We just started now!
Ask @KC_Ryosuke anything about AI user experience
teapaw.org/i/spaces/1NGarobDrMdJj
Tonight 8pm Taipei (UTC+8): KC from the Glows team hosts our AMA on X Spaces, in Mandarin. "How much can one person actually get done with AI?"
Set a reminder: teapaw.org/i/spaces/1NGarobDrMdJj
Glows bills per second, and a released instance stops billing: docs.glows.ai/docs/my-instan…
Clover-35B: open-weight diffusion LM (35B total, ~3B active), Apache 2.0.
Vendor figure: 680 tok/s on one GB300, BF16, batch 1, internal agent workloads.
BF16 weights are ~69 GB, so start with one 80 GB card.
How much can one person actually get done with AI?
This week KC from the Glows team hosts our AMA. Bring your tools, your shortcuts and the questions you have been saving.
🗓 Thu, Oct 29 · 8pm UTC+8
Set a reminder 👇
teapaw.org/i/spaces/1NGarobDrMdJj
Compute forecasting is hard enough to be an S-1 risk factor.
Anthropic commits to compute "months or years in advance."
It has "significantly underestimated" its needs before.
Guess high and take-or-pay bills you anyway.
How far ahead do you commit?
sec.gov/Archives/edgar/data/…
Same GPU. Same rate card. Very different bills.
On many clouds, spot GPUs don't have one fixed price.
There's a floor, it can float up close to on-demand, and it can change often.
Same card, different start time, different bill.
What did your last job actually pay per hour?
Open weights are free. 66GB of VRAM isn't.
Nanbeige4.3-33B: dense, Apache 2.0, 512K context.
Rough math: 33B × 2 bytes (BF16) ≈ 66GB, weights only.
4-bit in 24GB? Community says so. Untested by us.
GPU type, quantity, duration? We'll quote.
huggingface.co/Nanbeige/Nanb…
H200 rent: ~+160% since January 1.
B200: ~+120%. H100: ~+35%.
Even the six-year-old A100 rents for more than it did in January.
Tell us the GPU type, quantity and duration, and we'll quote.
Booking early, or waiting it out?
a16z.news/p/charts-of-the-we…
Flapping Airplanes raised a $1B Series A, co-led by Index and Lux, on one bet: more useful work from every GPU.
Builders can make it too: match the card to the job, keep it busy, give it back when done.
RTX 4090 from $0.49/hr. Send GPU model, quantity and duration for a quote.
GPU rental prices went up again this week, per Silicon Data:
H100 +10.3% to $2.88/hr
H200 +6.0% to $3.51/hr
B200 +9.0% to $6.29/hr
B300 +5.8% to $7.35/hr
Waiting for prices to fall has been a losing bet all year.
Tell us GPU type, quantity and duration — we'll send a quote.
Palantir says customers are fine-tuning NVIDIA Nemotron on their own data and keeping control of both. McCarthy tuned Nemotron 3 Nano 4B on RFIs.
By our math, a 4B LoRA starts on one 24GB card.
RTX 4090 from $0.49/hr. Send GPU model, quantity and duration for a quote.
GPU rental prices aren't going down.
IREN's co-CEO on CNBC: 2026 is largely sold out; new 3-year deals at more than 2x last year.
That's for people buying years. Most teams need days.
4090 from $0.49/hr, H100 from $2.96/hr, billed per second. Tell us GPU, how many, how long.
DGX Spark is getting eight-system clusters; 64GB units roll out Oct 23.
Local or rented? Ask the workload.
Steady and private: local.
Bursty or on a deadline: rent.
RTX 4090 from $0.49/hr, H100 from $2.96/hr. Send GPU model, quantity and duration for a quote.
Ornn's B200 rental index rose 19% in the week to Oct 21; its H100 index fell ~4%.
The newest cards swing hardest, and most fine-tunes don't need them.
H100 80GB from $2.96/hr, A100 80GB from $1.20/hr. Send GPU model, quantity and duration for a quote.
When demand is "off the charts," big buyers book whole data centers. The team that needs 2 GPUs for a week gets squeezed.
Nebius just lined up $3.5B from Citi to build more. Capacity is coming.
The question is who gets it. Glows is for that small team.
Nvidia CEO Jensen Huang says neocloud demand is 'off the charts' but capital is scarce cnbc.com/2026/10/22/nvidia-j…
