Welcome to 2026! I’ve been out sick with the flu for most of the past week, which means this wasn’t done in time for 2025. Hope you enjoy!
It’s been another year of relentless AI news - though this time, the headlines felt less like novelty and more like a reckoning. After fears of an “AI plateau” emerged in late 2024, 2025 answered with a barrage of competing models, eye-popping valuations, and the first real signs that artificial intelligence was reshaping jobs, geopolitics, and the tech industry’s power structure.
Looking back, a few themes dominated: the end of American AI exceptionalism, the scramble for talent and chips, and the growing tension between breakneck innovation and calls for oversight. Some stories played out over months; others landed like bombs. Together, they paint a picture of an industry in flux - simultaneously more capable, more competitive, and more uncertain than ever.
10. DeepSeek
In January, Chinese startup DeepSeek released its R1 reasoning model, demonstrating performance comparable to the best American AI while spending a fraction of the resources.
The release sent shockwaves through Silicon Valley, (arguably) caused Nvidia stock to plummet, and forced a fundamental reassessment of assumptions about AI development requiring massive capital. DeepSeek continued making waves throughout the year with R1-0528 in May, maintaining competitive performance despite export controls.
More than any other single release, DeepSeek changed the narrative: the AI frontier was no longer an American monopoly. More importantly, it challenged the entire premise of the AI arms race - that billions upon billions of capital would be needed to build and run the state of the art models.
9. OpenAI’s Ups and Downs
For better or worse, OpenAI’s story often shapes the narrative of the whole industry - and the company had a rollercoaster of a year. On the upside: a $40B raise at a $300B valuation (the largest private fundraise ever), completing its conversion to a public benefit corporation, and hitting $1B in monthly revenue by August. ChatGPT remains synonymous with “AI” in the minds of many, and it was the most-downloaded app of 2025.
But the year also brought challenges. GPT-5’s August launch was underwhelming, and OpenAI faced real competition for the first time from all directions - DeepSeek, Anthropic, Google, and xAI all took turns at the top of the leaderboards.
Still, the company seems well-positioned heading into 2026, even if its dominance is no longer assured.
8. AI Talent Wars
Competition for AI talent was already fierce, but this year took it to another level. In some cases, talent was being poached: Mark Zuckerberg’s “Superintelligence” lab reportedly offered $100M+ packages, and Tencent paid ~$66M for a single OpenAI researcher. Meta attempted to acqui-hire entire startups - Thinking Machines Lab, Perplexity, Runway - before ultimately freezing hiring in August after months of runaway spending.
In other cases, the talent left to build their own empires: Mira Murati raised $2B at a $10B valuation for Thinking Machines Lab; Yann LeCun departed Meta after clashing with Zuckerberg over LLMs vs. “world models”; Intel’s CTO defected to OpenAI. Even Nvidia got in on the action, “licensing” Groq for $20B in what was effectively an acqui-hire. The war revealed both AI’s perceived value and the industry’s dependence on a surprisingly small pool of elite researchers.
7. Licensing Deals
One of the stories I feel like I haven’t focused on enough this year is the evolution of licensing deals in the space. After years of lawsuits and copyright fights - which, to be clear, are still ongoing - we’ve seen bigger and more diverse licensing deals start to settle the landscape.
On the music side, Suno cut deals with major labels that legitimized AI-generated music in a big way. On the text side, Anthropic reached a multibillion-dollar settlement with book publishers whose work it had trained on, perhaps setting a template for other AI companies to follow. And perhaps most interesting of all, OpenAI and Disney struck a $1B deal that will finally bring legally-approved, AI-generated Disney characters to Sora - a watershed moment for AI in creative industries.
6. AI Comes for the Juniors
What had been theoretical became real in 2025. People have been worried about AI coming for white-collar jobs for two years now, and for the first time, it looks like it might be showing up in the data. To be clear, there’s still a lot of conflicting evidence around this - some studies show productivity gains without job losses, while others paint a grimmer picture.
But I’m most compelled by the research showing that job growth for entry-level programmers has collapsed over the last two years, even as openings for senior roles have continued to grow. The generation entering the workforce is facing a fundamentally different landscape than the one that existed just two years ago.
5. Vibe Coding
It’s only been ten months since Andrej Karpathy coined the term “vibe coding,” and yet it ended up as the Collins English Dictionary Word of the Year. In some sense, not surprising - it’s hard to escape coding models if you’re a programmer these days.
Multi-billion dollar industries have spawned from whole cloth in the past year or so. Cursor hit $1B in ARR by year’s end and turned down repeated acquisition offers. Claude Code has become a major revenue driver for Anthropic.
More importantly, I think coding agents have pretty solidly demonstrated the first real incarnation of the modern “AI agent” - autonomous systems that can take on complex, multi-step tasks with minimal human intervention. There’s some debate over whether the current form and tools of vibe coding will persist, but it’s undeniable that the fundamental task of having an LLM generate code is here to stay.
4. Chip Wars
This isn’t a new story, but it gained new importance in 2025. Biden’s AI Diffusion Rule imposed strict export controls in January; Trump rescinded it in May, then brokered a deal allowing H200 sales to China with a 25% revenue cut to the US government. The geopolitics of AI chips became increasingly tangled with the broader US-China relationship.
China, for its part, is working to wean itself off of its Nvidia dependence. In September, Chinese tech giants were banned from buying Nvidia chips entirely, and elsewhere in the year Huawei unveiled chips that it claimed matched Nvidia’s performance. Whether that’s true remains to be seen - but the direction is clear. The chip war has become inseparable from the AI race itself.
3. Model Mayhem
Despite fears of an “AI plateau” emerging in late 2024, reasoning models got dramatically better in 2025 - and not just from OpenAI. All three leading labs spent the year jockeying for the “state of the art” crown, with leadership changing hands almost monthly.
A few of the models that had the “state of the art” label when they were released this year: DeepSeek R1, Grok 3, Gemini 2.5 Pro, Claude 4 Sonnet, GPT-5, Claude 4.5 Sonnet, GPT-5.1, Gemini 3 Pro, Claude 4.5 Opus, and GPT-5.2.
But more than just the competition, we saw benchmark after benchmark toppled this year. It feels almost like old news at this point, but it’s still so remarkable that these models are now better than the vast majority of human programmers and mathematicians - at least when it comes to coding challenges and math olympiads.
2. Infrastructure Investment
A near-pervasive drumbeat this year was infrastructure spending - it’s hard to pick a single roundup to link to, because nearly every week brought new announcements of multi-billion dollar projects. There’s Project Stargate, with $500B in planned spending, but there’s also Meta’s planned $600B in US infrastructure through 2028, not to mention dozens of “smaller” deals from other tech giants.
Perhaps then, it shouldn’t be surprising that there’s growing backlash to the AI infrastructure buildout. Small towns across America have started rejecting data center proposals over water consumption, noise pollution, and rising energy costs - in some cases facing lawsuits from developers for their trouble.
The year exposed a tension at the heart of the AI boom: the industry’s ambitions require physical resources that communities and markets aren’t always willing to provide.
1. The Bubble Question
Is AI in a bubble? The question haunted 2025.
In March, Alibaba’s Joe Tsai warned of data center overbuilding. Throughout the rest of the year, we continued to see various reports that AI spending was becoming difficult to justify from an ROI perspective - one MIT study found that 95% of GenAI pilots at companies showed little to no financial impact. Even leading AI CEOs like Sam Altman and Sundar Pichai have admitted that valuations have become detached from reality.
There’s been a slight correction - AI stocks shed ~$800B in a single November week. But there’s also the revenue numbers to contend with. This isn’t the dot-com bubble where no real money was changing hands. OpenAI and Anthropic are reaching double-digit billions in ARR, and AI startups like Cursor are growing from zero to a billion in a handful of months.
Yes, there is irrational exuberance. But it’s still wrong to dismiss the entire AI technology wave entirely.
What’s on your top ten list? What stories do you think went underreported this year? Let me know in the comments!
Thanks for putting this together! I love this format and I might steal it tbh.
I agree completely about the rights issue being buried under all the (admittedly impressive) other headlines this year, but it's really really REALLY important. That's a non-technical barrier that has now moved somewhat to the side, IMO.
Nice recap, and I hope for no more flus this year!