HARDWARE

OpenAI wants to own more of AI’s future

OpenAI has finally unveiled its long-awaited AI chip. 

On Wednesday, the company announced Jalapeño, its first-generation "intelligence processor," developed in collaboration with Broadcom over nine months. The company said in its announcement that the chip was specifically designed around OpenAI's vision for "the future of LLM inference," and marks the first in a broader compute platform the companies are building together. 

The companies said the chip marks a major step in OpenAI's plan to build the full stack behind its models and products. In a statement, OpenAI president Greg Brockman said that, by designing more of the stack itself, the company can continue pushing for broader access to AI with greater efficiency. 

"The world is moving to a compute-powered economy,” Brockman said in the press release. “Jalapeño is part of our long-term full-stack infrastructure strategy to make compute more abundant, resulting in AI which is faster, more reliable, more affordable for people and businesses, and can be used to solve more important problems."

OpenAI said it designed the chip with the ability to work with all LLMs, and was guided by its understanding of the industry's current inference needs. 

• Jalapeño's architecture is built to limit data movement while balancing compute, memory and networking resources. 

• The company said it's still measuring final performance, but in testing so far, performance per watt is "substantially better than current state-of-the-art." 

• Engineering samples of the chip are already running machine learning workloads, including those of GPT‑5.3‑Codex‑Spark, the company said. 

OpenAI said that Jalapeño is part of the company's "full-stack advantage," developing not only models and products on top of them, but the infrastructure underneath. This is about strengthening its progress towards better compute efficiency, better training and more powerful models. By owning the entirety of the stack, each layer is designed for the same goal, it said: "making its models faster, more reliable, and more affordable for users." 

The chip comes after more than a year of snags in its mission to build its own hardware. The company had initially planned to build a network of chip fabs, but abandoned those plans in 2024. The company then announced its partnership with Broadcom to make chips in October last year, ambitiously aiming to deploy 10 gigawatts of custom AI accelerators.

Though this isn't the first time a major tech company has announced a custom AI chip, OpenAI's move has a few implications, Jeremy Roberts, senior director of research and content at Info-Tech Research Group, told The Deep View. 

"This move addresses the challenge of power consumption and efficiency of existing inference chips, reliance on Nvidia for AI hardware, and a potential entry into the datacenter hardware market, as OpenAI says that it will be compatible with other LLMs," said Roberts.

The Jalapeño chip is just the latest in OpenAI's quest to sink its teeth into a larger cross-section of AI's so-called five-layer cake. In May, the company started offering Guaranteed Capacity, giving customers “long-term access” to its compute. It also debuted Multipath Reliable Connection, an open source standard for GPU networking. Each of these initiatives target layers that have nothing to do with its core product offering. And it makes sense: OpenAI wants to be foundational to every part of the industry it believes will completely transform society, consistently preaching its goal of democratizing intelligence. And while leaders at OpenAI and other major tech firms have started to raise concerns about the centralization of power in the industry, by embedding itself within every level of the AI stack, it is inherently making itself a cornerstone upon which all AI innovation rests. 

TOGETHER WITH LIGHTFIELD

The CRM that does the work, not just records it.

"Lightfield has become my system of record and my system of action. It tracks everything — all the conversations, everyone I need to follow up with — and keeps me organized in a way that actually produces outcomes." — Anna Yuan, Co-founder, Scale

Lightfield is an agentic CRM with built-in agents that build your target account list, run personalized outbound in your voice, prep you before every meeting, send follow-ups, and keep every record current automatically.

One platform replacing your CRM, sequencer, enrichment tool, call recorder, and agent builder.

Connect your email and calendar — it starts working inside your first meeting.

Try Lightfield free →

WORKFORCE

Data: US workers are rapidly embracing AI agents

Despite growing anxiety about AI's impact on jobs, few workers appear to be resisting agents.

Only 2% of tech leaders report significant pushback from workers, according to new data from KPMG, as organizations rapidly deploy agents to automate tasks, support decision-making, and coordinate work across teams. Employee adoption of AI agents has already reached 68%, the firm found.

Instead, the findings point to a different challenge: helping workers keep pace with the technology. Among leaders who reported employee resistance, 78% said it stemmed from a lack of skills and fears about job security. More than half also cited concerns around trust and safety, followed by worries that AI is increasing workloads.

These fears are emerging at a time when employers are increasingly citing AI in layoff announcements and entry-level hiring continues to contract. As companies race to deploy AI across their organizations, workers are left trying to make sense of what the technology will mean for their careers.

KPMG's findings suggest that companies have largely moved beyond the question of whether employees will use AI. Instead, the workplace is entering a new phase where AI adoption is expected.

That pressure is already taking shape. Nearly half of tech leaders say AI literacy is a workforce priority, KPMG found, and many companies are introducing mandatory training, usage requirements, and performance metrics tied to AI adoption.

"As the majority signal growing adoption and acknowledge expectations for employees to become AI fluent, it's critical to rethink how tasks are executed by embedding human-machine collaboration into everyday work," said Kevin Bogle, KPMG's US advisory leader for technology, media and telecommunications.

Companies face other hurdles as they scale AI. KPMG found organizations are deploying AI without full visibility into costs, with average AI investments projected to reach $269 million over the next 12 months. As spending rises, tech leaders say that data privacy and a shift toward lower-cost, higher-performing models are shaping their AI strategies over the next six months.

Worker attitudes toward AI agents may change as the technology matures. Despite the hype around autonomous digital employees, companies are still deploying the technology cautiously. A separate KPMG study found that 63% of organizations require human review of AI agent outputs, meaning they can't be taken at face value. Meanwhile, deployments remain limited to low-risk tasks such as triaging IT support tickets or answering HR questions. In other words, the gap between the industry's vision of AI agents and today's reality remains wide. Workers may be embracing AI agents today because they still function more like assistants than replacements. But as companies release tools that make agents perform more like coworkers, such as Anthropic's Claude Tag, that mindset could shift.

TOGETHER WITH JETBRAINS

Junie is out of Beta – the AI Coding Agent by JetBrains

Great cooking takes the right tools in the right hands. A chef's knife isn't flashy – it's precise, balanced, and built for someone who knows how to use it.

That's Junie, and as of this month it's officially out of Beta.

Not a chatbot you paste code into, but an AI agent that wields 15 years of IntelliJ code intelligence – with the debugger, semantic refactors, and code review that CLI tools can't touch.

And like any serious chef, it writes the recipe first. Plan Mode maps requirements, design, and delivery stages before a single line of code. Plan on a top-tier model, implement on a cheaper one – with BYOK, local models, and zero vendor lock-in.

CLI tools are microwaves. Junie is a professional kitchen. The #1 coding agent on SWE-Rebench.

Use code DEEPVIEW20 for $20 in AI credits and give it a try.

Start cooking

PRODUCTS

Figma bets it can make AI a team sport

Most of AI's boost to productivity and efficiency has been focused on tools for individuals. Figma is making it more about teams.

On Wednesday at the Figma Config event in San Francisco, the company released a slew of new tools aimed at creating a full-stack design platform that lets knowledge workers better collaborate with each other, while integrating agents into the process where it makes sense. 

"Right now, design is such a differentiator and so important," said Dylan Field, Figma CEO, in a briefing with the media. "Everyone needs a point of view, and I anticipate that what we're going to see, in the months [and] years ahead, are bolder points of view and more risk-taking visually… We have a bunch of stuff that we've been working on… and I think it really adds a new dimension."

In the same briefing with the media, Loredana Crisan, Figma's chief design officer, said, "Instead of asking whether AI can replace designers, we ask what becomes possible when designers gain new materials and tools to bring their vision to life."

The highlights of what Figma announced include:

• Code becomes a design material: Code Layers allows teams to work with repositories, generate new directions with AI, and sync changes between design and production code without leaving Figma.

• Design moves closer to production-ready: New Motion and Shader tools let teams create animations, 3D effects, and advanced visual treatments inside Figma. The goal is to move beyond static mockups toward experiences that are closer to the final shipped product.

• AI workflows come to the canvas: Figma Weave workflows bring more than 20 AI tools into Figma, turning complex AI processes into reusable building blocks. The idea is to make AI a core layer of the design and development workflow.

• Agents take the annoying tasks: Figma is adding agent skills, deeper context, and generative plugins to allow teams to automate repetitive stuff, build custom tools, and share AI capabilities across a team.

Figma was especially clear about its point of view on how AI works with humans. 

"We do not expect that the creative breakthroughs that will differentiate your company and make you win will come from AI any time soon," said Field.

Figma has one of the clearest perspectives in tech right now and how it's integrating AI into its design tools that 95% of Fortune 500 companies utilize. While lots of companies are playing defense with agents and others are racing forward with automation, it's a lot more likely that we'll start to see more companies like Figma that deeply understand the nuances of integrating AI into the products and take a much more thoughtful, rigorous approach to figuring out where the tools have strengths that can augment human capabilities. That's a welcome step forward in both the conversation and the utilization of AI. 

LINKS

• Google will reportedly lose two more AI researchers to Anthropic

• Meta's upcoming prediction market app reportedly will use its AI

• WSJ finds major chatbots provide left-leaning outputs to political queries

• Microsoft's digital crimes unit says Copilot took down cybercrime tools

• Qualcomm unveils CPU designed for AI data centers, Meta first customer

• Humanoids firm Agility to go public in $2.5 billion SPAC deal

• GPT-5.5 Instant: OpenAI has released a new version of its Instant model, with a better understanding of the intent behind questions.

• Gemini 3.5 Flash: Google's flagship model now has computer use, allowing users to build agents that can interact across platforms.

• Runway: The AI video tool now allows users to localize ads, changing languages based on location.

• Perplexity Computer for Counsel: Built for lawyers, this connects to their research databases, document tools, and matter-management systems.

• Uber: Senior Applied Scientist – AI Red Teaming & Model Risk

• Anthropic: Research Scientist / Engineer, Honesty

• Google DeepMind: Research Scientist, Gemini Safety

• Deloitte: Research Engineer — Post-Training & Small Language Models (SLMs), Healthcare AI

GAMES

POLL RESULTS

Are you concerned about the centralization of power in AI?

Yes (80%)
Somewhat (11%)
No (5%)
Other (4%)