Weekly AI hardware review (2026-06-22)
Automated check from scripts/weekly-ai-hardware-check.mjs. Triage and either:
- Update
worker/src/ai-hardware.ts if a new datacenter chip shipped, redeploy the worker, then close this issue, OR
- Comment
noop and close if nothing actionable surfaced.
Current state of /api/ai-hardware
- lastUpdated:
2026-06-05 (17 days ago)
- Total chips tracked: 18
- Released within last 2 calendar years (2025 or 2026): 4
By manufacturer:
- NVIDIA: 8
- AWS: 3
- AMD: 2
- Google: 2
- Apple: 1
- Cerebras: 1
- Groq: 1
Recent chips:
| Year |
Mfr |
Name |
VRAM |
FP16 TFLOPS |
| 2025 |
NVIDIA |
NVIDIA GB200 (Blackwell) |
384 GB |
5,000 |
| 2025 |
NVIDIA |
NVIDIA B200 |
192 GB |
2,500 |
| 2025 |
NVIDIA |
NVIDIA RTX PRO 6000 Blackwell |
96 GB |
360 |
| 2025 |
NVIDIA |
NVIDIA RTX 5090 |
32 GB |
209 |
Chip-launch-flavored news, last 7 days
Matched 4 articles (keyword scan; many will be coverage of already-tracked chips).
| Date |
Source |
Title |
| 2026-06-22 |
WIRED AI |
OpenAI Launches Full-Scale Effort to Patch Open-Source Bugs as It Takes on Anthropic’s Mythos |
| 2026-06-22 |
NVIDIA AI Blog |
At ISC, JUPITER Shows What Exascale Science Looks Like |
| 2026-06-19 |
MIT Technology Review |
A startup claims it broke through a bottleneck that’s holding back LLMs |
| 2026-06-16 |
NVIDIA AI Blog |
Fastest, Largest, Strongest: NVIDIA Blackwell Sweeps MLPerf Training 6.0 |
Sources: NVIDIA AI Blog (2), WIRED AI (1), MIT Technology Review (1)
What "needs update" usually means
- A new datacenter chip shipped or was officially announced with public specs. Add an entry to
AI_HARDWARE_CATALOG.
- A tracked chip got corrected specs (sparsity adjustments, TDP errata, real-world memory bandwidth). Update the row.
- List price changed materially. Update
listPriceUSD.
- Bump
AI_HARDWARE_LAST_UPDATED to today.
What it usually does NOT mean
- Roadmap rumors or unofficial leaks (wait for vendor confirmation).
- Consumer GPUs (gaming RTX 5080/5090 etc) - those belong in
/gear when that ships.
- Cloud GPU rental price changes (those go in
/gpu-pricing).
- Specialty inference chips without published TFLOPS/VRAM specs (catalog requires real numbers).
After editing, redeploy the worker (cd worker && wrangler deploy) and bump last_reviewed for /ai-hardware in data/page-freshness.json.
Weekly AI hardware review (2026-06-22)
Automated check from
scripts/weekly-ai-hardware-check.mjs. Triage and either:worker/src/ai-hardware.tsif a new datacenter chip shipped, redeploy the worker, then close this issue, ORnoopand close if nothing actionable surfaced.Current state of
/api/ai-hardware2026-06-05(17 days ago)By manufacturer:
Recent chips:
Chip-launch-flavored news, last 7 days
Matched 4 articles (keyword scan; many will be coverage of already-tracked chips).
Sources: NVIDIA AI Blog (2), WIRED AI (1), MIT Technology Review (1)
What "needs update" usually means
AI_HARDWARE_CATALOG.listPriceUSD.AI_HARDWARE_LAST_UPDATEDto today.What it usually does NOT mean
/gearwhen that ships./gpu-pricing).After editing, redeploy the worker (
cd worker && wrangler deploy) and bumplast_reviewedfor/ai-hardwareindata/page-freshness.json.