Today I saw one of the most talented engineers I know and worked with leave DeepMind and now I probably know part of the "why". Dark clouds hovering over Google's AI game.
It's not really clear what their gameplan is. From the outside, it looks like they're asleep at the wheel. Qwen/Deepseek/Kimi are crushing them from the cheap-and-open side, and they're not remotely competitive with Mythos/Sol or even plain-vanilla Opus on the "premium" tier.
Gemini does actually have its uses, but they're very very marginal and niche.
From day one everybody was saying that Google would eventually capture the AI market, but it looks more remote than ever. Maybe Hassabis' personal inclination towards AI-for-science, and physics/chemistry in particular -- as opposed to consumer AI and coding AI -- has hurt them commercially.
Is Gemini really doomed? I'm still bullish on Google:
1) they have more free cash flow and capital than God due to the ads business
2) they have data - intent from web searches, youtube videos, google books and music
3) they have dedicated inference hardware
for all these reasons, is being 6 months behind the frontier actually a structural, long term disadvantage? some day the pace of improvement will slow, and google will vacuum up the market. they'll be able to compete with open-weight models just on pure cost advantage from their vertical integration
That also own one of the 2 major mobile operating systems, with Gemini tightly integrated and all of the data they can gather from that. Why do you think OpenAI wants to do hardware? Owning delivery is going to be important, and right now, Google and Apple own the delivery mechanisms (to consumers).
They may not capture enterprise use, but I don't think they have to. That's only one piece of the market. AI that's useful to consumers will still get delivered via a smartphone, and Google is in a great place to capture that.
I also don't think LLMs have to be a "winner takes all" situation. Value isn't going to come from having direct access to a chatbot or selling API inference, value is going to be in the form of a specific product (for most, devs aside here). Something a consumer, or a non-tech business can buy off the shelf and plug and play. A "ready made" customer service agent system, a "ready made" BI platform using AI, etc.
For consumers, that's probably going to look like whatever is bundled and tightly integrated into their mobile OS of choice.
My prediction is that the EU will eventually bring in legislation that will force platform providers to provide pluggable APIs so that you can use whatever LLM you want.
> is being 6 months behind the frontier actually a structural, long term disadvantage?
Yeah, this is one of the things I find so weird on the discourse. If you get there negligeably later, but without astonishing spend and waste, you might even be better off in the long term.
I have thought that for a while and assumed that was apples approach to AI. Wait until everyone burns through investor cash, invents the better tech, and can monetize. Then copy that business model and polish it, or buy out the competition and polish. That’s usually apples move and it makes sense for google to do something similar.
"Doomed" no, but it's pretty clear that they just had a bad cycle and are struggling to keep up with the frontier.
Whether this happened because they bet on "world models -> better reasoning" and that bet didn't pay off, or failed a frontier run for technical reasons like OpenAI did with 4.5, or something else went down? We don't know.
Will they bleed talent, fall further behind until they give up, or clean the organizational and infrastructural cobwebs and get back in the saddle? We don't know.
So long as search revenue isn't correlated to AI revenue or threatened by AI revenue, Google at any point can bail on AI spending and suddenly the free cash flow machine is back on. Yeah they have a far larger debt load they now have to service but cash flow from search revenue is so strong it wouldn't be much of a blip.
I don’t think it matters that much for Google. They need to not fall hopelessly behind, but I don’t think there’s a strong economic reason for Google to burn the kind of capex that the frontier labs are burning. Strategically, I think they’re probably doing better than OpenAI and Anthropic. The Gemini models are open, and they are what researchers are working with (see neuronpedia as an example). Over time, this will give them a strategic advantage for the same reasons that open source wins over proprietary. Meanwhile, OpenAI and Anthropic have massive capex that needs to be returned to investors while their margins are being undercut by Kimi/Deepseek/Qwen. Google can wait around for the coming frontier lab profitability crisis and cruise right on by with their Apple contract and owned data centers to pick up the pieces and exceed the existing frontier labs.
A deal with a company Google has a share of, announced a week before their IPO, with very non-committal terms and ramp period protections delivered in one large block on short term notice priced likely at the high end of what Google charges for A4X instances anyway.
I don’t think this reflects desperation as much as strategy.
I dunno if it's a sound strategy that involves repeatedly telling investors [1] and employees [2] over multiple quarters that you are desperate for compute, including leaving a triple-digit billion backlog on the table [3], and then spending so much on CapEx that you have your first negative cash flow quarter ever and taking the inevitable hit to the stock [4], while turning away a large paying customer (who also happen to be a competitor) [5] ;-)
The RPO can be anywhere from 3 - 6 years, sure, but even on an annual basis that’s like a hundred billion now. It was already in the double-digit billions since before AI took off and has only been spiking since then, which tells us 1) it’s been huge for 3+ years, and 2) it’s still growing faster than they can collect it. This matches what all the other hyperscalers are doing.
My point is that an ulterior motive is not necessary to assume when all their actions and statements point to them being severely crunched for compute.
I mean sure, if they had a choice between say, CoreWeave and SpaceX, they’d choose the latter for the nice bump to SpaceX’s financials and their stake… but not just for that, not when it contributes to their cash flow turning negative and their own stock taking a hit.
It is very suspicious when they’re paying for GB300 more than they turn around and charge those out as a4x instances, it was announced a week before IPO and they have a 90 day exit option.
> The RPO can be anywhere from 3 - 6 years, sure, but even on an annual basis that’s like a hundred billion now.
OpenAI and Anthropic deals are mostly 5 year and Anthropic’s starts in 2027, accordingly these RPOs are sized for projected compute needs and run rate in 2027 not today.
The only way either lab could pay 1 year of RPOs today (~80B for anthropic and ~150B for OpenAI) is with a lot more debt or circular financing, the former of which is difficult in this market.
Everyone spending crazy money on capex right now says they have a crushing backlog and need more compute to protect their share price. I highly doubt the demand exists today at current prices if the big 3 hyperscalers magically had an extra 2-3GW of compute. It’s not like Anthropic and OpenAI are turning away customers offering to pay API pricing..
> Everyone spending crazy money on capex right now says they have a crushing backlog and need more compute to protect their share price. I highly doubt the demand exists today at current prices if the big 3 hyperscalers magically had an extra 2-3GW of compute.
I don't get this though: The theory is all these hyperscalers are simultaneously spending buttloads of money on CapEx to the extent it affects their stock price, and then they would lie about the demand to protect their share price. Why would they do all that when they could just do nothing and keep their firehoses of existing business revenue untouched and maintain their stock prices on the upward trajectory they already were -- like Apple?
> It’s not like Anthropic and OpenAI are turning away customers offering to pay API pricing..
We don't know, but clearly Anthropic has been struggling to keep Claude's 9's better than GitHub's 9's even after paying through the nose for capacity from competitors like SpaceX and Google.
I suspect OpenAI is managing only because Altman scrounged for compute like a madman way in advance, and most of its traffic is free users who can be arbitrarily bumped down to weaker models whenever compute is low. Whenever Anthropic does that Claude Code degrades and people complain.
Paying customer at what price? My understanding of Meta’s request is API pricing per token (of course discounted for volume) for content moderation etc. which is a lower margin product than Gemini enterprise seats. They also prioritize internal training runs.
This doesn’t mean more compute at any price is worthwhile. It also doesn’t mean that the SpaceX compute deal would even be offered to Meta.
> I don't get this though: The theory is all these hyperscalers are simultaneously spending buttloads of money on CapEx to the extent it affects their stock price, and then they would lie about the demand to protect their share price.
It’s not lying - it’s optimistic revenue projections. Your Sam Altman point is an example, these RPOs are real but what’s questionable is whether the AI labs can generate enough premium token API revenue to actually pay those commitments. Today’s OpenAI annualized revenue estimate is only 40B. Will they actually be able to 10x that to pay those RPOs? I’m skeptical especially with offloading inference to cheaper models.
> Why would they do all that when they could just do nothing and keep their firehoses of existing business revenue untouched and maintain their stock prices on the upward trajectory they already were -- like Apple?
The hyperscalers with proven revenue streams and strong financials (Amazon, MSFT, Google, arguably Meta) benefit from making the game more expensive than everyone, will get at least 50% of their capex back from this peak supply/demand mismatch and maybe other than AWS could easily use any excess compute for internal needs.
What do you mean? Google bought SpaceX shares when it was a tiny startup. They are probably more than 100X on their initial investment. Even with a 99% drop in SpaceX stock, Google would still be positive.
Edit. Google invested 900 million in 2015 for roughly 5% of the company which comes out to 71.5 Billion dollars at 1.45 Trillion dollar current valuation. That's an 80x increase.
I think Larry Page individually might also have a very large stake as well.
It doesn’t matter what price they bought SpaceX stock for. They could have gotten it for free. Any gain prior to Google’s investment could be realized prior to Google giving SpaceX even more money. What matters is the return on investment for the $12B/year into SpaceX.
In order get that back by “pumping” SpaceX stock, SpaceX market cap would have had to increase $200B based on that investment, and then Google would need to sell the stock.
It would be extremely foolish for an early investor like Alphabet to liquidate SpaceX holdings en masse all at once.
Also, for better or worse, SpaceX did increase by about 200B today.
For all it's faults, it is forming a massive military and telecom monopoly that is nearly unassailable. In the next few years, it can start directly competing with Verizon. The Ukrainian military and civilian population heavily relies on those satellites
> It would be extremely foolish for an early investor like Alphabet to liquidate SpaceX holdings en masse all at once.
I agree, which is just one more reason why the original reason you proposed for Google’s investment is very likely not correct. More likely is that Google got a good deal on compute from them. Normal business reasons.
> Also, for better or worse, SpaceX did increase by about 200B today.
The relevant question is how much it would have increased without Google’s investment of an additional $12B. If the answer is “more than $0” then the fact that it only increased $200B means it was a bad investment to pump the stock.
I'm confused. I think you're confusing Gemini and Gemma. Gemini is Google's frontier offering which is closed-weight, API-only, like most frontier models. Gemma is Google's open-weight offering focused on deployment on consumer and edge hardware.
"The company raised its full-year 2026 capex forecast to between $195 billion and $205 billion, with further significant increases planned for 2027." - Alphabet.
I think ~$200B is just for AI infrastructure capex. Fun fact: that's nearly what the 3rd largest military in the world (Russia) is spending on a land war in Europe.
- They're making a lot of money selling Tensor to Anthropic. If Nvidia's $4T market cap is justifiable, Google's position as one of the other top AI chip seller is worth a lot.
- In a world where open source Chinese models decimate Frontier models ability to charge a high price, it's the operators of efficient inference data centers that will win. Like Google
- Google is probably the biggest provider of "free" AI because it's on Google.com. That forces them to focus on cost. And in a commodity market, low-cost providers are the ones that make the money.
> everybody was saying that Google would eventually capture the AI market
my take on this is eventually the money is going to run out and there's going to be acquisitions and consolidation. I think that's when Google will come out on top.
I largely agree but I don't know if it's quite so clear cut. From the pricing angle, all competitors except Google, including Chinese models, have incentive to gain market share at all costs, and may be serving tokens at or below cost. I am not sure though, Google could certainly decrease prices if they wanted to.
For agentic work, but especially for web search, 3.6 Flash has an important leg up, its fast speed, that no other model comes close to matching. I guess nobody pays attention to it because it's not the one big flashy number that you compare to other models.
The default model they are using for Web search is getting capable and is very visible. And now it invites people to keep asking questions.
They are quietly trying to become the chatbot that everyone uses to look things up. That strikes me as an intelligent move - not everything has to be done by an expensive frontier model.
Google doesn't need to compete. 'everyone' is locked into them via the Gapps (mostly Gmail and Maps) and Android ecosystems. Same with Apple, and Microsoft on the B2B side. It's only Anthropic, OpenAI and everyone else that _need_ to compete because switching models is painless. And they have no other revenue streams.
10 years from now, its gonna be Google, Apple and Microsoft left standing in the AI game. Well, until the US wakes up and starts attempting to break the oligopoly like the EU has recently started to do.
They are not "asleep at the wheel"; it's just that the people in charge (the "MBA types") have no clue what to do!
There's an old saying, if you judge a fish's smarts by how well it can ride a bicycle, it will always seem dumb.
The people who have risen to the top of at Google are built for a different environment than what's needed right now. They are good at playing their political games, sabotaging each other, etc.; i.e. all of the petty games that managers play in big companies. But the AI era demands a different skill set: how to bring together incredibly smart people and forge them into a battle group that will achieve victory in the ongoing battle for AGI! It's as if you have built an army of tanks, but the next battle is being fought on the high seas.
Are they measurably attracting more talent here? It seems like they're losing some of it right now, so is there public info about numbers of researchers they have or etc
They also have the money, the hardware and the talent to be the best cloud infrastructure provider, yet they're still far behind AWS (for good reason, as anyone who's dealt with their customer service will understand).
Agree that I would (and do) still place my bet on them for the long term. Maybe the outcome will be a couple of good startups seeded and DeepMind _really_ focusing on LLMs now.
social media is a completely different market though - since there are massive returns to scale, it's incredibly hard for a new entrant to break in.
model training and inference is the opposite. people switch LLMs like people change clothes in the morning. there are popular services (openrouter) that make moving as easy as changing a model string.
Which poses the question: Why does Google even need to catch up? At least currently, the name of the game is integration. The actual model is a commodity.
Unfortunately the new cow is cannibalizing the old cow, so Google is in a bit of a bind here. (So far I cannot imagine them monetizing AI overviews enough to compensate for the sharp loss of ads on SERPs.)
To its credit Google seems willing to disrupt itself before its competitors can.
I said that too at the start of the year, but that's a looong time in "AI years". I feel like by now they should have announced a Fable-killer model. They may still do it but looking back I am becoming less convinced now than I was 6 months ago.
I think Google realized that it's more profitable to sell compute to AI labs than to make AI. They are not an ideologically driven company like Anthropic. They very much turned into a conventional company that just wants to protect the bottom line. This was evident even back when they had LAMDA and refused to release it.
German system supports wealthy individuals who are planning to stay in Germany long-term, but is horrendous for anyone without cash in the bank willing just to try something out.
For a split second there I believed there is a new Distill publication! Their articles were the most inspirational and eye-opening resource on my beginnings of ML journey, the quality of visualizations definitely made lasting impact on my mental models of _what is going on_.
My favourite goes definitely to The Building Blocks of Interpretability (https://distill.pub/2018/building-blocks/), those images landed in a lot of my university presentations and the dog made everyone immediately interested ;-)
I would be interested in studies into impact of left hemisphere importantce on the right hand usage, possibly the more sophisticated and "logical" usage of our hands pressured it as well.
It's true that the creative vs. logical side of the brain is mostly a myth.
But the hemispheres absolutely DO specialize in very predictable ways. Core language faculties are almost always handled by the left hemisphere, for instance.
Face processing is almost universally handled by the right hemisphere.
We know these things from people who have suffered an injury to one of their hemispheres. A person with damage to the right hemisphere has a chance of not being able to recognize faces, but that's almost never seen in an injury that exclusively effects the left-hemisphere.
For the longest time Iain McGilchrist has been going on about left brain this, right brain that and it all felt very pop-psych stuff.
Not sure if because of that being sort of torn down but recent years he has been clarifying he wasn't talking about a literal left/right device but more an analogy to different modes of thinking.
There is some hemisphere function allocation but it feels far to over played in folks trying to offer easy answers to difficult things.
Thanks! Although I understand there is still some specialization in each of the hempispheres, which could influence it, but I probably went too strong with my imagination here.
Left-handed people are often excluded from participating in MRI studies. To my personal dismay, as these studies often paid 25 euros per hour ~20 years ago, a significant sum for my student self that I could not partake in. It has however given me significant doubts about any strong lateralization claims...
The vibe coded software designed from the ground up to contain ads will be something regrettable. Will be like a doctor smoking cigarettes while prescribing opiates.
Algorithmically and seamlessly weaving undisclosed advertising (or other editorial content) into conversational output is their holy grail. It's the endgame. There's a reason they're pushing so hard.
I'd even walk back on just calling it advertising, because we immediately think of the usual ads we see everywhere. The actual thing here might be much more subtle and worrying, you could call it undisclosed influence.
Probably endgame plus getting "too big to fail" and getting gov't bailouts if things don't work out. It's part of the lobbying theme that LLMs are the next great power struggle.
