DX Today AI Daily Brief - Monday, June 29, 2026

DX Today AI Daily Brief - Monday, June 29, 2026

Elon Musk's xAI moves Grok 4.5 into private beta at SpaceX and Tesla on a 1.5-trillion-parameter foundation, while OpenAI previews its next model, GPT-5.6 Sol, and confirms a $122 billion funding round at an $852 billion valuation. Memory giant SK Hynix prices a roughly $29 billion Nasdaq listing as AI fuels demand for high-bandwidth memory. Anthropic accuses Alibaba of the largest known distillation attack on its Claude models, and Alphabet sheds about $269 billion in market value as investors weigh the soaring cost of the AI race. Nvidia and TSMC bring AI inside the fab, Salesforce moves to acquire Contentful for Agentforce, Meta cuts roughly 8,000 jobs in an AI restructuring, ByteDance ships Seedance 2.0 video, AMD pushes its 2nm Venice chip at TSMC, ServiceNow trims staff amid AI adoption, and Broadcom unveils the OpenAI-designed Jalapeno inference chip.

DX Today - Your daily AI news briefing, delivered every morning.

Send us fan mail

#AINews #ArtificialIntelligence #Grok #OpenAI #DXToday

Transcript

SPEAKER_05 0:00

It's Monday, June 29th, 2026. You're listening to the DX Today AI Daily Brief. Today, Elon Musk's XAI puts a new model into private beta, while OpenAI previews its next system and confirms a massive funding round. Anthropic accuses Alibaba of harvesting its Claude model, and Alphabet sheds hundreds of billions as the cost of the AI race rattles investors. Let's get into it.

SPEAKER_00 0:30

Elon Musk says XAI's newest model, Grok 4.5, has entered private beta testing inside SpaceX and Tesla. Announced Sunday, Baum Mamum, Ma. The model is built on a 1.5 trillion parameter foundation, a 50% jump over Grok 4.4, which shipped in late May. Musk says it draws supplemental training data from Cursor, the popular AAI coding tool, and claims early internal evaluations put its performance close to or beyond Anthropic's flagship Opus. Deploying first inside his own companies gives XAI demanding engineering and manufacturing environments to stress test the system. The company has laid out an aggressive roadmap, saying it intends to train fresh models at SpaceX every month through the end of the year.

SPEAKER_02 1:21

Staying in San Francisco. OpenAI had a busy Sunday of its own, previewing its next generation model, GPT 5.6, codenamed SOL, and confirming an enormous new funding round. The company says it has raised $122 billion in committed capital at a post-money valuation of roughly $852 billion. The round was anchored by strategic partners, including Amazon, Nvidia, and SoftBank, with continued participation from Microsoft and co-led by SoftBank alongside a roster of major investment firms. The capital is earmarked to accelerate the next phase of AI development and crucially to fund the staggering compute and data center commitments OpenAI has made. The sole preview arrives after weeks of speculation that a June launch had slipped into July.

SPEAKER_04 2:22

Now to the chip supply. South Korean memory chip giant SK Heinex is heading to Wall Street. The company has priced a Blockbuster U.S. listing of American depository receipts at about $166 each, aiming to raise close to $29 billion on the Nasdaq, with trading expected to begin as soon as July 10th. SKH has become the top supplier of high bandwidth memory, a critical component for the AI data center boom, and its sole traded shares have climbed roughly 850% over the past year, pushing its market value above $1 trillion. Analysts at HSBC suggested the listing could be worth 20% more than its offer price, underscoring just how central memory has become to the AI supply chain. A serious allegation next.

SPEAKER_01 3:13

Anthropic is making a serious accusation against one of China's biggest tech companies. In a letter to U.S. senators, the company says Alibarbert what it calls the largest known distillation attack on its Claude models. Anthropic alleges that operators affiliated with Alibarber used roughly 25,000 fraudulent accounts to run 28.8 million exchanges with Claude between late April and early June, harvesting its reasoning patterns to train competing models. The technique, adversarial distillation, lets rivals copy an advanced model's behavior while sidestepping enormous research costs. Anthropics says the campaign targeted software engineering and complex reasoning tasks, and it follows earlier claims the company made against other labs. Alibarber has not confirmed the allegations.

SPEAKER_03 4:04

Turning to the markets, it's been a brutal stretch for Alphabet. The parent company of Google saw roughly $269 billion in market value erased over the past week, as investors grew anxious about the spiraling cost of the AI race and the threat it poses to Google's core search business. The sell-off comes even as Alphabet pours unprecedented sums into infrastructure. The company recently moved to raise more than $80 billion through equity sales to fund its buildout, including a $10 billion investment from Berkshire Hathaway, and has guided full-year capital spending toward $180 billion or more. The episode captures the market's central anxiety. AI is both Google's biggest opportunity and its biggest expense. From chips to fabs.

SPEAKER_00 4:54

Nvidia and the world's largest contract chipmaker, TSMC, are deepening their partnership in an unexpected direction, using AI to build the chips themselves. The two companies announced an effort to bring NVIDIA's accelerated computing and AI tools directly into TSMC's fabrication plants, aiming to speed up semiconductor design and manufacturing. The idea is that AI can optimize the fiendishly complex processes inside a modern fab, from chip layout to defect detection, helping squeeze more performance and yield out of each wafer. It's a notable example of AI being turned on its own supply chain, and it comes as demand for advanced chips built on TSMC's cutting-edge nodes continues to vastly outstrip what the industry can produce.

SPEAKER_05 5:42

Enterprise software is consolidating.

SPEAKER_02 5:45

Consolidation in enterprise software is accelerating, and Salesforce is the latest to make a move. The company has signed a definitive agreement to acquire Contentful, a content management platform, to give its Agent Force AI system a native content layer for building personalized digital experiences. It's part of a wider pattern this month. Enterprise vendors are racing to buy the so-called execution layer, the place where AI decisions actually turn into business actions. Asana moved to acquire Stack AI for no-code agent workflows, while Cooper picked up Rossom for intelligent document processing. The common thread is that generic AI cannot substitute for domain-specific data and workflow history, so the biggest software companies are buying the specialized pieces their agents need to get real work done.

SPEAKER_04 6:49

A workforce story now. Meta is reshaping its workforce around artificial intelligence, and for thousands of employees, that means layoffs. The company has begun cutting roughly 8,000 jobs as part of a broad restructuring while reassigning thousands more workers onto AI-focused teams. It's a striking move from a company that remains highly profitable, and it reflects a theme running through 2026. AI, cited simultaneously as a cost saver and a growth engine. More than 150,000 tech roles have been cut across the industry in the first half of the year, with AI named again and again as the rationale. For Meta, the message to investors is that it intends to concentrate talent and spending on the technology it believes will define its next decade. Shifting to AI video.

SPEAKER_01 7:45

TikTok's parent company, ByteDance, is pushing hard into AI video generation. The company has released CDance 2.0, the latest version of its video model, and it's a meaningful step up. ByteDance says the new release delivers noticeably better motion consistency, longer generation windows, and improved character animation, addressing some of the flickering and warping problems that have dogged AI video. It lands in an increasingly crowded field where rivals are racing to make synthetic clips that hold together long enough to be genuinely useful for creators and advertisers. With ByteDance's enormous distribution through TikTok, any leap in video quality has obvious implications for how short form content gets made and for the creators who make it.

SPEAKER_05 8:32

Competition in silicon.

SPEAKER_03 8:34

NVIDIA may dominate the headlines, but AMD is mounting an aggressive challenge in the data center. The company is targeting a market at values at around $120 billion, pushing new accelerators and supercomputing winds to chip away at its larger rival. Central to the effort is its Venice server processor, which AMD describes as among the first high-performance computing chips to enter mass production on a two-nanometer manufacturing process currently ramping up at TSMC. The competitive pressure is a welcome sign for cloud providers and AI labs who have chafed at relying on a single supplier for the scarce, expensive silicon that powers modern AI. A genuine second source could ease both supply constraints and pricing.

SPEAKER_00 9:21

More on jobs. It's a pattern becoming familiar across the industry. Firms that sell AI automation tools are also among the first to apply those tools internally, reducing headcount in the process. For ServiceNow, which builds workflow software for large organizations, the cuts land at a moment when it is positioning AI agents as a centerpiece of its product strategy. The episode adds to a growing tally of 2026 layoffs where employers explicitly cite AI, raising pointed questions about how quickly automation is reshaping white-collar work, and which roles prove most exposed.

SPEAKER_02 10:10

Finally, a new chip. Chipmaker Broadcom has stepped further into the AI spotlight, unveiling a custom inference processor co-developed with OpenAI called Jalopino. It is OpenAI's first in-house chip, and Broadcom provided the silicon implementation, networking, and connectivity with Celestica handling system design. Remarkably, the partners say they took it from initial design to manufacturing tape out in just nine months, an unusually fast cycle, helped along by using OpenAI's own models to accelerate parts of the design work. Built specifically for inference, the running of trained models, Jalapeno is said to deliver performance per watt substantially better than today's state of the art. Initial deployment is targeted for the end of this year, the first step in a multi-generation compute platform.