Meta Unveils Custom Silicon Plan, Atlassian Cuts 1600 Jobs, Judge Bans AI Browser Orders, Netflix Eyes AI Studio Buy, and more...

Good morning, it's March 12th and this is your daily brief in AI.

Here's everything you need to know.

Meta lays out a four-generation MTIA silicon roadmap through 2027 with MTIA 300 in production

for ranking and recommendations and MTIA 400, 450 and 500 aimed at Gen AI inference and

broader workloads.

Broadcom is handling design elements while TSMC is responsible for fabrication and Meta continues

to buy GPUs and processors from NVIDIA and AMD to diversify supply.

The goal is lower cost per ops and greater data center efficiency, reducing reliance on

external suppliers as infrastructure expands.

Valuation signals show a buy consensus with a target price around mid-800s, supported by

favorable price earnings, price to sales, and price to book ratios.

A conservative view suggests cutting external hardware spending by about 15 to 20 percent

over the next three to five years to boost operating margins and damp and hardware price

volatility.

Insider activity shows caution with multiple insider sales over the past three months signaling

risk considerations for investors.

Risks include regulatory scrutiny and competitive pressure and digital advertising with Meta's

stock showing higher volatility, reflected by a beta around 1.76.

MTIA aims to boost inference efficiency, delivering up to sixfold throughput and 1.5 times

power per watt gains by tightly coupling silicon with the pie torch software stack and prioritizing

dense 72 accelerator racks for cost-effective high-volume predictions.

Initial market reaction was modest, with shares dipping modestly after the MTIA announcements.

Investors will watch MTIA progress for potential cost structure improvements, infrastructure

efficiency gains, and scalable product impact.

Industry analysts tracking AI stock opportunities, note broad investor interest in AI equities,

while Meta plans substantial 2026 data center expansion, guided by a capital expenditure

range of $115 to $135 billion, including developments like the Hyperion Facility in Louisiana.

Atlassian is restructuring the company in a move that affects about 1,600 roles, roughly

10 percent of its global workforce, as it pivots to accelerate its artificial intelligence

and enterprise growth strategy.

CEO Mike Cannon Brooks says the changes aren't about AI replacing people, but about shifting

skills and the mix of roles in certain areas.

About 30 percent of the affected roles are based in Australia, with specifics disclosed

in an SEC filing.

Investory structuring charges are expected to be recorded in Q3 of fiscal year 2026, with

actions and cash outlays largely completed by Q4.

The cuts come as investors demand growth, profitability, speed and value creation, against

a backdrop of volatile market capitalization and recent stock performance.

Atlassian stock has fallen about two-thirds over the past year, trading near $75 to $105

per share, with after-hours movement rising after the announcement.

Founded in 2002 and listed in 2015, Atlassian has struggled to achieve consistent profitability

and hasn't been profitable since fiscal year 2017.

The company describes this as a developing story with updates expected as more details

emerge.

Saving-related charges are estimated at roughly $225 million to $236 million, including

severance and office space reductions, with substantial completion targeted by year end.

The layoff plan includes about $169 million to $174 million in future cash outlays for

severance, notice periods, transitions and benefits, plus $56 to $62 million in office

space exit charges.

While timing and regional specifics aren't fully detailed in the excerpt, the financial

impact and strategic rationale center the story.

Canon Brooks highlighted momentum in cloud revenue growth, strong remaining performance

obligations, and milestones like 600 enterprise customers spending over $1 million annually

and over 5 million users of the ROVO AI suite.

A federal judge has blocked Proplexity's Commod AI browser agents from placing Amazon

orders on a user's behalf, saying Amazon showed strong evidence the agent's access to counts

without authorization.

Proplexity has one week to appeal.

If it does not appeal, it must stop accessing password protected areas of Amazon and destroy

its copies of Amazon data.

Proplexity has not publicly commented.

Amazon says the decision protects a trusted shopping experience for customers.

The case highlights tensions between retailers and AI developers over automated shopping and

access to third-party marketplaces.

Proplexity argues for user autonomy in choosing AI tools and says it will continue to fight

for internet users' right to use any AI they want.

Proplexity has disputed the allegations publicly and vows to press its stance on user choice

in AI tools.

No financial terms of any settlement or damages are disclosed.

The focus is on the ruling and possible appeal.

The injunction raises broader questions about how commercial AI agents can operate for users,

balancing consumer choice, innovation, and online security.

The order includes a one-week stay to allow Proplexity to appeal and the case will proceed

in court.

Proplexity has already appealed the injunction, signaling an ongoing legal dispute with Amazon

over automated purchasing.

The ruling, reported by Bloomberg and issued by United States District Judge Maxine

Chesney, confirms Proplexity will have an opportunity to appeal.

Netflix is nearing a potential $600 million deal to buy Interpositive, an AI startup co-founded

by Ben Affleck, to boost post-production tooling for Netflix originals.

Interpositive's software functions as a post-production AI tool trained on a production zone footage,

helping editors with lighting, relighting, background changes, object removal, and VFX

enhancements rather than acting as an autonomous director.

The upfront cash is expected to be smaller than the total value, with additional payments

tied to performance targets for Interpositive.

Industry analysis frames the AI debate around data control, likeness rights, and financial

terms rather than a simple ban on AI.

Hollywood's public backlash over AI focuses on generative models trained on copyrighted

material, while production tools and synthetic performances are increasingly accepted.

Public outrage from stars targets generative models, yet many in the industry are quietly

adopting and licensing AI tools for onset use and post-production.

The broader sentiment identifies three AI categories, generative models, production tools,

and synthetic performances, with the latter two gaining acceptance.

The deal is reported as under consideration, with no confirmed closing within the article.

Industry concerns center on AI replacing jobs and potential unauthorized use of performers

work for training.

Interpositive is framed as augmenting, not replacing creativity.

Netflix's move sits within broader AI discussions about cost, copyright, and the creator-friendly

aim of using tools to aid filmmakers rather than replace them.

Netflix has publicly framed AI as a problem-solving tool that can empower artists and automate

labor-intensive tasks, while Netflix positions the move as creator-led and emphasizes ethical

use by training models on production footage rather than open data sets.

NVIDIA is unveiling Neumatron 3 Super, a 120 billion parameter open model optimized for

large-scale agentic AI, with 12 billion active parameters aimed at high reasoning and

efficiency for autonomous agents.

This release signals NVIDIA's plan to expand into enterprise and developer ecosystems,

aligning Neumatron 3 Super with its broader AI infrastructure strategy.

The open weights come with a permissive license and include full training methodology, more

than 10 trillion tokens of data, 15 reinforcement learning environments, evaluation recipes,

and fine-tuning support via the NEMO platform.

Valuation and sentiment place NVIDIA in a premium market position, with elevated multiples

and positive analyst sentiment, while momentum indicators show a neutral stance.

Analysts expect a strong outlook, reflected in high price to earnings, price to sales,

and price to book ratios, plus a strong recommendation score, with 67% institutional

ownership and some insider selling activity.

Neumatron 3 Super is being adopted across AI tooling, software development, life sciences,

and enterprise software to boost search, development accuracy, molecular analysis, and automation.

Enterprise deployments span platforms like AMD OCS, Palantir, Cadence, Disalt Systems,

and Siemens, targeting telecom, cybersecurity, semiconductor design, and manufacturing workflows.

A configurable reasoning budget lets developers tailor latency and compute, with modes from

full reasoning to capped budgets and low effort for faster responses.

The model employs a latent MOE technique that activates four experts for the cost of

one to improve accuracy and efficiency in token generation.

It uses a hybrid MOE architecture that combines transformer and Mamba components to balance

high reasoning with manageable compute costs.

A 1 million token context window enables zero re-reasoning in long workflows and reduces

latency.

Risks include elevated volatility and sector dynamics, but a high ALTMA and ZERO score

indicates low bankruptcy risk.

Potential catalysts include product launches and partnerships.

This has been your daily brief in AI.

To read more about these stories, follow the links in the episode bio.

You can also subscribe to these updates via email at www.brief.news.

For more daily podcasts about the topics you love, visit brief.news forward slash podcasts.

Tune in tomorrow.

We'll be back with everything you need to know.