
Ubaid Dhiyan
432 posts

Ubaid Dhiyan
@UbaidDhiyan
Infrastructure Software M&A. Engineer turned Banker turned Entrepreneur. Dad. Reader.
Katılım Ağustos 2011
786 Takip Edilen59 Takipçiler
Sabitlenmiş Tweet

My 2024 Outlook on Generative AI and LLMs. As an investor, customer, or potential employee, there are five developing trends you should pay attention to. Check out my latest post for insights on key players and why they matter.
linkedin.com/pulse/2024-out… via @LinkedIn
English
Ubaid Dhiyan retweetledi

Many roughly know how a transformer works
To REALLY understand modern neural LMs—MoEs, GPU tiling, kernels, RLHF, data—you need CS336
By @tatsu_hashimoto, @percyliang
The 2026 edition appears on yt with ~2 weeks delay
youtube.com/playlist?list=…
Materials
cs336.stanford.edu


English

@Akamai @AnthropicAI 5/ I wrote about what Akamai’s recent moves say about the consolidation of enterprise browsers, the company’s evolving strategy, and why its positioning relative to
@Cloudflare is worth revisiting.
Here - tinyurl.com/nb3kf5he
English

4/ The timing is especially interesting alongside @Akamai's @AnthropicAI infrastructure deal. @Akamai is pushing further into AI compute while also moving closer to the enterprise user and workflow layer through security.
English

1/ @Akamai has been in the news for two very different reasons: a $1.8B, seven-year cloud infrastructure commitment with @AnthropicAI , and reported talks to acquire browser security company @LayerxSecurity for ~$250M.
English

@AnthropicAI in talks to acquire @StainlessAPI for ~$300M, reports @theinformation. Developer infrastructure around AI models, including SDKs, documentation and MCP tooling, will become increasingly important. Expect developer tooling M&A to accelerate.
English

@AnthropicAI in talks to acquire @StainlessAPI for ~$300M, reports @theinformation. Developer infra around AI models, including SDKs, documentation and MCP tooling, will become increasingly important. Expect developer tooling M&A to accelerate.
More here - bit.ly/4d4RZFU
English


@OpenAI's MRC announcement extends the AI infrastructure discussion from chips and interconnects into the protocol layer.
As training clusters scale, the network fabric becomes part of the performance, reliability and economics of frontier AI.
More here: udadvisory.co/intel/2026/arc…
English

The SAP modernization wave is becoming a useful lens for enterprise AI.
@tesseralabsai's $60M Series A points to AI moving beyond developer productivity into the services-heavy work of enterprise transformation.
More here: udadvisory.co/intel/2026/eve…
English

AI coding is usually discussed in the context of modern developer environments.
@nova_ai is using AI to reduce the cost, complexity and risk of changing legacy systems where business logic, implementation debt have accumulated over decades.
More here udadvisory.co/intel/2026/mor…
English

SAP announced two acquisitions that sit squarely at the intersection of enterprise data infrastructure and applied AI. More here - udadvisory.co/intel/2026/sap…
English

@adityaag I'd argue a bigger chunk of Series A/B companies are not even at the $10-20mm ARR level. There is a third path besides the ones you point to - it is to run an efficient M&A process that returns (multiples) of capital to stakeholders, preserves tech and soft-lands the team
English

The hardest spot in venture is to a Series A/B company that is not growing.
You are at 10-20M ARR, 20-50 employees but are growing sub 25%.
This setup is ngmi (not going to make it). You are not going to optimize and iterate your way out of that.
My provocative take here is that instead of trying to iterate here...you should return back to Minus One. Figure out the core assets you have and what you can build that might be a bigger shot on goal.
This will be very very hard. Frankly, I am not sure that many founders have the courage and fortitude to pull it off.
But it is worth trying.
Because the other path just leads to a slow decline and death.
And that is much more painful.

English

Ubaid Dhiyan retweetledi

Much of Dwarkesh's argument hinges on this statment which *was* accurate but will be increasingly inaccurate on a go forward basis imo:
“American labs port across accelerators constantly. Anthropic's models are run on GPUs, they're run on Trainium, they're run on TPUs. There are so many things you can do, from distilling to a model that's well fit for your chips.”
As system level architectures diverge (torus vs. switched scale-up topologies, memory hierarchies, networking primitives), true portability is eroding. The Mi300 and Mi325 had roughly the same scale-up domain size as Hopper while Blackwell’s scale-up domain is 9x larger than the Mi355 scale-up domain, etc.
Many frontier models are now being explicitly co-designed for inference on specific hardware like GB300 racks. Codex on Cerebras is another example. Those models run less efficiently on other systems and the performance differentials will only widen. A model that runs well on Google’s torus topology will run less efficiently on Nvidia’s switched scale-up topology and vice versa - the data traffic is fundamentally different as a byproduct of the models being parallelized across the different topologies.
Google’s internal teams - and increasingly the Anthropic teams as they become the most important customer of almost every cloud - have the luxury of operating across the stack (models, chips, networking) - but that is not the case for the rest of the market and other prospective users. Anthropic is the exception, not the rule. To wit, Anthropic and Google allegedly have a mutual understanding where Anthropic can hire the TPU engineers they need every year to ensure that they can continue to get the most out of the TPU.
Given the overwhelming importance of cost per token to the economics of the labs, models will be run where they run best. Most extremely large MoE models will run best on GB300s given the importance of having a switched scale-up network like NVLink for MoE inference. When training was the dominant cost for labs and power was broadly available, labs were optimizing to minimize capex dollars. Model portability was a way to create leverage over suppliers. I think that drove a lot of the focus on portability.
Today, inference costs as measured by tokens per watt per dollar are everything. Inference is way more important than training costs (inference is effectively now part of training via RL). Labs are therefore now optimizing for inference. This means increasing co-design and higher go-forward switching costs for individual models between systems. I do think this explains why Anthropic and Nvidia came together: Anthropic needed Blackwells and Rubins to inference at least *some* of their models economically. And Mythos might just end up being released coincident with the availability of Rubins for inference.
TLDR: as labs shift their focus from training to inference, the costs of portability and the upside of co-design to maximize tokens per watt per dollar both rise. Portability is likely to begin decreasing as a result.
I think what I might have respectfully added to Jensen’s answer is that systems evolve under local selective pressures.
The evolutionary pressure in America is a shortage of watts so it makes sense for Nvidia to optimize, as an American company, for power efficiency and tokens per watt and stay on copper as long as possible. China has a surfeit of watts. Chinese AI systems are already taking advantage of this with the Huawei Cloudmatrix 384 and Atlas SuperPoD having an optical scale-up domain that is much larger than anything offered by Nvidia today at the cost of *much* higher power consumption and much lower tokens per watt. The networking primitives for this Huawei system are very different than those for Nvidia’s systems and a model that runs well on Nvidia will not run well on that system and vice versa. This means that if a Chinese ecosystem gets momentum, Chinese models might stop running well on American hardware. And when Chinese models run best on American hardware, America is in a better position as this gives America a degree of leverage and control over Chinese AI that it risks losing to an all-Chinese alternative ecosystem.
This architectural fork makes porting and distillation less effective and strengthens the pro-American national security case for selling China deprecated GPUs imo.
Also I will attest that I did not wake up a loser this morning.
English

Can’t wait to join the team at @openai building codex. Would love to hear what you love about it or want changed. We’re moving fast. DMs open.
English

@pitdesi Of all the funny ARR numbers out there, this one has to be the funniest
English

cool- but I don’t understand it.
There are 100s of GLP factories that prescribe GLP’s after a few Q’s
Medvi’s flow is particularly bad.
I assume margins would have been competed away
Hims did $2.35B of revenue with 2k employees. This 2 person co in the space is doing $1.8B?
Sar Haribhakti@sarthakgh
.@eringriffith: "His start-up, Medvi, a telehealth provider of GLP-1 weight-loss drugs, got 300 customers in its first month. In its second month, it gained 1,000 more. In 2025, Medvi’s first full year in business, the company generated $401 million in sales. Mr. Gallagher then hired his only employee, his younger brother, Elliot. This year, they are on track to do $1.8 billion in sales." nytimes.com/2026/04/02/tec…
English
Ubaid Dhiyan retweetledi

