DX Today AI Daily Brief - Wednesday, July 1, 2026

It's Wednesday, July 1, 2026. You're listening to the DX Today AI Daily Brief. Today, a Chinese food delivery giant open sources one of the largest AI models ever built. Google widens its media AI lineup to cut creator costs.

And venture capital pours into the picks and shovels of the AI economy. Let's get into it. We begin in Beijing, where Maiton, the company often described as China's answer to DoorDash, has released one of the largest open source AI models yet built. Called Longcat 2.

0, it carries 1.6 trillion parameters and a context window of 1 million tokens, meaning it can absorb enormous documents in a single pass. Its most striking claim is not size, but sourcing. May Tuan says the model was trained entirely on domestic Chinese chips, so-called ASIC superpods, rather than NVIDIA hardware.

With export controls limiting China's access to top-tier Nvidia processors, LongCat is a statement that Chinese firms can now train frontier scale systems on homegrown silicon. From Beijing to Mountain View. Google is widening its Gemini family with two new models aimed squarely at cost-conscious creators. The first, Gemini OmniFlash, is built for fast multimodal work across text, image, and audio.

The second, carrying the playful name Nano Banana Too Light, is a lighter, cheaper media model designed to make slick content generation affordable at scale. The pitch is efficiency. As rivals push premium, compute hungry systems, Google is betting that many businesses simply want good enough media generation at a fraction of the price. The launch underscores a broader theme of this year.

The race in AI is shifting from raw capability toward cost per result. Next, a new yardstick for science. OpenAI has turned its attention to biology with the release of GeneBench Pro, a research grade benchmark for measuring how well AI agents can reason through problems in computational biology. It expands an earlier test, Genebench, with harder, more realistic tasks that mimic the messy uncertainty of real laboratory science, and OpenAI has open sourced a set of representative questions.

Benchmarks like this matter because they shape where the field pushes next. By defining what good looks like in scientific reasoning, OpenAI is signaling that drug discovery and the life sciences are a priority frontier. Early results suggest even the strongest models still stumble, passing only a fraction of the toughest problems. Now to the enterprise.

Amazon Web Services is taking a page from the consulting playbook. The cloud giant has launched a forward-deployed engineering team, a group of specialists who embed directly inside customer organizations to accelerate the rollout of agenic AI. The idea is simple but telling. Many enterprises are stuck.

They have the licenses and the ambitions, but they struggle to move autonomous AI agents from pilot to production. By putting its own engineers on the customer's factory floor, so to speak, AWS hopes to close that gap and lock in loyalty. It is also a quiet admission that agentic AI is still hard to deploy, and that handholding, not just software, is what sells right now. Staying with agents and data.

The database company Couchbase is tackling one of the quiet problems holding back AI agents, memory. Its new offering, called the AI data plane, aims to turn an enterprise's scattered, fragmented data into something an agent can actually use as persistent memory. The promise is real-time context retrieval and consistent access, all through a single governed layer, so an agent can remember what it learned and act reliably across tasks. It is a reminder that the headline models grab attention, but the plumbing underneath decides whether AI really works inside a business.

As companies deploy fleets of agents, the winners may be the firms that solve memory and context, not just intelligence. Now follow the money. On to the day's funding, and it was a busy one. Leading the pack, LeapExpert raised $180 million in a growth round led by Riverwood Capital.

The New York company sells what it calls governed communications, capturing and managing business conversations that across the messaging apps employees actually use from WhatsApp to Signal. The thesis is that as regulated work moves off email and onto chat, enterprises need to keep those conversations compliant and increasingly ready to feed into AI systems. The jump is dramatic. Leap experts earlier rounds were a fraction of this size.

A sign investors now see governed conversation data as a category scale prize rather than a niche compliance tool. Building AI into the home building trade. HEGARC, a startup based in Durham, North Carolina, raised $95 million in a Series C led by Insight Partners, pushing its total funding past 170 million. The company builds AI software for the entire home building lifecycle, from design through construction.

Alongside the RAISE, it announced a partnership with the building materials distributor USLBM to bring AI-powered estimating into the supply chain. HECARC's argument is that housing demands structured, three-dimensional data that most AI systems handle poorly, and that owning that data model is its moat. Customers, it says, have compressed development timelines from months to weeks and lifted their margins. It is a bet that AI's next frontier is the messy physical economy.

The hidden layer of AI compute. Not every AI infrastructure bet is about chips or power. Stathera, a Montreal company, raised $55 million in a Series B led by Maverick Silicon to mass produce something most people never think about: timing. Stathera builds silicon-based clocks, the components that keep computing systems synchronized.

As AI data centers scale, precise timing, power efficiency, and integration become surprisingly valuable. And Stathera is pitching its MES-based technology as a modern alternative to traditional quartz. There is a geopolitical thread too. In a year when several Canadian chip firms moved south or were absorbed, Stothera is raising American semiconductor money while keeping its center of gravity in Montreal.

Sometimes the smallest components are the ones that matter most. Keeping the machines alive. Here's a corner of the AI boom you rarely hear about: keeping the machines from breaking. Omen AI raised $31 million in a Series A, led by NARVA Ventures.

The company builds sensors that attach directly to a machine's fluid systems and monitor wear, contamination, and degradation in real time. Its pitch is that as data centers run flat out to power AI, downtime becomes enormously expensive and predictive maintenance turns from a nice to have into a necessity. Omen says its systems already cover data centers managing 10 to 14 gigawatts of capacity. It is a vivid example of how the AI buildout keeps widening, pulling in cooling, fluids, and machine health as part of the investment story.

A robot behind the pharmacy counter. Q emerged from stealth with $12.6 million in seed funding led by Alicorp and a bold claim, the world's first fully autonomous robotic pharmacy. The Palo Alto Company has built a kiosk-style system that dispenses, verifies, and hands over prescriptions with no human behind the counter.

The logic is operational. Retail pharmacy is labor-heavy, compliance sensitive, and chronically short-staffed, so a reliable machine that handles fulfillment could ease a real bottleneck. It is ambitious, and the margin for error in medication is unforgiving. But Q fits a pattern investors backed all day.

Companies using AI, robotics, and automation to compress expensive real-world workflows in healthcare, construction, and beyond. Main Street gets an AI receptionist. Rounding out the funding, a startup called Pi came out of stealth with $19.5 million in a Series A led by Lightspeed Venture Partners.

Its target is the small business, the corner shop, and the local service provider that struggles less with quality than with being found and reached. Pi sells AI tools for growth, local discovery, and customer conversion, including a new product called Frontdesk that answers phone calls on a business owner's behalf. The appeal is concrete. Main Street does not buy strategy platforms.

It buys things that feel like an extra employee or more revenue. Pi is betting it can package AI in a way small operators will actually pay for. Finally, a deal in security. And we close with a deal in cybersecurity.

Aikido, a security software firm, has acquired a company called Root with a specific mission of patching open source software vulnerabilities without forcing disruptive upgrades on customers. Open source code underpins nearly all modern software, but fixing its flaws often means painful version changes that break things. Root's approach lets teams apply targeted patches while staying put. For Aikido, the acquisition deepens its pitch to developers who want security that does not slow them down.

It is a small deal against a day of billion-dollar AI headlines, but it speaks to a durable truth. As AI accelerates how fast code ships, keeping that code secure only grows more urgent. That's your briefing for Wednesday, July 1st, 2026. From the models coming out of Beijing to the sensors keeping data centers alive, today's through line was clear.

The AI story is moving from the model at the center to the infrastructure, capital, and workflows all around it. For DX today, stay curious.