I spent years predicting outcomes at Google, and at FutureSearch we've been accurately forecasting frontier lab outcomes since mid-2024. Now we ask: which lab will lead at the end of 2026?
Here I do a simple forecast. I rank each AI company on model quality, data, compute, talent, and R&D automation, mostly based on the state of affairs today but extrapolating to the near future, and then pick the overall winner from those rankings.
What I'm forecasting here isn't business objectives, but quality of the AI systems produced by the company, broadly construed.
Model Quality
Lab | Why |
|---|---|
Anthropic | Claude 4.5 Opus likely best overall |
Gemini 3.0 Pro, plus dominance in images & video | |
OpenAI | GPT-5.2-pro is strong on math, otherwise trailing |
xAI | Grok 4 & 4.1 underwhelming |
Meta | Llama 4 a bust, still waiting on "superintelligence" |
Anthropic, Google, and OpenAI are effectively tied, with Google's edge on non-text modalities likely less important Anthropic's edge on coding (see below on R&D automation). OpenAI could pull ahead again, as they did with o3 after they fell behind to Anthropic, but unlikely they will pull dominatingly ahead (see below on Talent). xAI has a chance but they have a lot of ground to cover, and Meta is a dark horse.
Data
Lab | Why |
|---|---|
Search, YouTube, Gmail & Docs, utterly dominant in data | |
Meta | Facebook, Instagram, WhatsApp comparable to Google data |
OpenAI | ChatGPT starting to produce highly relevant data |
xAI | Full X firehose, but Twitter data is narrow |
Anthropic | Likely not even ahead on code data |
Google's data advantage is structural and likely permanent. Meta's social graph and video data are underrated. The telling thing here is that Anthropic is leading on model quality while having by far the least access to data, which we take as pretty strong evidence that data matters less than people think.
Compute
Lab | Why |
|---|---|
Custom TPUs, massive internal capacity, no cloud dependency | |
xAI | Colossus cluster (100K H100s), expansion to 1M+ GPUs |
Meta | Huge GPU fleet, $40B+ capex planned for 2026 |
OpenAI | A bunch of partnerships inked, will they materialize? |
Anthropic | At least they are diversifying across clouds |
xAI's compute buildout is the story to watch. If Colossus delivers, they could eclipse Google in ability to train large models. (Is scaling dead though?) Meta is spending the money but unclear whether it'll cache out, since it didn't for Llama 4.
OpenAI and Anthropic are all using other people's compute, and the best they can hope for is to not fall further behind.
Talent
Lab | Why |
|---|---|
Anthropic | Highest talent overall, strongest mission, lowest attrition |
OpenAI | Would be leading if not for massive talent exodus |
DeepMind has many great researchers, but are they hungry? | |
Meta | Spending vast $$$, but is it working? |
xAI | Musk's pull; small elite focused team; uses $$$, not mission |
As we've written about before, OpenAI's talent hemorrhage is the most underappreciated risk. Anthropic is where the most brilliant and motivated people go. xAI might be able to attract more brilliant minds but probably the people there are only for the money, as with Meta.
Google doesn't have a culture of excellence and hard work, though they clearly have the raw talent if they could motivate people better.
R&D Automation
Lab | Why |
|---|---|
Anthropic | This is their plan. Claude Code is in fact increasing productivity. |
OpenAI | Using Codex but unclear how much it accelerates internal R&D. Claimed R&D automated in 2028. |
Long history of best-in-class dev tooling, but currently behind (why?) | |
Meta | Are Meta employees even using Meta's models for coding? |
xAI | No evidence of acceleration, but small focused team could pull it off |
Anthropic is explicitly premised around a coding-based R&D takeoff, and has been clearly delivering on that with Claude Code. OpenAI could do this too but are they as focused on it as Anthropic, with all their other businesses? xAI could get on board quickly.
Google is weirdly behind here, with anecdotes suggesting that they aren't even using agentic coding much, let alone with Google's unique dev tool advantages like codesearch.
All Rankings for the 2026 AI Race
Lab | Model | Data | Compute | Talent | R&D | Avg |
|---|---|---|---|---|---|---|
Anthropic | 1 | 5 | 5 | 1 | 1 | 2.6 |
2 | 1 | 1 | 3 | 3 | 2.0 | |
OpenAI | 3 | 3 | 4 | 2 | 2 | 2.8 |
xAI | 4 | 4 | 2 | 5 | 5 | 4.0 |
Meta | 5 | 2 | 3 | 4 | 4 | 3.6 |
1st
2nd
3rd
4th
5th
By average ranking, Google leads. But I weight Talent and R&D Automation more heavily, so my pick for #1 overall in AI, broadly construed, at the end of 2026 is Anthropic.
My view, and our view as co-authors of the AI 2027 timeline forecast, is it comes down to who builds the best AI R&D positive feedback loop. If agentic coding accelerates research velocity as much as Anthropic believes, this will be decisive. And I predict that data and compute problems won't slow Anthropic down that much, even though I think they're in last place on both fronts!
Google or OpenAI could pull it off too, so I give them perhaps a tied second place.
As of this writing, xAI just raised $20B, so this forecast could change quickly. Still, we like posting concrete forecasts, so we'll see at the end of 2026 how well this shapes up!