DX Today AI Daily Brief - Sunday, June 28, 2026

DX Today AI Daily Brief - Sunday, June 28, 2026

Today's briefing: OpenAI retires GPT-4.5 from ChatGPT as the GPT-4 era closes

Transcript

SPEAKER_05 0:00

It's Sunday, June 28, 2026. You're listening to the DX Today AI Daily Brief. Today, OpenAI retires the last of its GPT-4 generation. Anthropic takes a distillation fight with Alibarber to Washington, and a custom chip reshapes the economics of AI inference. Let's get into it.

SPEAKER_00 0:23

We begin with OpenAI. As of Saturday, June 27th, GPT 4.5 is officially retired from ChatGPT, closing a 30-day sunset that OpenAI first flagged in late May. Existing conversations now migrate automatically to the company's GPT 5.5 instant model, which has served as the consumer default since spring. The move ends the GPT-4 era inside ChatGPT's consumer interface. Though developers keep separate API access for now. GPT 4.5 launched back in February as OpenAI's last frontier model before the GPT-5 capability jump. For teams that tune workflows to its particular reasoning style, this is the notice that the behavior they relied on is now a historical baseline.

SPEAKER_02 1:12

Now to a brewing dispute. Anthropic has accused Alibaba and its Quen lab of running what it calls the largest known distillation attack against its models. In a letter to U.S. Senators made public this week, Anthropic says roughly 25,000 fraudulent accounts generated 28.8 million Clawed exchanges between late April and early June, harvesting outputs to train a rival system. Distillation breaks no firewall. The accounts simply used Claude like any customer at industrial scale, targeting agentic reasoning and software engineering skills. It's asking Congress to treat that extraction as a legal violation.

SPEAKER_04 2:14

Staying with Silicon. OpenAI and Broadcom this week unveiled Jolapano, OpenAI's first custom AI inference chip, handed physically to Sam Altman and Greg Brockman at an event on June 24th. Unlike a training chip, Jalapeno is built purely to run finished models, the work that piles up every time ChatGPT or Codex answers a user. Until now, OpenAI leaned entirely on NVIDIA for that, a cost disadvantage against rivals running their own silicon. The company says the chip went from concept to tape out in nine months, its own models helping speed the design. Early testing points to better performance per watt for inference. Deployment starts late this year, scaling through 2028, and it's tied directly to OpenAI's path toward profitability. Next, a new processor bet.

SPEAKER_01 3:08

Qualcomm is making its move into the data center. At its shareholder meeting on June 25th, the company revealed the Dragonfly C1000, a central processor built specifically for agentic AI workloads, those persistent, multi-step reasoning loops that don't map neatly onto graphics chips. Crucially, Meta has signed on to use it when production begins in 2028. Qualcomm built the chip on the Open RISC-V architecture and says it has already secured two custom deals with hyperscalers. It lifted its 2029 non-handset revenue target to $40 billion, with $15 billion from the data center alone. The stock jumped about 15%. For a company still defined by smartphones, this is its clearest diversification play yet.

SPEAKER_05 3:56

Now, a landmark listing.

SPEAKER_03 3:58

South Korean chipmaker SK Heinix is planning a NASDAQ listing to raise roughly $29 billion, with American depository shares set to begin trading as soon as July 10th. If it lands at that size, it would be the largest tech listing since SpaceX's debut earlier this month. SK Heinix is the world's leading supplier of high bandwidth memory, the specialized chips that Nvidia's AI accelerators cannot run without. Its market value passed Samsung earlier this year, making it South Korea's most valuable company. A U.S. listing gives it direct access to the capital markets where AI infrastructure spending is being priced. Memory is one of the least discussed but most genuine bottlenecks in the entire AI build-out.

SPEAKER_05 4:46

Turning to a costly week.

SPEAKER_00 4:48

Alphabet just lived through one of the largest non-earnings market wipeouts in tech history, shedding around $269 billion in value over several sessions. The trigger was talent. Four senior Google DeepMind researchers departed in six days, including a Transformer co-author bound for OpenAI and a Nobel laureate behind AlphaFold heading to Anthropic. Investors read it plainly. If you're spending nearly $190 billion on AI infrastructure while losing the people who make it produce frontier results, you may be buying depreciating assets. And yet, in the same stretch, Alphabet was added to the Dow Jones Industrial Average, replacing Verizon. Two facts, both true in one remarkable week.

SPEAKER_02 5:34

From Wall Street to Courtside. It's IREN, a digital infrastructure company that began in renewable-powered data centers and has pivoted hard into AI cloud computing, supplying GPU clusters for training and inference. The deal, unveiled this week, replaces longtime sponsor Rakuten and puts an AI data center brand on one of basketball's most visible franchises. Warriors ownership and a star player shared a stage with IRN's founder at Chase Center to market. It's a small detail with a big signal. The companies powering the AI boom now have the cash and the appetite for marquee consumer marketing.

SPEAKER_04 6:30

Now to the workforce. Meta is reportedly cutting around 8,000 jobs, close to 10% of its workforce, as part of an AI-focused restructuring, with several thousand remaining employees reassigned onto AI teams. It fits a pattern that's defined 2026. Across the industry, a majority of this year's tech layoffs now cite artificial intelligence or automation as a driver, even as the very same companies pour record sums into AI capacity. The uncomfortable tension is hard to miss. Firms are funding enormous AI build-outs partly by trimming the headcount that AI is meant to augment. For workers, the message is that restructuring and reinvestment are increasingly two sides of the same strategy. Onto regulation.

SPEAKER_01 7:18

Colorado is back in the spotlight as the first U.S. state to pass a comprehensive AI law. The Colorado AI Act was originally meant to take force on Monday, June 30th, regulating high-risk systems used in consequential decisions. Think hiring, housing, lending, and healthcare. But a spring amendment significantly narrowed it and pushed the effective date to January 1st, 2027, stripping out a duty of care around algorithmic discrimination and easing deployer obligations. Consumer groups say it was gutted. For the rest of the country, the lesson is directional. Colorado's retreat suggests U.S. states are converging on disclosure and transparency rather than the heavier, Europe-style risk management model. The first state AI law is also a cautionary tale.

SPEAKER_05 8:13

Now a shift in spending.

SPEAKER_03 8:15

After two years of what some call token maxing, throwing unlimited AI spend at every problem, enterprises are pivoting to efficiency. Reporting this week highlighted AI startup Lindy, whose chief executive moved 100% of the company's traffic off Anthropics Claude and onto China's DeepSeek, a cheaper open weight alternative purely to cut costs. It captures a broader trend. Chinese open weight models have surged from a sliver of usage to the majority of activity on some developer marketplaces, as teams route routine work to low-cost models and reserve premium Western models for the hardest tasks. For open AI and Anthropic, both eyeing public offerings, this commoditization pressure is now central to the story investors are weighing.

SPEAKER_05 9:05

Two more before we close.

SPEAKER_00 9:07

Oracle is the latest enterprise giant to reshape itself around AI. The company disclosed it has reduced its workforce by roughly 21,000 jobs, about 13%, over the past year, deeper cuts than previously understood, with savings redirected toward AI data centers, and the redirection is paying off on paper. Oracle posted billions in quarterly net income, up sharply year over year, and reported a towering backlog of contracted future revenue as cloud and AI demand climbs. It's the clearest example yet of a legacy software company financing an AI pivot by shrinking its traditional operation. The strategy is straightforward, if stark. Cut where growth has stalled and pour the proceeds into the infrastructure layer everyone now wants to own.

SPEAKER_05 9:58

And finally, physical AI.

SPEAKER_02 10:00

Nvidia is pushing artificial intelligence off the screen and into the physical world. The company introduced Cosmos 3, built as the first fully open Omni model for physical AI, folding vision, world simulation, and action generation into a single architecture. Alongside it came a healthcare robotics platform, including a surgical video data set and tools to generate synthetic training footage of rare procedures, data the real world simply can't supply at scale or safely. The aim is to give robotic surgical systems and clinical assistance the experience they'd otherwise never get. It's a reminder that as the model wars rage in software, Nvidia is quietly laying the groundwork for AI that perceives, simulates, and acts in the real world.