Jozef Soja
@JozefARK
Research Analyst @ARKInvest focusing on AI, cloud computing, and SaaS. Disclosure:
128 Following 2K Followers
Last week OpenAI released Astra, and the initial stats look insane. It all but maxed out the ARC-AGI-3 reasoning benchmark with a score of 99.9%. 5.6 Sol and Opus 5 scores were in the 30s. Progress is happening incredibly quickly.
Show more
Really incredible to see Meta making such a comeback in the LLM space. Llama 4 was released less than a year and a half ago, and now they're pushing the Pareto frontier forward. Very excited to see how they price the max version.
Show more
Muse Spark 1.3 (xhigh) in Muse Code sits on the Pareto frontier for Coding Agent Index vs. cost per task at $1.72. Muse Spark 1.3 (max) is excluded from cost comparisons as Meta has not announced pricing for the limited release, but on tokens it averages ~14M total tokens per task, around a third fewer than Claude Opus 5 (xhigh) in Claude Code (~22M) at the same score.
Show more
Public estimates suggest that revenues are continuing to grow at both Anthropic and OpenAI. Both are expected to turn to public markets in the near future to fund the next leg of their growth. Read more below, and check out @MattarARK's initial impressions of Grok 4.6, too!
Show more
Welcome to the team, @mattarARK! Definitely give him a follow for tips and tricks on the practical use of AI.
First ever X post!
Recently joined @ARKInvest as a research associate, focused on AI, cloud, and semiconductors.
As someone hands on with the latest tools, I’ll be sharing what I’m learning across the AI stack, different use cases, and some tips and tricks along the way.
Stay tuned for anything related to compute and infrastructure to models, applications, and the companies building them.
Lots to learn, lots to share. Excited to get going!
Show more
On Cloudflare's earnings last week, @eastdakota said non-human internet traffic could outnumber human internet traffic by 1000:1 in 5 years. This creates a massive monetization opportunity as the internet is fundamentally reimagined. Read more below:
Show more
Last week, @OpenAI launched GPT 5.6 and @SpaceXAI launched Grok 4.5. Both models are meaningfully more performant than the previous generations, but the real headline is how efficient they are, especially as companies begin to evaluate how much they spend on AI.
Show more
As AI labs like Anthropic and OpenAI continue to push the frontier of performance, government regulation is becoming a more meaningful headwind to adoption. Read more on Fable/Mythos, SpaceX, and more in this week's newsletter:
Show more
Last week, @AnthropicAI closed a round at nearly a $1 trillion valuation and released Opus 4.8. The newest model is more cost effective than the previous generation, which is starting to become more important as CTOs and CFOs begin to evaluate their spend on AI.
Show more
A lot of the conversation around AI is rightly focused on boosting productivity and automating knowledge work. But it also holds the potential to accelerate new discoveries in math and science. Last week, an @OpenAI model autonomously solved an open question in mathematics:
Show more
The platform-as-a-service layer is seeing growth from AI as tooling and data become increasingly important. Read our investment thesis for Databricks below:
Enterprise technology tends to create a new platform layer that becomes essential to operations. Mainframes had operating systems. The internet era had databases and middleware.
The AI era demands something different: a unified data and intelligence platform that can govern, process, and learn from an entire organization's information estate.
ARK believes Databricks is building that platform.
A few numbers that stand out:
• $2.6B in fiscal year 2025 revenue
• By early 2026, $5.4B+ annual revenue run-rate, with 65%+ year-over-year growth
• Net revenue retention above 140%, compared to Snowflake's 126%, meaning existing customers expand dramatically
• 800 customers spending over $1M annually, 70 spending over $10M
• Structured Query Language (SQL) business alone targeting $1B run-rate; AI product portfolio already there
We believe the architecture is the competitive moat. Databricks' "lakehouse" architecture collapses historically separate systems, data warehouses, data lakes, and machine learning (ML) environments, into one open platform. As enterprises move from AI experimentation to production-grade deployment, the platform governing the underlying data captures disproportionate value.
Open-source gravity around Delta Lake and MLflow anchors developer workflows in Databricks-originated standards, extending its relevance beyond any single product cycle.
The consumption flywheel is already turning. More workloads. More governed datasets. More inference. More value.
Read ARK’s full Investment Thesis ⬇️
Show more
Capex from the major hyperscalers continues to grow, but so does their revenue backlogs as demand outpaces supply. Read our thoughts on last week's earnings and the implications for AI:
Really interesting to see the price to rent H100s flatten out, and in the case of 1yr contracts, even start to tick back up. Lines up well with the inflection in usage we've seen for AI models over the past quarter. Data from @SemiAnalysis_
Show more
When @AnthropicAI revealed the capabilities of the unreleased Mythos model, it resurfaced concerns about AI's impact on cybersecurity. We think AI will continue to be a tailwind to cybersecurity companies- read more about why in this week's newsletter:
Show more
