The Biggest AI News This Week
🍎 Apple Sues OpenAI for Trade Secret Theft Related to the Development of Hardware Products
🏛️ Google DeepMind CEO Demis Hassabis Proposes Global AI Regulatory Board Led by the US
📉 IBM Stock Drops Sharply Over Drops in Mainframe and Software Sales Due to AI
🧩 Thinking Machines Ships Its First Model, and It’s Open-Weight
🚫 New York Governor Signs Bill Mandating a One-Year Ban on Building New, Large-Scale Data Centers
🕵️ US Weighs Options Over Adversarial Distillation of US Frontier AI Models by Chinese AI Companies
🦅 White House Launches Gold Eagle AI Cybersecurity Coordination Program
💰 DeepSeek Weighs New Funding Weeks After $7 Billion Round and Eyes a 2027 IPO
⚕️ CVS and Google Form a Consumer-Focused Healthcare Alliance
📖 Report You Should Read: 2026 State of AI: The Builder’s Economy from Iconiq Capital
📊 Data You Can Use: AI Start-Up Boom, Data Centers in Tornado Zones
📃 An Article You Should Read Right Now: Generative AI Is an Engineering Disaster from The Atlantic by Alex Reiser
🃏 Definitely Not AI: T-Rex Auction, Dog & Hot Dog, Butterflies & Bees Back, the Crazy Sun, and Johnny Cash
1️⃣ Apple Sues OpenAI for Trade Secret Theft Related to the Development of Hardware Products
Apple sued OpenAI on July 10, accusing the company of a “systematic effort” to steal trade secrets for its hardware business. The complaint singles out OpenAI’s Chief Hardware Officer Tang Tan and technical staffer Chang Liu, both former Apple employees, as well as Jony Ive’s IO Products, which OpenAI bought for $6.5 billion in 2025. Apple wants an injunction and the return of its intellectual property.
Deeper Dive
The suit, filed in the Northern District of California, alleges Tan emailed himself confidential supplier data before leaving Apple and told departing employees to bring “actual parts” to OpenAI interviews for “show and tell.”
Liu allegedly used a former colleague’s Apple laptop to download dozens of confidential hardware files and coached her on evading Apple’s security team.
Apple also says OpenAI approached a manufacturing partner to learn its proprietary metal-finishing technique.
More than 400 former Apple employees now work at OpenAI, according to the filing. Apple first flagged its concerns in a February letter it says went unanswered.
“OpenAI’s nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misappropriated trade secrets.”
- Apple, in its complaint
OpenAI spokesman Drew Pusateri said the company has “no interest in other companies’ trade secrets.”
The Takeaway
Apple is casting doubt on the viability of OpenAI’s hardware roadmap ahead the release of its devices in early 2027, calling the business “rotten to its core” by alleged theft.
The case is an early template for how IP disputes between incumbent hardware makers and AI labs will shape hiring practices and vendor risk across the industry.
2️⃣ Google DeepMind CEO Demis Hassabis Proposes Global AI Regulatory Board Led by the US
Demis Hassabis called for the US to lead a new AI standards body model, with authority to test frontier models before release and to coordinate an industry-wide slowdown if one is judged too risky. The group would be modeled after the Financial Industry Regulatory Authority, which serves as the private regulatory body for the financial industry. The proposal drew rare public support from rival lab chiefs.
Deeper Dive
In a post titled “A Framework for Frontier AI and the Dawning of a New Age,” the Google DeepMind CEO said the body should draw independent experts and open-source representatives, funded mostly by industry, and wants it operating before year-end. He told Axios the Trump administration’s private signals have been “very positive.” Hassabis has spent months briefing the White House, other labs, and European officials on the proposal.
“This is a thoughtful proposal from Demis.” - Sam Altman, OpenAI CEO
“An important piece from Demis. We need more of this kind of thinking.” - Satya Nadella, Microsoft CEO
Sundar Pichai and Elon Musk praised the idea publicly, an unusual moment of unity among bitter rivals. Altman had separately floated a similar, more explicitly global version in a Financial Times op-ed. The proposals represent a follow-up to President Trump’s June executive order, which asks companies to submit new models for a voluntary, 30-day government review before release, not the mandatory regime Hassabis envisions.
The Takeaway
Modeled on FINRA: industry-funded, independently governed, with authority to pause a risky model’s release.
Public backing from OpenAI, Microsoft, Google, and SpaceX is a rare show of unity among competing labs.
Whether this becomes real oversight or a lobbying vehicle for incumbents will shape how enterprises weigh model risk going forward.
3️⃣ IBM Stock Drops Sharply Over Drops in Mainframe and Software Sales Due to AI
IBM shares fell more than 25% on July 14, its worst trading day since the 1960s, after CEO Arvind Krishna warned investors that clients redirected spending toward AI infrastructure, leaving software and consulting deals unclosed. Revenue rose just 1%, with infrastructure revenue down 7%.
Deeper Dive
In a letter released eight days ahead of IBM’s scheduled earnings call, Krishna said the company “did not adapt and move quickly enough” as clients shifted capital toward scarce servers, storage, and memory ahead of expected price increases, compounded by “rapidly evolving, industry-wide cybersecurity concerns.”
“These conditions require our teams to execute perfectly, and this quarter we faltered.” - Arvind Krishna, IBM Chairman, President, and CEO
The warning revived fears of the “SaaSpocalypse,” in which AI agents will erode demand for software subscriptions.
Evercore ISI analyst Amit Daryanani attributed the miss mainly to mainframe and related software weaknesses.
EMARKETER’s Jacob Bourne called it a “triple whammy” of hardware-first spending, investor punishment of legacy vendors, and pressure from AI-native challengers like Anthropic, calling it “a disruption story, not an extinction one.”
IBM CEO Krishna’s candor drew praise from former Cisco CEO John Chambers, who called transparency “Rule 101 on setbacks.”
The Takeaway
IBM’s infrastructure revenue fell 7% as clients shifted spend to AI chips and servers, a case study in how fast procurement priorities flip when compute becomes the constraint.
Analysts are split on durability: Evercore frames this as an IBM-specific mainframe issue, while others expect more “casualties” as AI capex crowds out software budgets.
Krishna’s blunt shareholder letter is already being cited as a model for crisis communication other CEOs should study.
4️⃣ Thinking Machines Ships Its First Model, and It’s Open-Weight
Thinking Machines Lab, the startup led by former OpenAI CTO Mira Murati, released its first model, Inkling, on July 16. The 975-billion-parameter model is open-weight, meaning developers can download and modify it. Inkling was built to balance cost against power rather than chase state-of-the-art benchmarks.
Deeper Dive
Only 41 billion of Inkling’s parameters activate on a given query, making it cheaper to run than dense frontier models.
Thinking Machines said the architecture “largely follows” DeepSeek’s V3, and that part of the post-training used data from Chinese startup Moonshot AI’s Kimi K2.5 — notable given that OpenAI and Anthropic have both accused Chinese labs of distilling their models.
The model can be fine-tuned through Tinker, the company’s customization platform.
“Inkling is not the strongest overall model available today, open or closed... We trained it to be a broad, balanced foundation model: strong across many domains, flexible enough to adapt.”
- Thinking Machines Lab
CEO Murati frames the release as proof AI should be decentralized rather than concentrated in a handful of labs.
The Takeaway
Inkling’s reliance on DeepSeek’s architecture and Chinese training data is a rare admission that US open-weight efforts are still catching up to China.
Sparse activation (41B of 975B parameters) is the efficiency story enterprises should watch as inference cost, not benchmark scores, drives model selection.
5️⃣ New York Governor Signs Bill Mandating a One-Year Ban on Building New, Large-Scale Data Centers
New York Governor Kathy Hochul signed an executive order on July 14 pausing state environmental review for new data centers over 50 megawatts for one year — the first statewide moratorium of its kind in the country. Hochul is also proposing an end to data center tax incentives.
Deeper Dive
The order directs the state’s Department of Public Service to study data centers’ environmental and energy impacts and to develop a new generic environmental impact statement during the pause. It’s a lighter-touch version of the Responsible Data Center Development Act, a bipartisan bill capping facilities at 20 megawatts that has sat on Hochul’s desk since the legislature passed it in May.
“We have no choice but to address the challenges created by these massive facilities.” - Kathy Hochul, New York Governor
State Senator Kristen Gonzalez, the bill’s sponsor, said the pause would ensure “development and innovation do not come at the expense of all of us.”
Food and Water Watch’s Alex Beauchamp called it “a gigantic step forward.”
At least 13 other states have introduced similar moratoriums this year; Maine’s legislature passed one in April, but Governor Janet Mills vetoed it.
The Takeaway
Enterprises with data center or colocation plans in New York should expect delays and start scouting alternative markets now.
The gap between Hochul’s 50-megawatt order and the legislature’s 20-megawatt bill still sitting on her desk suggests New York’s rules could tighten further within the year.
With 13-plus states now weighing similar bans, site-selection risk for AI infrastructure is becoming a genuine multi-state variable rather than a one-off local fight.
6️⃣ US Weighs Options Over Adversarial Distillation of US Frontier AI Models by Chinese AI Companies
The White House and Congress are considering implementing countermeasures against “adversarial distillation,” the practice of mining a rival’s AI model with mass queries to train a cheaper competitor. Anthropic and OpenAI have accused Chinese labs, including DeepSeek, Moonshot AI, and Alibaba, of running the practice at scale against their models.
Deeper Dive
Anthropic told federal officials in late June that Alibaba ran the “largest distillation campaign yet,” an “industrial scale” attack, while separately citing more than 3.4 million suspicious exchanges traced to Moonshot AI. Google has also reported a rise in similar attempts. US officials estimate unauthorized distillation costs American AI labs as much as $6 billion annually.
Office of Science and Technology Policy director Michael Kratsios wrote in a memo that distillation used legitimately “is a vital part of the ecosystem,” but added there is “nothing innovative about systematically extracting and copying the innovations of American industry.”
A bipartisan group of lawmakers has proposed sanctioning Chinese actors involved in mass-scale distillation, and OpenAI and Anthropic are pushing for antitrust exemptions to share threat data with each other. Not everyone agrees there’s a real threat:
Independent consultant Harold Mansfield argues Anthropic is “using their relationship with the government” for competitive advantage, while Domyn CEO Uljan Sharka called the fight “unreasonable and unproductive.”
The Takeaway
Enterprises using Chinese open-weight models to cut costs should build a contingency plan; access could tighten quickly if Washington moves on sanctions.
The $6 billion estimated annual cost to US labs gives the distillation fight real commercial stakes beyond the geopolitical framing.
Skepticism from independent researchers is a reminder that leading labs have a financial incentive to frame competitive pressure as a security threat.
7️⃣ White House Launches Gold Eagle AI Cybersecurity Coordination Program
The White House launched Gold Eagle on July 14, the first program to emerge from its June executive order on AI cybersecurity. The initiative is a clearinghouse linking government agencies and companies to coordinate on AI-related vulnerabilities and fixes.
Deeper Dive
A senior White House official praised open-source software providers’ contributions during a briefing but declined to name participating companies. Asked about a rumored order targeting Chinese open-source technology, the official said,
“I could not be more clear that we are in full support of the U.S. open source community.”
Key parts of the underlying June order, including a voluntary framework for submitting new models to government review before release, are still being finalized and are due within 60 days of the order’s June 2 signing.
The Takeaway
Gold Eagle formalizes what had largely been an ad hoc vulnerability-sharing process between agencies and AI vendors, and it’s likely to become a reference point for federal expectations regarding disclosure.
Coordination alone doesn’t add engineering capacity: security teams already carry more remediation work than they can complete, and AI-accelerated discovery adds to the workload before it adds relief.
The still-unfinished voluntary pre-release review framework, due within 60 days, is the piece enterprises should watch for its impact on procurement timelines.
8️⃣ DeepSeek Weighs New Funding Weeks After $7 Billion Round and Eyes a 2027 IPO
DeepSeek has begun preparing an IPO filing for as soon as this year, targeting a 2027 listing, while separately seeking a new private round just weeks after closing a $7 billion fundraise. The new round targets a pre-money valuation of roughly $74 billion, up sharply from the ~$50 billion valuation on its last round.
Deeper Dive
The Hangzhou-based company is in talks with advisors and aims to finish financial reports by year-end to support a filing in late 2026 or early 2027. DeepSeek is seeking additional funding to expand computing capacity as it pushes into agentic AI.
Founder Liang Wenfeng told investors the company will prioritize open-source research and AGI over monetization. After DeepSeek’s last raise, Liang’s stake has made him the world’s richest AI model founder, with an estimated net worth of $36 billion, ahead of Anthropic’s Dario Amodei and OpenAI’s Greg Brockman.
The Takeaway
A second raise, weeks after a record round, signals that DeepSeek intends to keep pace with US labs on compute and not cede ground on model quality.
Continued heavy investment from Tencent, CATL, and Beijing’s national AI fund shows Chinese AI capital markets are accelerating despite US export controls.
Enterprises evaluating DeepSeek’s open-weight models should watch for new investors with closer ties to the state, which would sharpen Washington’s scrutiny.
9️⃣ CVS and Google Form a Consumer-Focused Healthcare Alliance
CVS Health and Google Cloud unveiled Health100, an AI-native platform meant to serve as an always-on health assistant for scheduling, prescriptions, and coverage questions. CEO David Joyner wants it to become an industry standard regardless of where patients get care.
Deeper Dive
The platform runs on Google Cloud’s AI to connect CVS’s existing apps and build a running health record for patients, with an agent layered on top. Joyner told Axios the company spent 18 months rebuilding the digital plumbing linking Aetna, pharmacies, and providers. CVS is betting that loosening federal interoperability rules will let it assemble a patient’s full care picture.
“We’re at a point where the government is encouraging interoperability, so there’s a race to figure out now what to do with all the data. Is it the technology companies that are going to organize it, or is it the health care companies?” - David Joyner, CVS Health CEO
Health100 launches with a small test group before wider rollout.
Leerink’s Michael Cherny called it CVS’s “most unifying effort” yet, reiterating an Outperform rating.
The Takeaway
CVS is betting that its 90 million-plus customers and multiple care touchpoints give it an edge over pure-tech players like Amazon and OpenAI.
Watch adoption from the initial test group before treating this as validated; health apps have a long history of weak consumer engagement.
🔟 OpenAI Debuts a Physical Keypad for AI Agents
OpenAI launched Codex Micro on July 15, a $230 physical keypad built with keyboard maker Work Louder for managing its Codex coding agents. It’s OpenAI’s first hardware product, arriving well ahead of the Jony Ive-designed smart speaker the company is separately rumored to be building.
Deeper Dive
The macro pad sits beside a standard keyboard and includes a rotary knob, a small joystick, and a push-to-talk button.
Its six backlit “Agent Keys” show the live status of Codex threads; idle, processing, completed, awaiting input, or error via color changes, letting developers glance at agent progress without switching windows.
The dial adjusts a task’s reasoning-effort setting.
Codex Micro ships in Clicky and Silent variants and is being sold as a limited, niche run for power users rather than a mass-market device; OpenAI has said it will only be available until it sells out.
The Takeaway
This is a hardware test balloon, not a consumer play: OpenAI is validating demand for physical interfaces among developers before its larger device ambitions arrive.
The status-at-a-glance design addresses a real problem in agentic coding - tracking multiple parallel agent threads without constant screen switching.
📖 Report You Should Read: 2026 State of AI: The Builder’s Economy from Iconiq Capital
About Iconiq Capital
ICONIQ Capital is a global investment firm whose growth practice works with founders and executives building software and AI companies. Twice a year, ICONIQ surveys hundreds of AI-building executives through its GenAI Surveys and layers in perspectives from its network of CIOs, CTOs, and technical advisors. This report is ICONIQ’s third bi-annual read on the AI builder landscape, and its central finding is a shift from proving AI works to proving AI pays.
AI Models & Infrastructure
Builders have consolidated at the application layer:
43% are building vertical AI products, and another 20% are building horizontal ones, with financial services and healthcare use cases climbing fastest. Agentic capabilities and agent reliability rank as the top two customer-facing investment priorities for the next year.
Model sourcing has broadened too. Companies now run an average of 3.3 providers, up from 3.1 six months earlier, and Anthropic moved to the top spot among respondents, rising from 51% to 81%.
Reliability and cost remain the leading selection criteria, but security, privacy, and SOC2/enterprise SLAs have climbed sharply, signaling pressure for enterprise readiness as products mature.
AI Go-to-Market & Economics
AI has become a margin story rather than an experiment:
AI products grew from 32% of revenue in 2025 to a projected 42% this year and 53% by 2027, while gross margins climbed from 45% to a projected 59% over the same span; infrastructure and platform products carry the highest margins, at 67%.
Pricing is shifting to match: consumption-based (42%) and outcome-based (23%) models are both rising, and companies now blend 1.7 pricing models on average, up from 1.5.
Hybrid go-to-market motions, splitting product-led and sales-led growth, gained the most ground of any approach.
Talent & Organization
AI is reshaping org charts, not just headcount:
Companies with 50%-plus of revenue driven by AI run flatter, more cross-functional structures with wider spans of control than their peers.
78% of companies are planning either a different mix of roles (45%) or an outright smaller team (33%) because of AI-driven efficiency gains. R&D, sales, and product teams are broadly expected to grow over the next two years, while customer support and G&A are expected to shrink.
Forward-deployed engineers are scaling fast: roughly half of companies plan to make the FDE model a permanent part of their go-to-market motion, primarily as a revenue driver.
AI for Internal Productivity
Internal AI spend is projected to rise from 11% of revenue in 2025 to 19% by 2027, far above the 1-3% companies estimated in earlier surveys:
Half of respondents call the true cost of internal AI “very difficult” to predict, with overruns concentrated in token spend, data infrastructure, and organizational enablement.
R&D teams show the deepest adoption, with 71% tool access and the highest self-reported maturity score, 3.7 of 5, versus 3.1 for G&A.
Coding assistance delivers the strongest productivity gains, averaging 48% at high-growth companies.
Ramp’s internal case study, cited in the report via a public perspective from Ramp’s Internal AI lead (source), credits its 99% tool adoption to more than 350 reusable, git-versioned workflows that let what works scale past a technical few.
📊Data You Can Use: AI Start-Up Boom, Data Centers in Tornado Zones
AI Start-Up Boom
While it may still spawn a jobs apocalypse, artificial intelligence is set to spark a record number of new entrepreneurs in the US. Many of these startups could fail quickly, as AI helps dubious business plans move forward — a new wrinkle on the “AI slop” that’s all over social media. But the AI-enabled startup surge is so robust that it should yield many lasting companies even after the weaker ones peter out, said Aaron Terrazas, an economist who works with the small-business services firm Gusto. Bloomberg Gift Link
Data Centers in Tornado-Prone Geographies
About 40% of US data center capacity — including existing, under construction and planned sites — is set to operate in zones with significant or very high tornado activity, according to a report from insurance giant Swiss Re’s research arm. Nearly 90 gigawatts of data center capacity is planned for areas with one to three tornado days per year, while around 60 GW is planned for areas with three to six tornado days per year. More than a quarter of US data center capacity will be clustered in areas that receive frequent large hail, according to the report. Hail can damage data centers’ roofs, causing water leakage that can affect server halls, often worth at least as much as the building itself, if not several magnitudes more. Bloomberg Gift Link
📃 An Article You Should Read Right Now: Generative AI Is an Engineering Disaster from The Atlantic by Alex Reiser
Here’s the money quote:
The problem with generative AI, in the industry’s own jargon, is that it does not scale. The cost of growing from, say, a thousand users to a million is a key factor that venture capitalists examine when they evaluate start-ups. They want to see that the cost of adding each new user decreases over time, so that the company can support millions of users and make increasing profits. This is achieved partly through the careful engineering of computer systems that can efficiently handle more users who want to post photos, hail Ubers, or stream music.
With generative AI, the work of building efficient, scalable systems has not been done.
First, this article will make you scratch your head, and then, if you have been through several platform shifts as we have here at AI to ROI, it will make you very angry. Read on. The Atlantic
🃏Definitely Not AI
This T-Rex just sold at auction for A LOT of money. Sad pooch finally gets a hot dog from the Miami Marlins. Bees and butterflies are back thanks to roadside wildflower verges.
The schizophrenic sun and everywhere Johnny Cash has been, according to the song.
How to Engage with Us
Contact us directly here with questions about the Big Book of AI Metrics, benchmarking data, podcast advertising, or newsletter partnerships.
The Big Book of AI Metrics
Download the Big Book here to access the tools you need to build a measurement framework that delivers high ROI AI use cases.
The AI to ROI Podcast
If you want to listen to the latest in AI in more detail, Ray and Peter break down our big AI story of the week on the AI to ROI podcast. We also interview leaders at AI product companies and mid- to large-sized enterprises about their challenges and successes in AI. Click here to listen or subscribe using your favorite podcasting app.