AI studio upgrades the entire vibe coding experience, now allowing users to build multi-player experiences, add databases and authentication, connect to real-world services, access a more powerful agent, and more. There's never been a better time to start building your own stuff.

TOGETHER WITH VIKTOR

Your Next Hire Isn't on LinkedIn

You need an analyst, a marketer, and more engineering hours. You can't hire for every gap. And OpenClaw showed everyone wants an AI agent, but DIY means Docker, plaintext credentials, and real security risk.

Viktor is the managed answer. It lives in your Slack, connects to 3,000+ tools via secure OAuth, and actually does the work. Ad audits, lead research, revenue reports, bug triage, internal dashboards. SOC 2 compliant. Credentials never touch the AI model.

It learns your company and suggests automations before you ask.

Try Viktor free. No credit card required. $100 in credits to start.

RESEARCH

What Anthropic’s seminal AI study leaves out

Though Silicon Valley might have a rosy picture of what AI can do for people, the broader narrative around the technology is murky. So Anthropic sought clarity. 

On Wednesday, Anthropic published the results of a December survey of more than 80,000 Claude users from 159 countries and 70 languages, who conversed with an AI interviewer built by Anthropic. The survey, which The Deep View’s Sabrina Ortiz took part in during her time at ZDNET, sought to illuminate the desires, fears and reality of AI to better define the “abstract projections of its risks,” Anthropic said.

Broadly, users crave a tool that can manage their work, their life and their time. While many clocked major time savings, thousands reported that AI isn’t making the grade just yet. 

Here are the highlights: 

• What users want: Almost 19% of users reported wanting a tool to help them professionally by handing off routine tasks so that they could focus on more complex work. However, large percentages of users also reported wanting tools to help with life management and personal transformation, with those categories coming in at 13% each. 

• What users fear: The biggest concern was reliability, with more than a quarter (27%) of users surveyed fearing domino effects regarding inaccuracy and hallucinations. But beyond this, users largely feared AI doing more than it should, with respondents reporting concerns of job displacement, autonomy and cognitive atrophy. 

• The reality: The biggest benefit that users have realized by far has been productivity, with 32% reporting that the tech has dramatically sped up work and tedious tasks. However, 19% reported that the tech hasn’t delivered on its promises and isn’t yet capable of doing what they envisioned. 

The goal of this report was to help Anthropic identify what it’s doing well and what it could be doing better. The throughline, the company said, is “that AI helps them live better, not simply work faster.”

Anthropic’s study begs the question of how people really feel about AI. Surveys from organizations like Pew Research tell a different story: Americans are more concerned than excited about AI, believing it will worsen our ability to think for ourselves and harm our ability to form meaningful relationships.

Though Anthropic’s survey is one of the largest research studies on AI ever conducted, it may have a blind spot: It only studied Anthropic users. Because Anthropic’s report seeks to inform Anthropic about its own shortcomings, we should be wary about taking this data at face value. The narrative around AI among the broader public may not be nearly as enthusiastic, with many distrusting and fearing the technology. However, for these companies to succeed in the long term, especially among consumers, they know they need to shift that perception. 

LINKS

• Jeff Bezos in talks for $100 billion fund for AI manufacturing

• OpenAI to merge ChatGPT, Codex, and Atlas into one superapp

• Waymo analysis finds that robotaxis are safer than human drivers

• OpenAI to acquire Astral, a startup offering high-performance Python tooling 

• Google launched new capabilities to enhance agentic shopping experience

• DoorDash launches app that pays people to collect data for AI 

• Samsung to spend more than $73 billion this year on chip capacity, research

• Lovable expands its platform beyond coding to other general-purpose task

• Adobe Firefly: The generative AI creative suite got a series of updates including Custom Models 

• Google AI studio: Google introduced a brand new full-stack vibe coding experience

• ElevenLabs: The AI voice platform introduced a Music Marketplace where artists can now publish their tracks and earn money from it

• Google Drive: With ask Gemini in Drive you can ask questions across your files, calendar, and email with one prompt

Automate the Work That's Slowing Your Team Down

Want to know what world-class talent actually costs in 2026?

Athyna's Salary Report breaks down real salary data across AI, Tech, Data, Design, and more—so you can see exactly where the savings are.

The numbers might surprise you.

Download the report

GAMES

POLL RESULTS

Have you tried any AI agents yet?

Yes (46%)
No (47%)
Other (7%)