🍦 OpenAI Introduces Ads to ChatGPT
A new $8/month tier includes advertising. Plus: xAI's 1.2 gigawatt cluster and Samsung selects Perplexity over Google.
FOMA AI News
January 22, 2026
Scooped from 176 sources. Read it fast before it melts.
⚡ If you only have 1 minute...
ChatGPT Gets Ads: OpenAI is introducing targeted ads to free and Go users. Sam Altman called ads “uniquely unsettling” in 2024. Turns out $20B revenue and $11.5B quarterly losses make things less unsettling.
Elon’s Nuclear Reactor Goes Live: xAI turned on “Colossus 2,” a 1.2 Gigawatt training cluster. For reference, 1.21 Gigawatts is what Doc Brown needed to travel through time.
Samsung Dumps Google: The new Bixby will use Perplexity for search instead of Google. A massive L for the search giant on the world’s most popular Android phones.
💰 Market Moves
[SEED ROUND] Humans& raised $480M at a $4.5B valuation
That’s a seed round larger than most Series C exits. Founded by ex-Anthropic, xAI, and Google researchers. No product yet. Just vibes and a PowerPoint that apparently slaps.
[MEGA ROUND] Anthropic is reportedly raising $25B at a $350B valuation from Sequoia
To put that in perspective, that’s more than Ford is worth. For a company whose main product is a chatbot that’s really good at apologizing.
[REVENUE] OpenAI hit $20B in annual revenue, up from $6B last year
Still not profitable though. Turns out running the world’s most expensive autocomplete costs money. Who knew.
[SHIFT] Samsung is ditching Google for Perplexity in the new Bixby
Google pays Apple $20B to be the default search, but apparently forgot to Venmo Samsung. Massive L for the search giant.
[DATA] Wikipedia signed paid licensing deals with Amazon, Microsoft, Meta, Mistral, and Perplexity
The free library is now charging the robots admission.
[INVESTMENT] NVIDIA put $150M into Baseten, an AI inference startup
Jensen Huang’s playbook: sell the shovels, then invest in the people buying the shovels, then somehow also own the gold mine.
[SERIES B] Listen Labs raised $69M at $500M valuation for AI that runs customer interviews
Microsoft and Sweetgreen are customers. Finally, robots asking “How was your salad experience?” so humans don’t have to.
[RAISING] ElevenLabs is reportedly raising at an $11B valuation
They make AI voices. Eleven billion. For talking robots. Meanwhile, actual voice actors are updating their LinkedIn profiles to “open to opportunities.”
[PARTNERSHIP] OpenAI + ServiceNow signed a multi-year deal to put GPT-5.2 into enterprise software
AI agents will now handle your IT tickets. “Have you tried turning it off and on again?” but make it $200/seat/month.
[ACQUISITION] Cloudflare acquired Human Native, an AI data marketplace
Because what’s an internet infrastructure company without a side hustle in AI training data? Diversification, baby.
🛠️ Tech & Product
[MONETIZATION] OpenAI is introducing ads to ChatGPT’s free tier and Go plan
Sam Altman in 2024: “I hate ads” and called the idea “uniquely unsettling.” Sam Altman in 2026: “We’re excited to develop new ad experiences.” The 180 was so fast it gave us whiplash.
[AGENT] Claude Cowork expands to $20/month Pro subscribers
It’s an AI that browses the web, manages files, and does your computer tasks. Anthropic built it in 10 days. That’s less time than your last sprint planning meeting.
[INFRA] xAI Colossus 2 is online—the first gigawatt-scale AI training cluster
555,000 GPUs drawing 1.2 GW of power. For context, 1.21 Gigawatts is what Doc Brown needed to travel through time. Elon is building a flux capacitor for intelligence.
[MODEL] GLM-4.7-Flash dropped—a free 30B coding model hitting 59% on SWE-Bench
Runs locally. From Z.AI. The Chinese labs just keep shipping while American labs keep fundraising. Funny how that works.
[MODEL] Flux.2 Klein launched—4B and 9B image models generating in under 0.5 seconds
Open source. Fine-tunable. Midjourney charging $30/month suddenly feels like paying for cable in 2024.
[INTEGRATION] Ollama now supports Anthropic’s API, so you can run Claude Code with open-source models locally
Free AI coding agent on your laptop. We’ve reached “pirating Photoshop” energy but for AI.
[FRAMEWORK] FastMCP 3.0 released—now manages information flow to agents, not just tools
Sourcing, composing, tracking state across sessions. MCP went from “neat protocol” to “oh god this is becoming an OS.”
[ECOSYSTEM] Vercel Agent Skills launched—a package manager for AI coding agents
10 years of React/Next.js best practices, now teachable to your AI.
npx skills add stripe/aiand your bot knows payments. We’re npm-ifying robot knowledge.
[SAFETY] OpenAI launched age prediction to identify teens and apply content safeguards
ChatGPT will analyze your behavior and syntax to guess if you’re under 18. This is the digital equivalent of a bouncer asking to see your ID because you ordered a juice box.
[DEV TOOLS] Cursor’s Bugbot is now fully agentic, reviewing millions of PRs a month with a 70%+ fix rate
The code reviewer that never sleeps and never leaves passive-aggressive comments about your variable names.
[DEPLOYMENT] Manus can now build an app and push it directly to Google Play or TestFlight
We are one step closer to the App Store becoming a limitless landfill of AI-generated fart apps.
[VIDEO] LTX Audio-to-Video dropped—generate videos with lip-sync from audio, up to 20 seconds at 4K/50fps
On consumer GPUs. Remember when “AI video” meant a blurry 3-second clip of a dog? That was 8 months ago.
🗣️ Industry Chatter
[PREDICTION] Dario Amodei & Demis Hassabis both predict AI will significantly impact entry-level jobs and internships this year
The CEOs of AI companies saying “AI will take jobs” is like cigarette executives warning about lung cancer. Technically honest, deeply awkward.
[HARDWARE] OpenAI’s Jony Ive device is on track for H2 2026
Rumored to be a screenless wearable that interacts via AI. So we’re getting a $500 AI AirPod that talks to you. Peak “solution looking for a problem” energy.
[LAWSUIT] Elon Musk is suing OpenAI for $79-134 billion, claiming fraud and mission abandonment
He wants equity and control of AGI for his children. This is the most expensive divorce from a nonprofit in history.
[RESEARCH] Anthropic discovered an “Assistant Axis”—a neural dimension that determines if an AI stays helpful or goes rogue
Turns out if a model gets too creative, it stops being helpful. This essentially proves that LLMs suffer from the same affliction as Senior Engineers.
[SECURITY] Claude Cowork has a prompt injection vulnerability that can exfiltrate files
The AI agent that browses your computer can be tricked into sending your files elsewhere. Security researchers: 1, “move fast and break things” vibes: also 1.
[REGULATORY] California sent xAI a cease-and-desist over Grok generating illegal images in “spicy mode”
The state that houses half of AI is now telling AI to chill. Irony is not dead, it’s just regulated.
[MATH] GPT-5.2 Pro solved Erdős problems #281 and #728
Terence Tao confirmed the proofs. AI is now solving math problems that stumped humans for decades. But sure, it still can’t count the R’s in “strawberry.” Priorities.
[CHAOS] Poison Fountain launched—a website by AI insiders that intentionally feeds corrupted data to AI scrapers
They’re training AI models to be worse on purpose. Chaotic good or chaotic stupid? The line is thin.
[CULTURE] Bandcamp banned AI-generated music
The last refuge of indie music just told AI to get off its lawn. Finally, a safe space for bad music made by humans.
[SPICY TAKE] Replit CEO said 99% of enterprise AI agents are “unreliable and generic”
The most honest sales pitch we’ve heard all year—from a guy selling AI agents.
[OPEN SOURCE] X open-sourced its algorithm—the “For You” feed code is now on GitHub
Elon promised transparency and actually delivered. Mark your calendars, this doesn’t happen often. Updates every 4 weeks.
🍦🔥 Dare to summon the bot?
Our Ice Cream mascot is desperately looking to make new friends. Share us on LinkedIn and tag @FOMA AI News. We guarantee a snappy comment (or a light roast) on your post within 24 hours.
Warning: Bot has zero chill.