Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡️ 

Here’s what we got for ya today:

• 👩🏻‍💻 GPT 4.5 is ready

• 🏭 $200 billion AI data center

• 🗣️ Text-to-speech by ElevenLabs

• 🦾 Humanoid robots in the home before 2025

Let’s get right into it!

Sign Up | Advertise | Tools

STARTUPS

GPT 4.5 is ready

GPT-4.5, OpenAI’s latest generative AI model, is larger and more compute-intensive than its predecessors, promising improved comprehension of user prompts. Available as a research preview for ChatGPT Pro subscribers at $200 per month, its release follows a year of AI innovation, including Anthropic’s hybrid reasoning model and DeepSeek’s budget-friendly yet powerful R1. OpenAI continues its strategy of scaling up models, believing a larger size reduces hallucinations and enhances interactions.

While benchmarks show mixed results, GPT-4.5 is expected to excel in writing and programming tasks. Unlike OpenAI’s “o” series reasoning models, this is the last non-chain-of-thought model before future iterations shift toward automatic model selection.

BIG TECH

$200 billion AI data center

Meta Platforms is in discussions to build a massive AI-focused data center campus, potentially costing over $200 billion, making it the largest in the company’s history and one of the biggest globally. This project, far exceeding the scale of Meta’s recently announced Louisiana data center, signals the company’s anticipation of a long-term surge in AI demand across its platforms.

Meta, alongside rivals like OpenAI, Google, and Microsoft, is ramping up AI infrastructure investments, with its capital expenditures set to rise nearly 70% this year to as much as $65 billion. Despite the emergence of cost-effective AI alternatives like DeepSeek, Meta remains committed to extensive AI expansion to keep pace with competitors, including OpenAI’s $500 billion joint venture with SoftBank. Concerns over the rapid development of xAI’s Memphis data center have further accelerated Meta’s urgency in scaling its AI infrastructure.

STARTUPS

Text-to-speech by ElevenLabs

ElevenLabs, an AI startup valued at $3.3 billion, has expanded beyond its core audio-generation capabilities by launching Scribe, its first stand-alone speech-to-text model. The model, supporting over 99 languages, boasts high accuracy, with 25 languages—including English, French, German, and Spanish—achieving a word error rate below 5%. In benchmark tests, Scribe outperformed Google Gemini 2.0 Flash and OpenAI’s Whisper Large V3.

The model features smart speaker diarization, word-level timestamps, and auto-tagging of sound events, making it useful for video subtitles and captions. Initially designed for pre-recorded audio, a real-time low-latency version is in development.

STARTUPS

Humanoid robots in the home before 2025

Figure 02 will begin alpha testing in residential settings in late 2025. This shift is driven by the company’s proprietary Vision-Language-Action (VLA) model, Helix, which enables faster task learning by processing visual and natural language inputs. After parting ways with OpenAI, Figure is focusing on its own AI models to enhance robotic capabilities.

While industrial applications, such as its pilot at a BMW plant, remain a priority due to their structured environments, the company sees long-term potential for home assistance, particularly in supporting aging populations. Despite challenges like varied home layouts and unpredictable conditions, Figure’s move signals growing interest in domestic robotics alongside competitors like Tesla and Apptronik.