Baker cites research showing every 10x increase in training data doubles AI quality.
“and it's been very well established in multiple papers from both Google and Microsoft research that for every order of magnitude increase in the data you use to train an algorithm, the quality of the AI doubles.”
Baker states data quantity is the single most predictive element of AI quality, not algorithms or infrastructure.
“The single most predictive element of knowledge about AI quality is the quantity of data used to train the algorithm.”
Baker argues AI revolution stems from cloud computing power and mobile-generated data, not algorithmic advances.
“The only thing that has enabled the AI revolution that we're living through, which I think we're at the bottom of the first inning in, is one, we had the ability to do cloud computing, so just apply significantly more computational power to old algorithms, and then b, we had dramatically more data.”
Baker predicts Tesla FSD will achieve 100x improvement quickly as compute scales to GPT-4.5 levels.
“I think they're going to go really fast to GPT-4.5 compute, which means you're going to get, using these orders of magnitude, you're going get a 100x improvement really fast.”
Baker identifies three axes of AI scaling: pretraining, inference time compute, and now reasoning as the third multiplicative dimension.
“And then we started scaling around inference time compute. And it's very clear that we have now added a third axis of scaling performance, and that is reasoning.”
Baker predicts big businesses will pay a million dollars for AI to think for six weeks on key questions.
“I think you're if you're a big business and you can pay a million dollars to let an AI think for six weeks about the most important question for your business, that's gonna be a profound advantage relative to small businesses that can't afford that.”
Baker predicts inference compute will be the derivative winner and compute will run out in 2025.
“inference compute is gonna be the the kind of derivative winner of that. We're gonna run out of GPUs, accelerators, compute in 2025 the same way we did in '23.”
Baker states only NVIDIA and Amazon Trainium have functioning switched scale-up networks for inference today.
“And the only two functioning switched scale up networks in the world today are the ones that power NVIDIA GPUs and Amazon's Trainiums.”
Baker emphasizes only 0.1% of the world uses AI models properly yet there's massive shortage despite trillions spent.
“And we're in an insane shortage despite spending cumulatively trillions of dollars. What happens when 5% of the world's population is using these models the way the cutting edge 10 basis points are?”
Baker reveals Atreides could have invested over $50 million in CoreWeave at $1.1 billion valuation but was conflicted out.
“I could've Atreides could've invested over $50,000,000 in the round at 1,100,000,000, And I was conflicted out by Crusoe,”
Baker predicts Trainium will dominate 2026 like TPUs did in 2025, with Trainium 3 ramping in second half.
“Tranium is going to be to 2026, especially in the second half of this year when Tranium three really ramps, as TPUs were to twenty twenty five.”
“What happens when 5% of the world's population is using these models the way the cutting edge 10 basis points are? Like, it's just it's unimaginable. This is why orbital compute is a necessity.”
Baker argues Trainium is most underestimated because frontier mixture-of-expert models require switched scale-up networks for inference.
“And so Trainium is for sure the most underestimated, not only because of those design choices, but because the all of these frontier models are what are called mixture of expert models.”
Baker predicts Trainium will be to 2026 what TPUs were to 2025 as Trainium 3 ramps.
“Tranium, by far. Tranium is going to be to 2026, especially in the second half of this year when Tranium three really ramps, as TPUs were to twenty twenty five.”
Baker says AI models shifting to usage-based pricing with overage reveals no ceiling on spending yet.
“We're just moving from these all you can eat plans to usage based plans with overage, where those usage tokens cost a lot more, and we're finding out that there's we're nowhere near the amount of, you know, people ceiling price for how much they'll spend.”
Baker says understanding frontier AI now requires enterprise usage-based plans, not $250 monthly subscriptions which are rate-limited.
“To understand what Frontier AI is capable of today, even for a non coding use case, need to have Cloud Code or Codex five point Codex.”
Baker says AI shifting from flat pricing to usage-based is extremely bullish as people consume more AI.
“AI is just shifting from all you can eat to pay by the drink. Then it turns out people really like to talk to their friends long distance.”
Baker estimates Anthropic would be doing $100-150B ARR if not compute-constrained, versus current $50B.
“And I think maybe a true statement is that Infantropic could just wave a magic wand and get all the compute they wanted. They'd probably be doing well north of $100,000,000,000 today, maybe 150.”
Baker predicts OpenAI and Anthropic will exceed $200B ARR this year due to shift to usage-based pricing.
“So I think the shift to usage based pricing is probably why you will see OpenAI and Anthropic exceed well over $200,000,000,000 in ARR this year.”
Baker explains harness engineering matters significantly and harnesses are increasingly co-developed with models.
“And it turns out that harness engineering is not as important as the model, but it really matters. And these harnesses in these models are increasingly being co developed.”
Baker says understanding frontier AI now requires enterprise usage-based plans, not consumer subscriptions, due to rate limiting.
“To understand what Frontier AI is capable of today, even for a non coding use case, need to have Cloud Code or Codex five point Codex. And you need to be on an enterprise plan.”
Baker explains continual learning as models dynamically updating weights in real-time, unlike current reinforcement learning approaches.
“Continual learning is a model that dynamically adjusts its weights or adjusts in some way in real time.”
Baker argues America will consume all available compute, reducing edge AI bear case concerns.
“And I just think the same is true of compute. It's why I'm probably less worried about like an edge AI bear case than I was.”
Baker argues America will consume all available compute, making him less worried about edge AI bear cases.
“It's why I'm probably less worried about like an edge AI bear case than I was. We're going to consume as much compute as we can.”
Baker predicts GPUs will have 10-15 year useful lives due to prefill-inference disaggregation, extending older chips' value.
“The disaggregation of inference means that I think these GPUs are going to have ten or fifteen year lives.”
Baker argues GPU useful lives will extend to 10-15 years due to inference disaggregation, contradicting AI skeptics.
“The disaggregation of inference means that I think these GPUs are going to have ten or fifteen year lives. The AI skeptics are like, oh, these companies are all cooking their books.”
Baker argues GPU useful lives will extend to 10-15 years due to prefill/decode disaggregation, not 1-2 years.
“The useful life of GPU is only a year or two. The useful life of CPU is only four years because the rapid technological change.”
Baker predicts OpenAI and Anthropic will exceed $200B ARR this year due to usage-based pricing shift.
“I think the shift to usage based pricing is probably why you will see OpenAI and Anthropic exceed well over $200,000,000,000 in ARR this year.”
Baker believes Anthropic is likely already generating cash or will start this year.
“I think Anthropic probably starts generating cash this year if they are not already generating cash, which I think is probably the case.”
Baker calculates rendering Monopoly Go with VEO3 would cost over 100x the game's revenue.
“Monopoly Go is a game where we have the revenue and the hours played and it is to render it using list prices for something like VEO3 is more than two orders of magnitude greater than its revenue. So what is the role for human creativity, man? I don't know.”
Baker explains AI recomputes answers probabilistically each time, enabling superhuman capabilities but requiring extreme computational expense.
“AI, even if you put a harness on it, even if you do the chain of thought, even if you have multiple agents, it's probabilistic that it is recomputing the answer each time.”
Baker cites analysis showing compute margins, quantity, and inference margins all rising simultaneously, driving lab acceleration.
“The amount of compute is going up and inference margins going up. And if you multiply those three, that's how you're getting this crazy acceleration into some of the labs plus open source,”
Baker reports GPU rental prices doubled from mid-$2 to nearly $4 per hour over seven months for identical clusters.
“And they had rented a cluster of several thousand black wells, and we'll just call it somewhere in the mid $2 per GPU hour.”
Baker reports inference cloud expects to pay 100% more for Blackwells when contracts expire, showing hyperscalers are underearning.
“They went on a podcast, and they essentially said, we are planning to pay 100% more for Blackwell's when our contract expires. And that just means that essentially all the hyperscalers are under earning.”
Baker estimates SpaceX monetizes compute at $50B per gigawatt versus $73B consensus, with Grok and Cursor hitting $10B ARR quickly.
“And they're monetizing at something like 50,000,000,000 a gig and consensus estimates for next year are 73,000,000,000. So forget Starlink v three, forget Starlink direct to sell, Grok 4.5 and Cursor.”
Baker argues contracted compute trades at massive discount to spot, repricing will accelerate cash flows and answer ROI questions.
“And so, essentially, you have the contracted base of installed compute trading at a massive discount to the current spot market.”
“Really impressive from @cursor_ai The future is trending hybrid. Next iteration will likely be allowing customers to use their own proprietary data to post-train an open-source model. Excited for Grok 4.6. https://t.co/ZTE2VAskTG”
Baker announces interviewing Lin Qiao of Fireworks AI at Tomorrow X Summit November 17-18, praises specialized intelligence vision
“Super excited that I will be interviewing @lqiao CEO and founder of @FireworksAI_HQ at the @tomorrowxsummit in Austin, TX November 17-18. Lin’s vision of specialized intelligence is compelling and she is making it a reality. Fireworks is at the forefront of helping enterprises https://t.co/MK7CFgAsAU”
Baker says Grok 4.6 matches Fable 5 Max at 85% discount, 80% cheaper input and 88% cheaper output, Pareto dominant
“Grok 4.6 is roughly the same performance as Fable 5 Max at an 85% discount. 80% cheaper for input tokens and 88% cheaper for output tokens. Pareto dominant. Grok 4.7 will be significantly better as is a much larger model with the Cursor and SpaceX data included in pretraining. https://t.co/4e8CXbzs2D”
Baker says future is multiple models working together, lower model margins mean more AI infrastructure demand
“Important. The future is almost certain to be multiple models working together. Lower margins at the model layer whether from open-source or vertically integrated players = more AI infrastructure demand. https://t.co/YBWlCNnrDD”
Baker estimates his personal AI usage is up 100x and calls Grok Bot another Claude Code moment
“I think @bot is another “Claude Code” moment for AI. I would estimate my personal AI usage is up something like 100x. And for everyone who reached out about how to build a “podcast summarizer” it took me about 15 seconds in Grok Bot and is better than what I had before. https://t.co/AaoeHbEqtl”
Baker would not take under on 250 billion for Anthropic unless they stumble or fail to secure compute
“@altcap Love the expectations management! Let me say this: I would not take the “under” on $250 billion *unless* they stumble or fail to secure enough compute. Maybe you can discuss with @DarioAmodei at @TomorrowXSummit before I interview you about public policy!”
Baker shares GrokBot origin story and notes they lack sharing feature and iPad app but both coming soon
“The GrokBot origin story as told by @bot Sadly they lack both a sharing feature and an iPad app but I’ve been told they are both coming soon: https://t.co/4YaISohWW6”
Baker acknowledges having more sophisticated use cases than podcast summarizer but cannot post them
“@goldilock3din Laughing I do like my “podcast summarizer” There are much more sophisticated uses cases I promise but can’t post those here!!!”
Baker thinks at least two other labs believe they might be creating God
“@firesidealpha Funny and true. I do think at least two other labs think they might be creating God.”
“From yours truly on Friday’s @theallinpod “On the margin, Anthropic is losing share to OpenAI, to open source, and to Grok, and they're still growing so fast.” If the $65b is correct, then someone leaked it to help Anthropic on their forthcoming IPO roadshow. Basically, they https://t.co/oIYSS1PZhT”
Baker says Fireworks and Baseten are two largest inference clouds and first time seeing SpaceXAI on fastest growing list
“Fireworks and Baseten are the two largest inference clouds and I think of them as *an* index for open-source AI adoption. Also first time seeing SpaceXAI on the list of “fastest growing” IIRC. SambaNova! Will wonders never cease. https://t.co/HlmFsKRtVk”
Baker says Stripe is growing net revenue and FCF almost 2x faster than Adyen with declining share count
“Empirically impressive metrics from Stripe in the Axios report about their acquisition of Openrouter. Growing net revenue and FCF almost 2x faster than Adyen YTD with the share count declining over the last 3 years. I think Stripe will be a good steward of Openrouter and it is https://t.co/mnxU0XqP0N”
Baker suggests Anthropic may have shifted from gross to net ARR reporting making 65 billion more conservative than 47 billion
“Good chart from @TMTBreakout July accelerated both MoM and YoY for the sum of OpenAI and Anthropic. Some chance Anthropic shifted from gross to net reporting for ARR. Meaning the 65b is a more conservative metric than the $47. Smart thing to do before an IPO. Openrouter data https://t.co/FOwqXizXIL”
“@Shaughnessy119 @TMTBreakout 2 of the 3 parts of your equation are correct (first and last, not the middle). An open-source token costs just as much compute and energy to produce as a frontier token for a model of comparable size. And they are often less token efficient.”
“@ThainRos @TMTBreakout Yes. OpenAI drove the July acceleration.”
Baker says sum of OpenAI and Anthropic accelerated QTD with sharper August acceleration while Grok and open-source grow faster
“Yes. And the sum has accelerated slightly QTD with a sharper acceleration in August vs. July. And note that Grok Bot, Grok and Open Source are all likely growing faster than both of the two leading frontier labs. But at some point going to have to move past second derivatives https://t.co/L4r1aExjrI”
Baker explains Anthropic and OpenAI have more leverage over infrastructure providers than open-source does over cloud providers
“@Shaughnessy119 Y is higher than X - Anthropic and Open AI have more leverage over the infrastructure/cloud providers. For a given token over time all else equal, the cloud and infrastructure players charge more for open source tokens vs. what Anthropic and OpenAI pay per token b/c open source”
Baker agrees and thinks Poolside news is important indicator for future of American open-source
“@DavidSacks @sarthakgh Yes. Although I do think the Poolside news is an important indicator for the future of American open-source.”
Baker calls AFFN disaggregation the holy grail but notes it is a hard networking problem
“@AccBalanced @Atreidesmgmt AFFN disaggregation is the holy grail just such a hard networking problem.”
Baker says Atreides internal AI spend will be 100x higher in August 2026 versus March 2026 and still doubling monthly
“Suspect we will start to hear about a “Pareto optimal” balance of computationally efficient humans, cheaper open-source tokens and frontier tokens. Our internal AI spend @Atreidesmgmt will be roughly 100x higher in August 2026 vs. March 2026. Still roughly doubling every month.”
Baker explains open-source lowers compute spend by ratio of frontier to open-source margins because frontier marks up compute more
“@itsvlady Well it lowers compute spend by the ratio of the margins of the frontier tokens to the margins of the open-source tokens - frontier tokens mark up underlying compute dramatically more than open-source.”
Baker thinks Anthropic likely building in acceleration for roadshow and year-end but less true for OpenAI
“@jweil130 @vercel @rauchg Likely true of Anthropic IMO - likely building in an acceleration for roadshow and end of year. Probably less true for OpenAI.”
Baker says ratio of frontier inference margins to open-source margins will determine the outcome
“@JosephJacks_ The ratio of frontier inference margins to average open source inference margin will determine the result.”
“More data than open-source AI is taking share from OpenAI and Anthropic. Open source has gone from 28% token share to 62% token share @vercel over the last 2 months. Chart from @rauchg Super impressive given that the sum of OpenAI and Anthropic accelerated in July. So net https://t.co/wjDtlcl6ic”
Baker says big models are expensive and Opus 5 is a Mythos or Fable distillation doing well
“@F1NewsWire @TMTBreakout Big models are expensive. I don’t think super important one way or another. Opus 5 is a Mythos/Fable distillation and doing well.”
Baker argues cheaper open-source tokens are net positive for AI infrastructure because they cost same compute as frontier tokens
“@arakharazian Respectfully this is not correct from my perspective. Cheaper open-source tokens are net positive for the AI infrastructure names, although maybe not for the model companies. An open-source token broadly costs the same amount of compute (watts and wafers) as a frontier token to”
Baker says frontier labs claim their tokens use less compute than open-source tokens and have more intelligence density
“@JoshPurtell @arakharazian Frontier labs would say their tokens use less compute than comparable open-source tokens but that would break people’s brains. And for sure they have more intelligence density per token which compounds their efficiency advantage.”
Baker says future is multi-model with specialized open-source models on customer data working with frontier models behind custom router
“@IbrahimAjami Yes. The acceleration in OpenAI/Anthropic as open-source takes share is so interesting. The future is multi-model with specialized open-source models that were post-trained on customer data working together with frontier models behind a customer specific router. Firework Nexus”
Baker argues Ultrafast is flagship model from second largest frontier lab and OpenAI wants it as default mode
“@anni_sen @stefanveliki137 Hmm. This looks like the flagship big model from the second largest frontier lab. And IIRC @thsottiaux from OpenAI said they wanted this currently capacity limited Ultrafast mode to be the default mode 6 days ago. I’m not sure I would describe “default mode” for the flagship https://t.co/OQqwwW8CD9”
Baker says more interesting question is where teacher models stack up versus frontier and where they will be in months
“@firesidealpha This was a good podcast. The more interesting question is where those teacher models stack up vs. the frontier. And where the teacher models will be in a few months. Those ASICs sure were fun to develop!”
Baker calculates Ultrafast timeline from December deal through wafer and server production to August debut
“@DratchCap @anni_sen @stefanveliki137 @thsottiaux I am also curious about this. Along with why Ultrafast mode is capacity constrained. Deal struck in late December, 4.5 months to make a wafer, 6 weeks to make a server, a month to energize a server. So logical that Ultrafast debuted in August and would think it would be less”
Baker asks whether OpenAI or production constraints are limiting Ultrafast capacity
“@anni_sen @stefanveliki137 @thsottiaux You think OpenAI is constraining the capacity or production volume constraints elsewhere are constraining the capacity? Genuinely curious.”
“I literally have no idea why anyone is concerned about the FT data on Fable 5. Just shows customer preference for ZDR. Actually surprised so many are still using it after Opus 5 became available. https://t.co/u5Rzi13JmV https://t.co/GCqX5bMgYg”
Baker views Grok Bot as a great router and harness
“@iabdul_wasey Grok Bot feels like a great router and harness to me.”
“@StevanBoljevic @insane_analyst @IanCutress Yes also disaggregated inference is the future IMO.”
Baker has high conviction that large enterprises will use frontier harnesses for Pareto dominance while keeping data safe
“High conviction hypothesis especially for large enterprises. This is the best path to Pareto dominance while keeping data safe. SMBs and individuals probs just use the frontier harnesses. https://t.co/4petLDY4kt”
Baker argues open-source gaining share is net positive for everyone except model companies
“@arakharazian Markets do irrational stuff all the time! But open-source gaining share is net positive for everyone except for the model companies.”
Baker says real comparison will be Jalapeño versus Rubin Ultra and Feynmann given current shipping timelines
“Should note the real comparison will be with Rubin Ultra given Rubin shipping today so kinda have to wait. And Feynmann coming fast. But still impressive debut from Jalapeño.”
“@Stefan_Tee8 What is Maia is my take.”
Baker credits Jalapeño as first good ASIC outside TPU and Trainium but says disaggregated GPU plus SRAM will outperform
“Impressive that Jalapeño outperforms the comparable TPU and is in the mix with Rubin. Credit where credit is due - first good ASIC outside of TPU/Trainium. However, will likely significantly underperform a disaggregated GPU/Trainum plus SRAM accelerator setup. Especially with”
Baker notes Jalapeño participated in third-party benchmarks before TPUs did, suggesting competitive confidence
“Also pretty funny that Jalapeño participated in a third party benchmark *before* TPUs! Tells you a lot!”
Baker observes different chips hit different points on interactivity versus throughput tradeoff for model sizes
“@ChrisMurphyHQ Seem to hit different points of interactivity vs. throughput tradeoff for different size models. Basically everyone plays to their traditional strengths is early read. Really good AFFN disaggregation could change this.”