Sunday roundup: from curl -w to Colossus
One post went up this week. The news cycle ran harder than the publishing schedule.
This was the last week before I shifted gears into a daily cadence. May 11 would mark the start of posting every day, but for now, Tuesday was still the publishing day. And the rest of the week belonged to an AI infrastructure deal that made everything else feel small by comparison.
Tuesday: TIL: Measuring API Response Times With curl -w
The only post this week was a TIL about curl’s -w flag, which prints timing variables after every request. I wrote it because I was spending a lot of time profiling API latency at work, and -w was the fastest way to get DNS, TCP, TLS, and response time from a single command.
Looking back at it, this post is a good snapshot of where I was in May 2026. Deepgram was a speech-to-text and text-to-speech API company, and latency was the thing everyone cared about. Single-digit millisecond improvements mattered. The -w flag let me isolate where the time went without pulling out a separate profiling tool. Quick, practical, no ceremony.
That post turned out to be the last of the monthly TIL series. By Monday the daily routine had started, and the relationship between publishing and my actual work changed.
What else happened this week
Tuesday was the quietest day of the week by far. By Wednesday the AI news cycle had lit up.
On May 6, Anthropic announced it had leased the entire compute capacity of SpaceX’s Colossus 1 data center in Memphis. Over 220,000 NVIDIA GPUs online, 300 megawatts of power. The deal immediately doubled Claude rate limits and killed peak-hour throttling. For anyone using Claude Code in long sessions, this was the practical difference between planning around limits and just working.
The Colossus deal was also a signal about where the industry was headed. Compute supply was becoming the binding constraint, and the companies that could secure it would pull ahead. Not on model quality, not on pricing, but on raw capacity. Anthropic had been public about hitting walls since late 2025. This was the fix.
OpenAI had a busy week too. New voice API models shipped with lower latency and better emotion handling, plus the quieter rollout of ads in free-tier ChatGPT. The voice AI market was visibly heating up. Specialists and generalists were both chasing the same latency targets.
Two reports landed covering AI venture funding. The numbers were enormous: $212 billion in 2025, up 85% year over year. Enterprise generative AI spending hit $37 billion, a 3.2x increase. The application layer captured $19 billion of that. These numbers are too big to act on directly, but the trend was clear: companies were past the pilot phase and buying real products.
One week, two speeds
Looking at the week as a whole, there was a disconnect between the publishing calendar and the industry. I posted one TIL about terminal profiling. The world moved on a $1.25 billion per month compute deal. That tension between practical daily work and industry-scale news is not unusual. Most weeks are like this. Most of the time you are working on one small piece of a much bigger machine.
May 11 would start the daily cadence and close the gap between what was happening and what was being written about. But this week, May 4-10, was the last week where that gap existed.
Frequently asked questions
What was the Anthropic-SpaceX Colossus deal?
Anthropic leased the full capacity of SpaceX’s Colossus 1 data center in Memphis, bringing 220,000 NVIDIA GPUs and 300+ megawatts of capacity online. The deal doubled Claude rate limits and eliminated peak-hour throttling. Reported terms were roughly $1.25 billion per month through 2029.
What happened with OpenAI’s voice API models that week?
OpenAI released three new voice API models with lower latency and improved emotion handling. The same week, ads appeared in free-tier ChatGPT. The voice API market was becoming contested between specialists and generalist providers.
How much did AI venture funding reach in 2025?
AI venture funding hit $212 billion in 2025, up 85% from $114 billion in 2024. Enterprise generative AI spending reached $37 billion, a 3.2x year-over-year increase. The application layer accounted for $19 billion of enterprise spending.
Was this the last weekly post before daily publishing started?
Yes. The post on May 5 was the final monthly TIL post. A daily publishing cadence began on May 11, starting with an AI industry analysis covering the same week’s events.