PLUS: IBM's worst stock drop and NY bans new data centers
Happy reading
A major breakthrough in on-device AI has arrived. Startup PrismML has successfully compressed a massive 27-billion-parameter model, making it the first of its size to run directly on a modern iPhone.
This breakthrough shifts powerful AI from the cloud to your pocket, unlocking new possibilities for private and persistent on-device agents. Will this major leap in 'intelligence density' fundamentally reshape the balance between local and cloud computing?
In today’s Next in AI:
PrismML’s 27B model that runs on an iPhone
IBM's stock plummets amid AI hardware race
Meta sued for using biased AI in layoffs
New York halts new AI data center construction
An iPhone Heavyweight
Next in AI: AI startup PrismML has released Bonsai 27B, the first 27-billion-parameter model compressed to under 4GB, allowing it to run powerful agentic tasks directly on a modern iPhone.
Explained:
The model comes in two variants: a 5.9GB version for laptops that retains 95% of performance and a 3.9GB version for phones that retains 90%.
Crucial capabilities for agentic work like math and coding remain largely intact, allowing the model to perform complex, multi-step reasoning locally.
Apple is already evaluating the technology, which could enable faster, more private versions of Siri that handle tasks directly on your device.
Why It Matters:
This breakthrough shifts powerful AI from the cloud to your pocket, unlocking new possibilities for private, offline, and persistent on-device agents. By dramatically increasing "intelligence density," this approach changes the economics of AI and could reshape the balance between local and cloud computing.
Big Blue's Black Tuesday
Next in AI: IBM's stock suffered its worst single-day drop in over 50 years following a warning from its CEO that the frantic race for AI hardware is causing customers to shift budgets away from its software and mainframe businesses.
Explained:
Customers are aggressively reallocating capital to buy supply-constrained AI infrastructure like servers, storage, and memory before anticipated price hikes, a trend IBM admitted it failed to adapt to quickly enough.
The company's mainframe business took a significant hit, with infrastructure revenue falling 7% in the second quarter, a much steeper decline than the low-single-digit drop it had forecasted.
The news sent shockwaves through the broader software sector, with companies like Microsoft, Salesforce, and ServiceNow also seeing their shares decline as investors weigh the industry-wide impact.
Why It Matters: This event shows the real-time cost of the AI hardware gold rush, where enterprise spending is being funneled toward foundational infrastructure at the expense of traditional software. It signals a major test for legacy tech giants and the beginning of a potential budget squeeze for software providers everywhere.
Meta's AI Pink Slips
Next in AI: A new lawsuit alleges Meta used biased AI systems to select employees for its May layoffs. The complaint claims the company's automated tools disproportionately targeted workers on protected medical and family leave.
Explained:
The lawsuit argues that Meta's AI measured productivity using inputs like performance ratings and AI token consumption, which inherently penalize employees for being absent. This means workers on protected leave were disproportionately selected for layoff, as their metrics could not be accumulated during their time away.
Plaintiffs claim Meta used a "constellation of internal artificial-intelligence systems" to rank workers, including its internal LLM assistant "Metamate" and other tools that monitored keystrokes, activity, and communications to generate productivity scores.
This case marks the first major challenge to a U.S. company's use of AI in layoffs and reflects a growing legal trend. It follows a similar ruling against Workday for its AI hiring tools and comes as new state laws emerge to regulate automated employment decisions.
Why It Matters: This legal challenge puts a spotlight on the hidden risks of using AI in HR, forcing companies to prove their algorithms are fair and unbiased. The outcome could set a critical precedent for how employers can use automated systems for workforce management and accountability.
The Empire State Says No
Next in AI: New York has become the first U.S. state to halt the construction of new, large-scale AI data centers. Governor Kathy Hochul signed a landmark executive order imposing a one-year moratorium, citing massive energy consumption and rising electricity costs for residents.
Explained:
The temporary ban specifically targets new “hyperscaler” facilities that consume 50 megawatts or more of power, giving the state time to develop a comprehensive environmental impact framework for future projects.
The decision is driven by widespread public concern over utility bills, which have surged nearly 68% since 2019, and the immense strain data centers place on local power grids and water supplies.
While New York is the first to enact a statewide ban, it's a move being watched closely across the country, as legislatures in fourteen other states have introduced bills to restrict data center development.
Why It Matters: This action marks a pivotal moment where the exponential growth of AI infrastructure collides with real-world energy and environmental limits. It establishes a critical precedent for how states may seek to balance attracting tech investment with protecting public resources and grid stability.
AI Pulse
Tesla approved a new $29 billion pay package for CEO Elon Musk, explicitly linking the award to the company’s strategic transition towards becoming a leader in AI and robotics.
CEOs boast that shrinking headcounts are a positive signal of AI adoption and efficiency, reframing staff cuts as a strategic accomplishment rather than a sign of trouble.
Rod Stewart sparked fan debate after using AI-generated visuals in a concert tribute that showed the late Ozzy Osbourne alongside other deceased music legends like Freddie Mercury and Amy Winehouse.