DX Today AI Daily Brief - Saturday, June 27, 2026

DX Today AI Daily Brief - Saturday, June 27, 2026

Today's briefing covers twelve stories from across the AI ecosystem. OpenAI opens a limited preview of its GPT 5.6 family, Sol, Terra, and Luna, while Google delays Gemini 3.5 Pro into July and Anthropic accuses Alibaba of the largest known distillation attack on Claude. In the markets and infrastructure, SK Hynix plans a 29 billion dollar Nasdaq listing, Qualcomm unveils its Dragonfly C1000 data center CPU with Meta on board and explores an 8 to 10 billion dollar acquisition of Tenstorrent, Microsoft weighs running Claude on its Maia 200 chips, and SpaceX crosses 80 billion dollars in Colossus compute contracts with a new Reflection AI deal. Liquid AI ships a tiny but mighty LFM2.5 edge model, Mirendil launches with a 200 million dollar seed, Baseten raises 1.5 billion dollars for AI inference, and Getty Images surges after a display partnership with OpenAI.

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

Send us fan mail

#AINews #OpenAI #AIChips #AIFunding #DXToday

Transcript

SPEAKER_05 0:00

It's Saturday, June 27, 2026. You're listening to the DX Today AI Daily Brief. Today, OpenAI opens a preview of its new GPT 5.6 models. Google delays Gemini 3.5 Pro into July, and Anthropic accuses Alibaba of the largest distillation attack it has ever seen. Let's get into it.

SPEAKER_00 0:25

OpenAI opened a limited preview of its GPT 5.6 family on June 26th, three tiers named Sol, Terra, and Luna. Sol is the company's most capable model yet, with gains in agentic coding, biology, and cybersecurity. Terra matches the older GPT 5.5 at roughly half the price, while Luna targets high volume, low-cost work. Pricing runs from $1 per million input tokens on Luna up to $30 per million output on SOL. The release adds a new maximum reasoning effort and an ultra mode that spins up subagents for complex tasks. For now, access is limited to trusted partners through the API and codex, with general availability promised in the coming weeks.

SPEAKER_02 1:10

Now, a notable delay. Google has quietly pushed the general release of Gemini 3.5 Pro from June into July, according to insider reports and prediction market data. Early testers flagged problems with token efficiency and long horizon task performance, the kind of extended reasoning that Google does not want to ship before it is ready. The model was announced at Google I.O. on May 19th, where chief executive Sundar Pichai committed to a June launch, drawing audible groans from developers who wanted it that day. By June 26th, market odds of a June 30th release had collapsed to under 5%. The confirmed specs still impress, a 2 million token context window and a deep think mode. But Google now needs a firm date.

SPEAKER_04 2:09

Onto a serious accusation. 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 United States Senators made public this week, Anthropic says operators tied to Alibaba used roughly 25,000 fraudulent accounts to run 28.8 million exchanges with Claude between late April and early June. Distillation is the practice of flooding a rival model with prompts, harvesting the outputs, and using them to train your own system. No accounts were hacked, no firewalls breached. Anthropic says the campaign targeted Claude's agentic reasoning and software engineering skills, and it is asking Congress to act. Alibaba did not respond to requests for comment. Turning to the markets.

SPEAKER_01 2:59

South Korean chipmaker SK Heinex is preparing a Nasdaq listing that could raise $29 billion, targeting a debut as soon as July 10th, according to CNBC. If it lands at that size, it would be the largest tech listing since SpaceX went public earlier this month. SK Heinex is the world's second largest memory maker and the leading supplier of high bandwidth memory, the specialized chips that Nvidia's AI accelerators depend on. Its market value has already passed Samsung Electronics, making it South Korea's most valuable company. A United States listing would make it the first major Korean chipmaker to dual list stateside, plugging it directly into the capital markets pricing the AI build-out.

SPEAKER_05 3:47

More chip news ahead.

SPEAKER_03 3:49

Qualcomm used its investor day on June 25th to unveil the Dragonfly C1000, a data center processor built specifically for agentic AI workloads. Meta has already signed on to deploy it when production begins in 2028. The chip is built on the Open RISC-V architecture and is designed for persistent, multi-step reasoning rather than raw parallel math. Qualcomm also raised its 2029 non-handset revenue target from $22 billion to $40 billion, with $15 billion of that coming from data centers. Investors approved, sending the stock up roughly 15%. For a company still rooted in smartphones, the data center push is now its clearest path to diversification. A small model surprises.

SPEAKER_00 4:36

Liquid AI released LFM 2.5, a 230 million parameter model that is tiny by today's standards yet punches far above its weight. Built on a non-transformer architecture, the company says it matches the performance of transformer models three times its size on core edge reasoning and sequence tasks. The base and tuned versions are available now on Hugging Face, aimed at developers who want capable AI running directly on phones, laptops, and embedded devices rather than in the cloud. It is a pointed counterpoint to the gigawatt scale data center race, a reminder that some of the most useful AI may run on the device in your pocket.

SPEAKER_05 5:16

Back to inference hardware.

SPEAKER_02 5:18

Microsoft is in early talks to run Anthropics Claude on its custom Maya 200 chips through Azure, according to CNBC. The Maya 200, launched in January on a three-nanometer process, is built specifically for inference, the work of generating responses once a model is trained. Microsoft claims it delivers more than 30% better performance per dollar than comparable Nvidia hardware for those tasks. For Anthropic, a deal would add a third compute relationship alongside Amazon's Traininium chips and its SpaceX arrangement. At the scale Claude now operates, even a fraction of traffic moved to cheaper silicon has real consequences for the company's path to profitability ahead of a possible public offering.

SPEAKER_04 6:15

And a major acquisition. Qualcomm is reportedly in early talks to acquire TenSTorent, the AI chip company led by legendary designer Jim Keller, at a valuation between $8 and $10 billion. If completed, it would be the largest chip acquisition since Broadcom bought VMware. TenSTorrent stands out for building on the Open RISC-V standard, rather than the proprietary architectures of NVIDIA and AMD, an approach increasingly attractive to governments and enterprises worried about supply chain sovereignty. The deal would hand Qualcomm a credible data center story and Keller's engineering reputation, the man behind chip designs at Apple, AMD, and Tesla. It is another sign the AI chip market is consolidating in earnest. Compute deals keep growing.

SPEAKER_01 7:08

SpaceX has signed its fourth major compute deal for its Colossus complex in Memphis, leasing capacity to the open source startup Reflection AI at $150 million a month from July 1st. Run through 2029, the contract totals roughly $6.3 billion. With that deal, Colossus has now crossed $80 billion in committed compute revenue across tenants including Anthropic, Google, and Reflection. The site is staggering in scale, a planned two gigawatts of power and more than half a million in video chips. The takeaway is hard to miss. The companies that control compute at this scale now hold real leverage over every AI lab that depends on them.

SPEAKER_05 7:52

Now the funding rounds.

SPEAKER_03 7:54

A new lab called Mirendil launched this week with a $200 million seed round at a $1 billion valuation, led by Andreessen Horowitz, Kleiner Perkins, and Nvidia. Founded by former anthropic researchers, Myrendil is building AI that accelerates AI research itself, neural networks designed to automate the manual work of constructing frontier models. The founding team of around 20 researchers was recruited from Anthropic, XAI, Google Deep Mind, and OpenAI. It is one of the largest seed rounds the sector has seen, and a clear bet that frontier research can happen outside the handful of giant labs that dominate today. Whether that thesis holds is the question investors are paying to answer.

SPEAKER_05 8:42

Another inference mega round.

SPEAKER_00 8:43

Inference infrastructure drew another enormous check this week as Bastin raised $1.5 billion in a Series F round, valuing the company at up to $13 billion. The round was led by Ultimeter Capital, Conviction, and Spark Capital. Baston now processes more than 1 billion inference calls every day, and says revenue has grown roughly 20-fold year over year as demand for serving AI models in production explodes. The raise underscores a shift the whole industry is feeling. Inference, the cost of actually running models for users, has become AI's most contested infrastructure layer, and the capital is flowing there accordingly.

SPEAKER_02 9:24

One more deal today. And finally, Getty Images is having a remarkable week after announcing a multi-year display partnership with OpenAI. Under the agreement, Getty's licensed image libraries will appear inside ChatGPT's search and discovery features. Investors reacted sharply, sending Getty shares up as much as 145%. It is a striking reversal for a company that once banned AI-generated art from its platform and sued an AI image startup for copyright infringement. The companies did not disclose financial terms or whether Getty's images would be used to train future models. For a stock photo business reshaped by generative AI, it marks a pivot from fighting the technology to partnering with it.