Morning AI tech roundup: chips, models, and budget signals

A Techmeme item says a US official reported that TSMC will add $100 billion to build four more chip fabs in the U.S.

The figure raises the total pledge in the broader plan to $265 billion.

A related Techmeme report in the same batch lifts 2026 capex guidance to $60B-$64B from $52B-$56B.

That same update raises the revenue-growth outlook to 40%+ YoY, citing the AI megatrend as the driver.

Microsoft appears to be prepping its sales team to get more competitive with the other major players in the AI industry.

At an internal meeting on Tuesday, the company’s executives outlined a plan for salespeople to negatively compare AI products from companies like OpenAI, Google, and Anthropic to its own, according to a new report from Bloomberg.

“Everyone else is selling parts — we’re selling the full end-to-end system. That’s the story that we all need to get out there and tell in FY27,” Executive Vice President Jay Parikh reportedly told the room.

Copilot executive vice president Jacob Andreou reportedly went further, delivering a presentation comparing Copilot directly to Anthropic’s chatbot Claude. According to Bloomberg, Andreou noted that, when it came to performance within Microsoft’s office apps, Anthropic’s model was “slower and less accurate, and lacked the proper security integrations.”

TechCrunch has reached out to Microsoft and Anthropic for comment and will update this story if we hear from either outfit.

The post title on Hacker News is straightforward: 'Grok Build is open source.'

It points directly to the xAI GitHub repository.

The thread has 446 upvotes and 495 comments, which is a strong signal for the AI developer community.

The submission itself is evidence of growing demand for inspectable, modifiable model tooling.

Applied Computing , a London-based startup that’s building a foundation AI model for the oil, gas, and petrochemical industry, has raised a $20 million Series A led by engineering giant KBR, with Databricks Ventures participating.

Essentially, Applied Computing is pitching speed: It claims Orbital can flag anomalies, investigate what caused them, and model whether a proposed fix could create problems elsewhere in the facility, all within minutes. Adamson claims the product can compress investigations that previously took days or weeks into seconds, helping operators reduce energy use and maintain output.

That promise of speed seems to have found believers. The startup says it has gone from stealth to double-digit millions in annual recurring revenue in under 18 months. Adamson said Orbital is in use at some “large, publicly listed” upstream oil and gas, downstream refining and petrochemicals companies, although he declined to mention how many customers it has.

Its partners include Indian energy company Wipro, and KBR, which has integrated Orbital into its INSITE 3.0 digital platform for energy projects, and is using the product for ammonia production. Adamson said the startup is also working with a “major U.S. upstream operator” and plans to announce a partnership with a European oil major in the coming weeks.

Still, Applied Computing is entering a market that has entrenched industrial software suppliers, as well as more focused AI startups. AspenTech sells simulation and AI-powered modeling software for upstream, refining, and chemical operations, while AVEVA offers physics-based process simulation, optimization, and “what-if” modeling for industrial plants.

A post titled 'Running Gemma 4 26B at 5 tokens/sec on a 13-year-old Xeon with no GPU' reports concrete inference throughput.

The setup is explicitly CPU-only on older hardware.

HN reaction is active at 287 upvotes and 187 comments.

For teams constrained by budget or data locality, this post is a practical signal that useful model workloads can start without new accelerators.