DX Today AI Daily Brief - Thursday, June 25, 2026

DX Today AI Daily Brief - Thursday, June 25, 2026

OpenAI unveils its first custom AI chip built with Broadcom

Transcript

SPEAKER_05 0:00

It's Thursday, June 25, 2026. You're listening to the DX Today AI Daily Brief. Today, OpenAI unveils its first custom AI chip. Micron posts blockbuster earnings as memory becomes AI's hottest commodity. And a wave of acquisitions reshapes the agent economy. Let's get into it.

SPEAKER_00 0:22

OpenAI has unveiled its first custom AI chip, designed in partnership with Broadcom. According to Reuters, the processor, reported to be codenamed Jalapano, is built for inference, the costly work of serving answers to the hundreds of millions of people who query OpenAI's models every day. The company aims to begin deploying the chip by the end of 2026. The move pushes OpenAI directly into the custom silicon race, already reshaping Google, Amazon, Meta, and Microsoft, and it's a bid to ease its dependence on scarce, expensive graphics processors. For OpenAI, owning more of the stack, from chips to data centers, is increasingly the path to controlling both cost and supply. The era of the flashy demo is giving way to the economics of compute.

SPEAKER_02 1:13

Now to chip strategy. Qualcomm is making its boldest move yet beyond smartphones. The company agreed to acquire the AI infrastructure startup modular in an all-stock deal valued at roughly $4 billion. Modular's software helps developers run AI models across different kinds of chips without rewriting their code for each one, a direct challenge to the lock-in created by NVIDIA's CUDA platform. Reuters reports the deal gives Qualcomm a much stronger software layer to pair with its push into data centers and edge computing. The strategic logic is clear. In the AI chip wars, raw hardware performance is no longer enough. Whoever controls the software that lets models run anywhere controls the developers and increasingly the market. Qualcomm wants to be an AI infrastructure company, not just a mobile one.

SPEAKER_04 2:16

From deals to earnings. Memory chips are now among the hottest components in the AI buildout, and Micron technology just proved it. After the closing bell, the company reported fiscal third quarter results that blew past Wall Street expectations, a double beat on both revenue and earnings. Revenue came in around $41 billion, well above forecasts, and the company posted record gross margins north of 80%, driven by surging demand for high bandwidth memory inside AI data centers. Shares jumped more than 16% in after-hours trading, even after sliding earlier in the day amid a broader sell-off in memory names. The result is a reminder that the AI boom isn't only about the chips that do the thinking. It's also about the memory that feeds them. Still in memory.

SPEAKER_01 3:08

Staying with memory, SK Heinex is reportedly preparing one of the year's largest stock offerings. According to CNBC, the South Korean chipmaker plans to raise around $29 billion through a US listing, with trading expected to begin as soon as July 10th. The goal is to fund a massive expansion of high bandwidth memory capacity as AI demand explodes. It's a striking signal of how central memory has become to artificial intelligence. Nvidia's processes grab most of the headlines, but those systems are starving without enough fast memory beside them. If the listing goes ahead, it could reshape how capital flows into the semiconductor sector, pulling tens of billions of dollars toward the unglamorous but essential business of remembering.

SPEAKER_03 3:55

Turning to the enterprise. Amazon Web Services unveiled a new collection of tools aimed at making AI agents more effective and reliable inside real businesses. The push reflects a shift, sweeping the industry. Companies have stopped asking whether AI agents are real and started asking which part of their operations gets handed to an agent first. The challenge AWS is targeting is trust. Agents that can take actions, move data, and complete multi-step tasks are only useful if they're predictable and governable at scale. By building the orchestration and guardrails layer, Amazon is betting that the winners of the agent era won't just be the model makers, but the platforms that make those models safe to deploy.

SPEAKER_05 4:45

Now into your home.

SPEAKER_00 4:47

Google is pushing artificial intelligence deeper into the home. The company is updating Google Home so its smart home AI can recognize familiar people even when their faces aren't visible, using signals like body size and the color of their clothing. The Verge reports that AI-generated descriptions of camera events will also get more detailed, identifying specific sounds such as a barking dog, an alarm, or footsteps. It's a glimpse of where consumer AI is heading. Less typing into a chat bot, more ambient intelligence quietly watching and listening in the background. That convenience comes with familiar tensions. The more context a camera infers about who you are and what you're doing, the sharper the questions become around privacy, consent, and how much our homes should infer about us.

SPEAKER_02 5:37

Next, the trust layer. As AI written text floods inboxes and classrooms, knowing what's human is becoming a feature. The email company Superhuman has agreed to acquire GPT-0, the AI detection startup founded by Edward Tian. TechCrunch reports the deal folds content authenticity directly into everyday productivity software. The logic is telling. As machine-generated writing spreads across email, schoolwork, and publishing, companies are beginning to treat provenance. The question of where a piece of text actually came from as a built-in product feature rather than a niche tool. It marks a quiet turning point. AI detection is moving from a standalone service into the infrastructure of the software we use every day. In an AI-saturated world, trust itself is becoming something vendors will compete to provide.

SPEAKER_04 6:39

Onto marketing. The marketing software world is being rebuilt around AI agents. MoEngage, a customer engagement platform, announced its first ever acquisition, buying the San Francisco startup AMP. As reported by The Economic Times, AMP builds autonomous AI agents that personalize how companies communicate with each individual user. The deal captures a broader shift. Marketing tools are moving away from dashboards and broad audience segments toward agents that can test, adapt, and act on their own across millions of customers, one person at a time. Instead of a marketer choosing a message for a group, the agent decides what to send, to whom and when. It's another sign that agentic AI is quietly moving out of the lab and into the systems that decide what lands in your inbox. From offices to factories.

SPEAKER_01 7:36

AI is also heading to the factory floor. The industrial cybersecurity firm Dragos has launched Ember AI, an assistant built specifically for the teams that protect operational technology, the systems running utilities, factories, and critical infrastructure. Silicon Angle reports the tool is designed to put Dragos' threat intelligence in the hands of analysts at every experience level. This matters because industrial environments are notoriously hard to defend. They often run legacy equipment that can't be patched quickly, and where downtime can mean a halted production line or a disrupted power supply. By giving smaller, stretched security teams an AI assistant that can triage threats faster, Drygos is betting that artificial intelligence can help close a dangerous skills gap before attackers exploited.

SPEAKER_05 8:28

Now the money.

SPEAKER_03 8:51

Tactile builds software that lets banks, lenders, and fintech companies design and run automated decisions, things like approving a loan or flagging a risky transaction without leaning on engineering teams for every change. The Rays reflects a steady investor appetite for applied AI that plugs directly into how money moves. While foundation models dominate the headlines, it's these practical, revenue-adjacent tools that are quietly winning some of the biggest checks.

SPEAKER_05 9:21

More fresh funding.

SPEAKER_00 9:22

Vertical AI is reaching into industries that rarely make tech headlines. ProBook, a New York startup, has raised $34 million in a Series A round led by Andreessen Horowitz following an earlier seed round backed by Sequoia. As Fortune reports, the company is building what it calls an AI operating system for home service businesses, the plumbers, electricians, and contractors who keep things running. It's a telling bet. Home services remain deeply fragmented and operationally messy, full of scheduling, quoting, dispatching, and customer calls, exactly the kind of repetitive workflow AI is well suited to streamline. Investors are increasingly chasing these unglamorous vertical markets, where a focused AI tool can find a clear path to revenue and a real-world problem worth solving.

SPEAKER_02 10:14

One final deal. Our last story takes us to healthcare. Assort Health has raised $120 million in a Series C round, led by Menlo Ventures and Lightspeed Venture Partners, pushing its total funding past $220 million. The company builds AI voice agents that handle patient phone calls for medical practices, the scheduling, the routing, and the endless administrative back and forth that clogs the front desk of nearly every clinic. It's a pointed example of where applied AI is finding traction. Rather than promising to diagnose disease, a sort is targeting the mundane operational burden that exhausts healthcare staff and frustrates patients. As investors hunt for AI with clear returns, the unglamorous work of answering the phone is turning out to be a surprisingly valuable place to start.