Maybe when chip would infer at 10-20x the speed of a 5090 RTX, otherwise it's really..bad.
Beside the 35b, the others are too slow to do anything vs cloud models, unless you have a local B200+. And the Qwen 35b has the reasoning capacity of a 2yo doing maths.
I wasn't "impressed" with any models for coding until 1year ago, only then they started to be "usuable" with the fault that we know and still exists.
What new bike should I get? How should I rephrase this email? What to do about my persistent headaches? Recipes based on these ingredients?
It'll become easier and more affordable to distill/train/optimize smaller local models to meet your specific work needs. Multi-agent efforts will orchestrate the balance between using on-prem hardware for most tasks, and then defer to cloud models for more demanding/edge-case tasks. Especially if an organization can manage what data is shared with which model.
I also expect cloud providers to eventually start charging different rates for on-peak and off-peak usage, which would incentivize running what you can with the local compute you have otherwise sitting idle across dozens to hundreds of machines in your fleet.
I could be wrong and all your future hardware purchases will be glorified thin clients, with less local processing capabilities than today's laptops, servers, and towers.
In the mainframe's heyday, it was difficult to grasp a world in which compute became so dense and so cheap that it made sense for each individual to have all processing and memory on their desk, talk less of on their lap and in their hand.
Of course this time could be different. But, when we look at previous relentless progress in compute and efficiency, then also couple that with the breakthroughs AI itself unlocks, it's hard for me to imagine that centralized AI will meet both the scale and flexibility desired for all use cases.
Just as centralized compute never fully disappeared -- it took the form of client-server computing and the cloud, there's no question that centralized inference will continue to exist.
But it wouldn't surprise me if, ten years from now, 90% of global inference happens at the client.
While simultaneously, the 10% of inference that takes place in the datacenter is wildly lucrative and orders of magnitude greater than today.
Put simply: the inference pie may become so large that both client and datacenter win.
Beyond inference, even local HPC or large in-memory workloads. Steve Jobs dreamed that the NeXT Computer could be used to cure cancer in a dorm room (whatever tha tmeans), and I guess a bunch of Mac Studios chained together wouldn't be bad for a lot of computationally tough biomedical "basic science" applications.
Qwen flash-next is at least as good as Sonnet for my uses, and the speed is perfectly acceptable.
I really think the big labs are going to soon start feeling pricing pressure for everything smaller than Opus-tier models, and it will come for those next.
Only if you're comparing apples with oranges. There are no local LLMs at 128GB RAM which compete with Sonnet 5.5 intelligence. And the gap is a chasm when comparing against Opus 5.5, which is fast and cheap and plentiful on event he $20 subscription.
People saying this actually know how to run local models.
People claiming otherwise are Claude customers who know nothing beyond its point-and-click interface.
Consider that fact.
Does it actually? Sure, it is operating at huge scale and this design choice allows enormous models (presumably enormous MoE models). But is it achieving economies of scale?
Economies of scale are rooted more in the idea of a profitable thing becoming cheaper as the market grows, allowing cheaper manufacturing and further growing the market.
The big two or three LLMs are cheap because they are substantially subsidised by investor cash, and they are not profitable at all in any sustainable sense: the current pricing can only be sustained for as long as they can raise money, and the market will (and does!) shrink when it is asked to pay a less-subsidised price.
There appears to me to be no route to true profitability for this design strategy at all. Even the open weights router companies are downstream of largesse from state-affiliated companies.
No, it's exactly the opposite.
The big money is in small(ish) but finely tuned for specific tasks models.
Whoever invents an app store for LoRA adapters will win this AI race.
I don't think so. LLM inference enjoys huge economies of scale. At no point in the foreseeable future will a local model be able to compete with a cloud model in terms of price. Local models will have the privacy angle, but very few people will be willing to pay top dollar for that. There will be some people, for sure, but the vast majority will just use cloud models.