AI Roundup January 2023: Microsoft Bets $10B on OpenAI, ChatGPT Goes Vertical, and the Copyright Wars Begin

TL;DR

January 2023 was the month the ChatGPT shockwave turned into structural change. Microsoft committed a reported $10 billion to OpenAI in a multiyear deal, ChatGPT reportedly hit 100 million users to become the fastest-growing consumer app ever, and the legal system finally showed up: artists filed a class action against Stability AI, Midjourney, and DeviantArt, and Getty Images sued Stability separately. Google, meanwhile, dropped the MusicLM text-to-music paper and sat on the model.

Microsoft Commits a Reported $10 Billion to OpenAI

On January 23, Microsoft confirmed a new "multiyear, multibillion-dollar" investment in OpenAI. Microsoft did not put a number on it publicly, but the widely reported figure was $10 billion, and it marked the third phase of a partnership that started with $1 billion in 2019. Azure stays OpenAI's exclusive cloud, Microsoft gets to weave OpenAI models into its product line, and both sides keep the right to commercialize independently.

Why builders should care: this is the moment the frontier got expensive. Training and serving these models is a capital game now, not a clever-architecture game, and the deepest pockets just doubled down. The flip side is that Azure OpenAI Service became a real, enterprise-grade way to ship GPT-backed features without running your own cluster. If you build on hosted models, your roadmap now partly lives inside a Redmond partnership. That is great for availability and uptime, and a little uncomfortable if you care about not being locked to one vendor.

The frontier stopped being something a clever team could bootstrap. In January it became something you buy your way into.

ChatGPT Becomes the Fastest-Growing Consumer App in History

Two months after launch, ChatGPT reportedly reached 100 million monthly active users in January, per a widely cited UBS analysis. For scale: TikTok took roughly nine months to hit that mark and Instagram took about two and a half years. UBS analysts said they could not recall a faster ramp in any consumer internet app in twenty years of covering the space.

This is the number that explains everything else that happened this month. The Microsoft check, the lawsuits, Google's internal scramble: all of it is downstream of a chat box that went vertical. For people who build with this stuff, the signal is demand, not novelty. Normal users, not just AI nerds, now expect a conversational interface that can write, summarize, and reason. That expectation is the new baseline you are building against, and it showed up almost overnight.

The Generative-AI Copyright Wars Officially Begin

January is when the legal bill started arriving. Two cases landed within days of each other:

• Artists' class action (filed January 13). Sarah Andersen, Kelly McKernan, Karla Ortiz, and a proposed class sued Stability AI, Midjourney, and DeviantArt in the Northern District of California. The claims include copyright infringement, DMCA violations, and right-of-publicity violations, centered on the LAION dataset of roughly 5 billion scraped images used to train Stable Diffusion.
• Getty Images v. Stability AI (filed mid-January). Getty brought UK proceedings against Stability, alleging it copied and processed millions of Getty-owned images without a license to train Stable Diffusion. Getty's number: more than 12 million images scraped.

These suits would grind on for years and get trimmed in places, but the point for builders is the precedent risk that arrived this month. The legal question "is training on scraped, copyrighted data fair use" stopped being a forum-thread argument and became something courts will actually rule on. If you fine-tune or ship image models, your data provenance is now a liability surface, not a footnote. "We scraped the web" is not a license, and January 2023 is when the industry got served the reminder.

Google Quietly Publishes MusicLM (and Doesn't Ship It)

On January 26, Google researchers published MusicLM, a model that generates high-fidelity music from text prompts like "a calming violin melody backed by a distorted guitar riff." It produces 24 kHz audio that stays coherent over several minutes, can be steered by a hummed or whistled melody, and beat prior systems on both audio quality and prompt adherence. The team also released MusicCaps, a dataset of 5.5k expertly captioned music-text pairs.

Two things stand out. First, this was a genuinely strong audio result in a month dominated by text, and it showed the generative wave was coming for every modality, not just chat and images. Second, Google did not release the weights or a public demo, explicitly citing the same copyright and misappropriation concerns now playing out in the image lawsuits above. That gap, a lab publishing a capable model but declining to ship it, became a recurring theme. For the open-weight crowd, it was a clear sign that the most interesting audio models might stay locked behind research-paper glass for a while.

Key Takeaways

• The frontier got capital-intensive. Microsoft's reported $10B into OpenAI means competing at the very top is now a balance-sheet question, which raises the value of strong open-weight models for everyone who is not a hyperscaler.
• ChatGPT reset user expectations. Hitting 100M users in two months made conversational AI a baseline feature, not a differentiator, so plan your product around it.
• Data provenance is now legal exposure. The artist class action and Getty suit turned training-data sourcing into a real liability, especially if you fine-tune or ship image models.
• Every modality is in play. MusicLM showed text-to-music had arrived, even as Google held the weights back over the same copyright fears driving the lawsuits.
• Publish-but-withhold became a pattern. Capable models behind paper-only releases mean the open-source community will keep doing the heavy lifting to put these capabilities in builders' hands.