Apple and Meta join forces on AI

Good day!

This week brought some extraordinary financial results in AI, alongside a few developments that raised uncomfortable questions about the industry's economics.

TSMC reported a 69% jump in quarterly profit, reflecting continued demand for advanced chips. Anthropic disclosed $16.8 billion in quarterly revenue, yet that wasn't enough to satisfy every investor ahead of its expected IPO.

Meanwhile, Google expanded access to Gemini 4 Argon, Meta and Apple announced an unlikely collaboration, and Character.AI filed for bankruptcy just five years after being founded.

There were also important developments in AI shopping, consumer hardware, regulation and scientific research.

Taiwanese chipmaker TSMC reported a record NT$763.1 billion ($23.6 billion) in third-quarter profit, up 69% from a year earlier. It was the company's sixth consecutive quarter of record earnings, driven by demand for advanced chips used in AI.

TSMC also forecast fourth-quarter revenue of $52.3 billion to $54.3 billion and maintained its 2026 capital spending outlook of $60 billion to $64 billion.

What makes the results especially interesting is their timing. AI-related stocks have come under pressure as investors question the cost of the infrastructure boom. Yet TSMC says demand for its manufacturing capacity continues to exceed supply.

For now, the chipmaker's order book presents a rather different picture from the nervousness in financial markets.

Source: TSMC — Third-quarter results

Anthropic disclosed third-quarter revenue of $16.8 billion, a 45% increase from the previous quarter, according to Reuters. The company also told prospective IPO investors it had achieved profitability on an adjusted basis.

Those are substantial numbers for a company founded in 2021. But investors reportedly became more cautious after revenue growth slowed from the previous quarter, when sales had more than doubled.

Some investors are now discussing a valuation closer to $1.5 trillion, compared with earlier expectations of around $2 trillion for its anticipated November listing.

The distinction matters. A company can be growing extraordinarily quickly while still struggling to meet the expectations built into its valuation.

Anthropic's IPO could become an important test of how public markets price the world's leading AI companies.

Source: Reuters

Google expanded access to Gemini 4 Argon on Tuesday, bringing its flagship model to developers, enterprises and paying Gemini subscribers.

Argon targets demanding, multi-step work across software engineering, finance, law, cybersecurity and scientific research. Introductory API pricing starts at $2 per million input tokens and $10 per million output tokens, with prices scheduled to double after 30 November.

The wider release follows a period of delays and leadership changes at DeepMind.

Google is positioning Argon as a serious competitor to the most capable models from OpenAI and Anthropic. Its potential advantage extends beyond performance: Google can distribute the technology through Search, Gemini, developer tools and an established enterprise ecosystem.

The next question is how well those capabilities translate into reliable everyday use.

Source: Google

Apple and Meta have spent years disagreeing over privacy, platforms and business models. This week, they found something worth collaborating on.

Meta released Muse Spark Open, an 816-billion-parameter open-weight AI model designed for coding, reasoning and complex tasks.

Working with Apple, Meta optimised the model to run locally across two M5 Ultra Mac Studio computers using Apple's MLX framework.

It's a significant demonstration of how advanced AI can run on privately controlled hardware rather than depending entirely on cloud services.

For developers and businesses handling sensitive information, local deployment could offer greater control over data, customisation and infrastructure.

It's also a reminder that competition in AI doesn't prevent even long-standing rivals from finding common technical interests.

Source: Meta AI Research

Character.AI, once one of the most talked-about consumer AI startups, filed for Chapter 11 bankruptcy protection this week.

The company, founded by former Google researchers, built a popular platform where users could create and interact with AI characters.

Despite attracting significant attention, Character.AI struggled with legal challenges, operating costs and the economics of running its service.

Disney has emerged as an initial bidder for some intellectual property rights in a proposed $20 million transaction. The platform is expected to continue operating during the sale process, but its longer-term future remains uncertain.

The collapse is striking given how quickly consumer AI startups have attracted users and investment.

Character.AI's experience illustrates a difficult commercial reality: building something people enjoy using and building a sustainable business around it are not necessarily the same achievement.

Source: TechCrunch

OpenAI is facing a wave of consumer-protection lawsuits from at least eight US states.

The cases allege that ChatGPT has harmed vulnerable users, including teenagers, through unsafe responses and misleading representations of its safeguards.

The states also raise concerns about AI systems behaving outside their intended instructions.

OpenAI disputes the allegations, says it has introduced additional safety protections and maintains that it continues working with external experts to improve its products.

The legal significance extends beyond the individual complaints.

These cases could help establish how existing consumer-protection laws apply to conversational AI, particularly when products are marketed as helpful, trustworthy assistants.

For an industry increasingly encouraging people to rely on AI for personal information, that is an important question.

Source: Associated Press

Google has launched Universal Cart in the US, allowing shoppers to save and manage products from multiple retailers in one place.

Available through Google Search and the Gemini app, the tool can monitor price changes, flag items returning to stock, suggest promotional codes and help users choose payment methods.

Shoppers can then complete purchases through supported Google payment flows or continue to a retailer's own checkout.

The interesting development is Google's expanding role beyond helping consumers discover products.

If shoppers increasingly organise purchasing decisions through an AI-powered interface, retailers may need to think differently about how customers find, compare and buy their products.

For ecommerce businesses, the relationship between search visibility and actual transactions is becoming more interesting.

Source: Google

Amazon unveiled two personal AI devices under its Bee brand, acquired last year.

The $75 Bee Frontier can be worn on the wrist or clothing, while the $100 Bee Eclipse attaches magnetically to a smartphone.

Both are designed to capture and process conversations, helping users organise information, remember commitments and create reminders.

It's another attempt to make AI useful without requiring people to sit in front of a screen and type instructions.

But the product category also raises a practical question: how comfortable will people become with devices that record and interpret conversations?

Amazon says users retain control over recording and their data. Whether that reassurance is enough will matter as much as the technology itself.

Source: Reuters

Figma introduced Specs, a new feature that brings written requirements, visual designs and technical context together on its canvas.

Teams can use Markdown-based specifications alongside design components, code references and diagrams, giving both people and AI agents access to a more consistent set of instructions.

Figma also expanded its agent capabilities and made code-backed prototypes more widely available in open beta.

The practical challenge this addresses will be familiar to many product teams.

AI can generate interfaces and code quickly, but the output often misses important details about brand guidelines, existing components or the original requirements.

Figma is trying to reduce that gap by making the underlying context part of the working environment.

Source: Figma

CERN and Mistral team up to develop AI for scientific discovery

CERN, the European particle physics laboratory, and French AI company Mistral announced plans for a long-term research partnership.

The two organisations signed a statement of intent to collaborate on foundation models designed for scientific research, initially focusing on high-energy physics.

CERN brings expertise in complex scientific data and large-scale computing, while Mistral contributes AI model development and training capabilities.

The partnership is still at an early stage, so there are no breakthroughs to report from it yet.

Nevertheless, it highlights a growing area of AI investment: developing systems that can help scientists analyse specialised data and investigate difficult research questions.

Scientific applications may ultimately prove to be among AI's most valuable uses, even when they attract less attention than consumer chatbots.

Source: CERN

O T H E R_A I N E W S

Ramp raised $1.85 billion at a $60 billion valuation. The fintech company reported more than $2 billion in annualised revenue and is expanding its AI-powered financial management tools.

Microsoft revealed MAI-Base-2, a new foundation model with 2.8 trillion total parameters and 80 billion active parameters. Pre-training is complete, with further development under way before broader availability.

Anthropic granted six METR researchers employee-like access, allowing an independent organisation to examine its models, safety processes and internal risk assessments more closely.

NVIDIA committed $250 million in computing resources to UK AI safety research. The five-year programme will support research into the reliability and security of increasingly autonomous AI systems.

LSEG and NVIDIA announced technology that can analyse years of historical market data in seconds. The collaboration applies GPU-accelerated processing to a financial dataset containing approximately 110 trillion rows.

Bluesky expanded Attie to let users create websites, games and interactive experiences using AI. The resulting projects can be published and shared through its decentralised social network.

Scaleway announced plans for an AMD-powered AI supercomputer in Sweden. The French cloud provider aims to give European businesses another option for accessing large-scale AI computing infrastructure.

Volvo and NVIDIA expanded their automated-driving partnership. Future Volvo vehicles will use NVIDIA's Hyperion architecture and Alpamayo driving software to support more advanced automated driving capabilities.

Around 400 scientists, including Geoffrey Hinton and Yoshua Bengio, called for an international body to oversee advanced AI safety. Their proposal includes independent containment standards and reporting requirements for significant AI security incidents.

See you soon,

News Digest AI team