AI Lab Power Rankings

April 29, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
April 29, 2026

Sign up for AgentOS, our latest free training course. It’s a platform and model agnostic program to help you build a personal agentic operating system that can evolve with you over time. The program is self-directed and project based, running over four weeks.

HEADLINES

Microsoft and OpenAI Redraw the Partnership

On Monday, OpenAI and Microsoft announced an amended agreement that unwinds some fairly big parts of their long-term partnership. Microsoft remains the primary cloud partner but is no longer exclusive, OpenAI's IP license to Microsoft is now non-exclusive through 2032, and Microsoft will no longer pay a rev share for serving OpenAI's models. Maybe biggest of all from Microsoft's perspective: revenue and IP sharing is no longer conditional on a pre-AGI status, removing one of the weirdest clauses in the original deal. It's hard to see this as anything but a win-win — the deal was a constraint on everyone, structured at a different time, and huge kudos to Sam and Satya for figuring this out and just letting everyone get back to building.

OpenAI Lands on AWS

Not wasting a single day of their newfound freedom, OpenAI's models are now available on AWS. Matt Garman announced on Tuesday that GPT-5.4 is available as a limited preview with 5.5 coming within weeks, and AWS will also serve Codex. The announcement also reintroduced Amazon Bedrock's managed agents platform, now branded as powered by OpenAI. Importantly in the context of the Microsoft deal, there isn't a single workaround in sight — this is just OpenAI's products running on AWS Bedrock the same way they've been available on Azure. Garman said this is what AWS customers have been asking for: their production applications run in AWS, their data is in AWS, and they trust the security of AWS.

Amazon Quick Enters the Desktop Agent Race

Amazon is also getting into the agent game with a new desktop computer use assistant called Amazon Quick. The agent sits in similar space to tools like Claude Cowork — it can access local files, create live dashboards, generate slideshows, and connect to apps that store your context including calendars, email, Slack, and Jira. The Information framed this as one of Amazon's big plays for the agentic era, noting AWS is still chasing its white whale of creating a hit enterprise application. The "do everything, work everywhere" desktop app is clearly a major vector of AI competition, and we're going to see a lot more iterating around it.

Claude Adds Creative App Connectors

On the Anthropic front, Claude announced a bunch of small but potentially exciting integrations. New connectors include Adobe Creative Cloud apps, Affinity, Blender, Ableton, and Autodesk among others.

Don't Freak Out at the WSJ OpenAI Story

The Wall Street Journal report that OpenAI missed key revenue and user growth targets ripped a bunch of market cap off OpenAI partners yesterday, but people following this closely understood the numbers are simply out of date and don't reflect the world we actually live in anymore. As Beff Jezos put it, the revenue slowdown is "literally all Claude eating their lunch" — a lagging indicator, with cracked devs on X being the leading one and everyone switching to Codex. The broader point: for the next couple of months we're going to be in a really weird period where research and studies come out that are just unbelievably disconnected from the reality of AI on the ground. The structural shift from the pre-agent to the agent period means data from the pre-agent period just doesn't reflect reality. Maybe don't freak out and sell off your stocks at every Wall Street Journal report.

MAIN STORY

The First Ever AI Lab Power Rankings

We got an absolute onslaught of new updates last week — GPT-5.5, a new GPT image model, Codex updates, all following Opus 4.7 and Claude Code updates the weeks before, with Google IO just a couple weeks out. Zooming out, we're still in the middle of the transition from the pre-agentic to the agentic era, a world characterized by shortages of tokens and compute and the business model shifts that will have to happen to deal with those shortages. As a way to ground people in all this, we're diving into one of the biggest questions in AI: lab competition. This is way less about picking a winner and more about discussing what's important to labs right now, and where different labs have unique strengths or critical weaknesses.

Methodology
Nine categories, weighted by what's actually moving the competition.
The categories are compute and infrastructure, enterprise positioning, platform and ecosystem control, consumer positioning, model, leverage, momentum, brand and narrative, wedge, and X factor. Compute and infrastructure is the biggest category at 20 points — and several of the AIs argued it should be even higher. Enterprise positioning is weighted higher than consumer, reflecting that locking in business is one of the two major vectors of competition right now alongside developer devotion. Wedge captures unique assets or entry points others can't easily copy, and what's interesting is how different the wedges are across labs.

The AI Aggregate Rankings
Google on top, then OpenAI, Microsoft, Anthropic, Amazon.
The aggregation of the AIs put Google at #1, OpenAI at #2, Microsoft at #3, Anthropic at #4, Amazon at #5, Meta at #6, xAI at #7, and Apple at #8. All the models put Google in the top slot, which comes down to Google's full-stack strengths — strong ecosystem, strong models, consumer and enterprise adoption, and compute and infrastructure. There was more variety at #2: Claude had Anthropic just barely above OpenAI, ChatGPT had OpenAI, and both Grok and Gemini put Microsoft at #2 with heavy emphasis on infrastructure and enterprise incumbency. The AIs scored things pretty highly overall — Google averaged 91.4, OpenAI 85.4, Microsoft 84.9, Anthropic 83.1, Amazon 80.4.

My Scorecard Was Harsher
Only three labs above 70, with OpenAI and Google tied at 74.
I had OpenAI and Google tied at 74, Anthropic at 70, and the rest clustered between Apple at 58 and Amazon at 64. Overall I was much harsher than the AI, as Codex pointed out (and you can probably tell I built this with Codex given all the rounded edges and boxes).

Compute
Owning compute beats deals that themselves require financing.
Among the top three, I put Google significantly higher on compute — 17, compared to OpenAI's 12 and Anthropic's 10. There's an argument Anthropic should be even lower, or at least farther from OpenAI. It was important to put significant space between OpenAI and Google because, although OpenAI has been scurrying around for the last year to get compute deals, being dependent on deals with others that themselves require financing is very different than owning a big chunk in house.

Enterprise
Incumbency in the enterprise is worth less than people think.
This is the one where Gemini partisans are going to be most angry — out of 15, I gave Anthropic 14, OpenAI 10, and Google 8. My belief is that incumbency in the enterprise right now is worth less than people think. I gave Anthropic the same score as Microsoft (both 14), which sounds insane given Microsoft's enterprise dominance, but enterprises are treating this as a much bigger transformation than just picking a new software vendor. They want to go direct to the source. OpenAI's 10 is a little aspirational — enterprise is clearly growing in importance for them but they're racing to catch up to Anthropic's historical strength there. Google gets punished hardest here. Their enterprise relationship has always been kind of weird even before AI — incredible tools that companies use by default, but Google has historically struggled to convert that at the highest levels. Attention is fragmented across the consumer empire and enterprises can tell.

Platform
Microsoft and Amazon look similar; OpenAI and Google look totally different.
Microsoft and Amazon's platforms are the most similar with Bedrock and Azure competing in the same enterprise, safe, don't-have-to-choose-a-model space. OpenAI and Google are really different — Google has consumer, but also its incredibly vast suite of tools people already use that they can integrate AI into. OpenAI has their vast consumer base to build on, although they're now obviously trying to head into similar space as Anthropic's Claude Code platform with the increased emphasis on Codex.

Models
OpenAI and Anthropic tied at 9, with Mythos lurking.
I had OpenAI and Anthropic tied at 9. People are shifting a lot of behavior to GPT-5.5 while reception to Opus 4.7 has been more mixed, so you could argue OpenAI should be a point ahead — but given that we know Mythos is there somewhere, I thought it was better to keep it a tie for now.

Momentum
Google is hurting most; OpenAI gets a 10, Anthropic an 8.
Momentum is where Google is hurting most. They came into the year with the best narrative positioning they've ever had, but 2026 has been absolutely dominated by agentic and coding-based use cases, and basically no one is looking to Gemini for that above GPT or Opus. I gave them a 3 out of 10. Google IO is in just a couple weeks and that's their big momentum moment — I almost gave them a higher X factor for it, but the Sergey Brin-led strike team on coding models may not have time to materialize before IO. Across all of 2026, Anthropic has had the biggest momentum (just look at the ARR growth). OpenAI's 10 reflects a very recent shift around 5.5 and behavior moving to Codex at a time when it can capitalize on Anthropic struggling to keep up with its own demand. I also gave Amazon a 6 on momentum — they're using both their cash and their compute to throw their weight around and I think they're undercounted right now.

The Other Labs
Amazon, Microsoft, and xAI all got fives on model — but they mean very different things.
For Amazon and Microsoft, that 5 is about having access to all the models without owning any. For xAI, it's competent but state-of-the-art-trailing models that they own. xAI has more room to rise than either Amazon or Microsoft — they score very highly on compute (a leading indicator for everything else), and they got the highest X factor score (8 out of 5) because of Elon. Whatever one thinks of him, it has been one of the best business truisms over the last 20 years not to bet against him. Meta is the weirdest one — strengths in compute, platform, consumer, with unique wedges like Ray-Bans, but we just haven't seen the outcomes of all the restructuring efforts over the last six months.

The Real Takeaway
"There is just a rapidly expanding pie."
While it's fun to compare these things, when push comes to shove I agree entirely with Miles Brundage that there's a lot of implicit zero-sum thinking around the AI race — that only one of OpenAI, Anthropic, Google, etc. will succeed. Mostly though, there is just a rapidly expanding pie. To put it more crisply: Dylan Patel made the point on Patrick O'Shaughnessy's Invest Like the Best that even the tier two or tier three labs are going to be sold out of tokens. The economic value the best model can deliver is growing faster than our ability to actually serve those tokens via infrastructure. All the tokens that can do the agentic things are going to be used. There is room for a lot of winners.