Live deals Have held stock, project excess or a cancelled order? Sell it

Posts are shown as published on LinkedIn on the date given. Availability changes quickly: for what we can offer today, see Live deals.

in
Looking for H200? Read this first. We have 100 AMD MI350P GPU servers in stock today. πŸ₯Š

In Rocky III, the champ got comfortable and a hungry challenger took the title. Our next fight looks a lot like that.

In the green corner: an NVIDIA HGX H200 8-GPU server. In the red corner: a Supermicro server with 10Γ— AMD Instinct MI350P.

Judges' scorecard:
Round 1, price per GPU: MI350P. About 30% less than an 8-GPU H200 SXM server.
Round 2, GPUs per server: MI350P. 10 against 8, in a 5U chassis.
Round 3, GPU memory per server: MI350P. 1.44TB against 1.13TB, about 25% more in one box.
Round 4, FP8: MI350P. 2.3 PFLOPS per GPU against 1.98.
Round 5, BF16: MI350P. 1.15 PFLOPS against 0.99.
Round 6, FP4: MI350P. 4.6 PFLOPS. The H200 has no FP4.
Round 7, lead time: MI350P. About 3 weeks after compliance. New H200 orders are being quoted at 25 to 30 weeks.
Round 8, memory bandwidth: H200. 4.8TB/s against 4.0.
Round 9, FP64: H200. 67 TFLOPS against 36.
Round 10, software: draw. PyTorch, vLLM and SGLang run on AMD's ROCm today. CUDA still leads for TensorRT-LLM and custom kernels.
Round 11, GPU-to-GPU link: H200. NVLink is still the champ's best punch.

Seven rounds to three, one drawn. If your stack runs on PyTorch, vLLM or SGLang, the challenger wins on price, memory per rack and time to deploy.

🚨 One lot: 100 servers (1,000 GPUs) in stock. Once they're gone, it's back to factory lead times.

If H200 was on your list, enquire today. Message us or email contact@blade-ai.co.uk with the quantity you need.

Sold as complete servers only. Subject to prior sale and end-user compliance.

#Rocky #AMD #MI350P #NVIDIA #H200 #AIInfrastructure
Image from the Blade AI LinkedIn post of 12 November 2026
in
🚨 IN STOCK: NVIDIA HGX B300 8-GPU servers, available immediately ⚑

Brand-new HGX B300 systems with a 2TB memory build are in stock with the OEM πŸ“¦ Delivery is 2–4 weeks from KYC completion, shipping globally 🌍

Most B300 orders are still quoting months ⏳

🟒 8Γ— NVIDIA B300, 288GB HBM3e per GPU
🧠 2TB DDR5-6400 (16 Γ— 128GB) per system
βš™οΈ 2Γ— Intel Xeon 6 processors
πŸ’Ύ 15.36TB NVMe E1.S + 2Γ— 960GB NVMe boot
🌐 8Γ— 800G OSFP networking
❄️ Air-cooled 10U, no liquid cooling needed

βœ… Brand new, factory fresh
🚚 2–4 week delivery from KYC
🌎 Global shipping

πŸ”₯ When they’re gone, they’re gone.

πŸ‘‰ Enquire now: blade-ai.co.uk/contact
πŸ“© contact@blade-ai.co.uk

Subject to prior sale, KYC, export control and NVIDIA end-user compliance.

#NVIDIA #B300 #HGXB300 #BlackwellUltra #Blackwell #GPU #GPUServers #AIInfrastructure #DataCenter #AIServers #Neocloud #LLM #GenerativeAI #HPC #CloudComputing
Image from the Blade AI LinkedIn post of 12 November 2026
in
πŸ”₯ US stock: NVIDIA HGX B300 8-GPU servers, available immediately.

πŸš€ New HGX B300 systems with a 4TB memory build are in stock in the US and ready to ship, subject to KYC.

⏱️ While most B300 orders are quoting months, these can be racked this quarter.

β–ͺ️ 8Γ— NVIDIA B300, 288GB HBM3e per GPU
β–ͺ️ 4TB DDR5 (32 Γ— 128GB) per system
β–ͺ️ 2Γ— Intel Xeon 6776P, 128 cores total
β–ͺ️ 30.72TB Gen5 NVMe Β· ConnectX-8 Β· BlueField-3 DPU

US based stock, for US-based buyers with US deployment sites only.

πŸ‘‰ Enquire now: blade-ai.co.uk/contact

Subject to prior sale, KYC and NVIDIA end-user compliance.

#NVIDIA #B300 #Blackwell #AIInfrastructure #DataCenter #readystock #neocloud #GPU #AIInfrastructure #AIGPU #DataCenter #GenerativeAI #AIComputing #HPC #GPUServers #Procurement
Image from the Blade AI LinkedIn post of 10 November 2026
in
B300 vs B200!!!! Went the distance. B200 Took the title!!!!!!!!!!!

In the first fight, Rocky went fifteen rounds with Apollo and lost on points. In the rematch, he took the title.

Our B200 vs B300 fight ends like the rematch: the challenger wins.

Judges' scorecard:
Round 1, price: B200. About USD 550,000 against USD 650,000 and up.
Round 2, lead time: B200. About 1 week from stock against around 36 weeks.
Round 3, FP8 and BF16: draw. Same throughput on NVIDIA's own figures.
Round 4, FP64: B200. 296 TFLOPS against 10.
Round 5, memory: B300. 288GB per GPU against 180GB.
Round 6, FP4: B300. 1.5x the dense throughput.

Three rounds to two, one drawn. If your model fits in 180GB per GPU and you are not FP4-only, the B200 takes it on points, and it is the one that can ship before Christmas.

Supermicro NVIDIA HGX B200 8-GPU servers, from cancelled orders, available from a single unit. The full comparison: https://blade-ai.co.uk

Enquire now. Subject to prior sale.

#Rocky #B200 #B300 #NVIDIA #Blackwell #AIInfrastructure
Image from the Blade AI LinkedIn post of 6 November 2026
in
Welcome to the party, pal.

Every December the same debate comes back: is Die Hard a Christmas film? We say yes. But the most famous line needs an update for 2026.

Now I have a GPU. Ho-ho-ho.

While new Blackwell builds are quoting around 30 weeks, we have GPU servers ready to go before Christmas:

- ASUS HGX B300 8-GPU servers: 24 units
- Supermicro HGX B300 8-GPU servers
- NVIDIA HGX B200 8-GPU servers, built and already in Europe
- NVIDIA RTX PRO 6000 Blackwell 8-GPU servers
- NVIDIA networking in UK stock: switches, ConnectX NICs and optics

No crawling through air vents. No waiting until next summer. Tell us the configuration and quantity, and we'll come back with availability, lead time and price.

See what's live: blade-ai.co.uk/live-deals
NVIDIA networking in UK stock: https://lnkd.in/esBaPsjU

Enquire now. Subject to prior sale.

Yippee-ki-yay.

#DieHard #Christmas #GPU #NVIDIA #AIInfrastructure #DataCentre #H200 #B200 #B300 #server #hardware
Image from the Blade AI LinkedIn post of 3 November 2026
in
Need Mellanox / Nvidia Networking parts???

NVIDIA is quoting up to 52 weeks on ConnectX cards and optics.

We have 500+ units sitting in a UK warehouse right now!!

Brand new, factory sealed, ready to ship!!

UK Stock list: https://lnkd.in/efS3Yuz4

βœ… ConnectX-7: 400G NDR, 200G, 100G and 25/50G (138 units)
βœ… ConnectX-6 Dx / Lx: 200G, 100G and 25G (140 units)
βœ… 800G twin-port OSFP transceivers, plus 400G and 100G (77 units)
βœ… Spectrum-3 SN4700 400GbE and SN4600 100GbE switches (15 units)
βœ… NDR copper, splitter and MPO fibre cables

If your build is waiting on networking, message us today. Stock like this moves in days.

πŸ“© contact@blade-ai.co.uk
🌐 blade-ai.co.uk/contact

#NVIDIA #InfiniBand #AIInfrastructure #DataCentre #GPU #Mellanox
Image from the Blade AI LinkedIn post of 3 November 2026
in
64 Γ— NVIDIA H200 nodes. Ready in 2 weeks. πŸ‡ΊπŸ‡Έ

New H200 servers are getting hard to find. OEMs are closing their HGX H200 lines as Blackwell takes over, and most new orders now join a 30-week queue.

This lot doesn't.

64 brand-new Dell PowerEdge XE9680 servers, in the USA, ready in about two weeks. Each node:

β†’ 8Γ— NVIDIA H200 SXM, 141 GB HBM3e each (1.1 TB GPU memory per node)
β†’ 2Γ— Intel Xeon Platinum 8580
β†’ 2 TB DDR5
β†’ 8Γ— ConnectX-7 NDR 400G InfiniBand
β†’ 1Γ— BlueField-3 B3220 DPU
β†’ 2Γ— 3.84 TB NVMe
β†’ 3-year Dell ProSupport, next business day

That's 512 H200 GPUs and 72 TB of HBM3e in one lot. Enough for a full training cluster, or to add capacity to the one you already run.

It will move fast. First PO wins, and we work with compliance-checked buyers only (end-user and export checks up front, so delivery isn't held up).

If you need H200 capacity before the end of the year, comment "H200", DM us, or email contact@blade-ai.co.uk

#NVIDIA #H200 #Dell #PowerEdge #AIInfrastructure #GPUServers #DataCentre #BladeAI
Image 1 from the Blade AI LinkedIn postImage 2 from the Blade AI LinkedIn post
in
Your GPUs arrived. Your optics didn't.

The GPUs are racked, but the network isn't. That's where a lot of AI builds are stalling right now.

Some NVIDIA optics are being quoted at 50+ weeks. Industry estimates have 800G transceiver supply falling 40–60% short of demand through 2027.

We supply the parts that hold racks up:
β†’ 800G switches
β†’ ConnectX-7 / ConnectX-8 NICs
β†’ 800G and 1.6T OSFP optics
β†’ cables

New, factory sealed, NVIDIA original. 7–10 working days from PO.

Send us your network BOM and you'll get price and lead time back within one working day.

contact@blade-ai.co.uk Β· blade-ai.co.uk

#AIInfrastructure #DataCentre #NVIDIA #Networking #GPU
Image from the Blade AI LinkedIn post of 29 October 2026
in
Waiting got expensive.

Renting a B200 now costs 22% more than a year ago. B300 rental is up 14% in just 90 days. Server memory rose up to 18% in a single quarter.

Every month you wait for an allocation costs more.

The fastest way out is to buy the servers someone else cancelled: new, factory-sealed Supermicro HGX B200 and B300 systems, here in weeks, not the 30+ weeks of a fresh allocation.

Comment "LIST" or DM us for this week's stock. The first PO wins.

#NVIDIA #B300 #B200 #GPU #AIInfrastructure
Image from the Blade AI LinkedIn post of 29 October 2026
in
Grant approved. Cluster booked up.

Your funding is in, but the shared GPU cluster has a queue measured in days or weeks, and your students are the last to get time.

A DGX Station GB300 gives your lab its own AI supercomputer:
β†’ 748 GB of coherent memory
β†’ models up to a trillion parameters
β†’ up to seven researchers on one box
β†’ plugs into a normal wall socket. No server room, no cooling project.

The paperwork is simpler too: one lab, one end user. It arrives in about 2 weeks, so it's on campus before the January term.

We supply the ASUS ExpertCenter Pro ET900N G3, new with a 3-year ASUS warranty.

PIs, lab heads and research IT directors: DM us or email contact@blade-ai.co.uk for the catalogue.

#HigherEducation #AIResearch #DGXStation #GB300 #ResearchComputing
Image from the Blade AI LinkedIn post of 29 October 2026
in
Your AI lab doesn't have to wait 30 weeks for a server.

If you run a university lab, a research group or campus IT, you know the maths: a full NVIDIA GPU server is a capital bid, a 30-week lead time and a power-and-cooling project. Meanwhile the students are here now, the grant clock is running, and the shared cluster has a queue.

The fastest route to frontier AI on campus is a workstation.

NVIDIA's DGX Station GB300 platform puts a Grace Blackwell Ultra Superchip and 748 GB of memory in a tower that runs off a wall socket. Models up to one trillion parameters, locally. Same software stack as a GB300 NVL72 rack, so what a PhD student builds on the desk moves to the cluster later without a rewrite.

This is already how the best labs work:

β†’ Stanford prototypes training and evaluation pipelines on a deskside unit, then scales the ones that work to the cluster
β†’ Harvard's Kempner Institute iterates on epilepsy genetics models without waiting for cluster time
β†’ NYU runs radiology AI locally so patient data never leaves the building
β†’ Mississippi State teaches AI engineering hands-on on deskside systems
β†’ The Naval Postgraduate School just commissioned a DGX GB300 for 1,500 students and 600 faculty, the same chip family in a rack

One desk-side GB300 unit gives up to seven researchers their own slice, no queue, no cloud bill, data on campus. Two or three of them and a department is training and serving large models this term, not next year.

We've added the ASUS ExpertCenter Pro ET900N G3, built on DGX Station GB300, to the Blade AI range. New, factory sealed, 3-year ASUS warranty, first units in around two weeks. End-user and export checks done up front, so delivery is quick.

If you're a lab head, PI or IT director planning next term's AI capacity, message me or email contact@blade-ai.co.uk and I'll send the catalogue.

#HigherEducation #AIResearch #NVIDIA #GB300 #DGXStation #AIWorkstation #ResearchComputing #BladeAI
Image from the Blade AI LinkedIn post of 26 October 2026
in
Don't let NVIDIA lead times spook you. πŸŽƒ

This week a cluster builder sent us the networking section of their BOM for an 8-node HGX B300 build. The market was telling them 38 to 52 weeks on the fabric. Some lines over a year.

Our answer came back in 24 hours: 7 to 10 working days from PO. Brand new, factory sealed, NVIDIA original, full NVIDIA warranty.

What's in the bucket:

🍬 NVIDIA Spectrum-3 SN4700 400GbE switches (MSN4700-WS2FC)
🍬 NVIDIA SN2201 1GbE out-of-band switches (920-9N110-00F1-0C0)
🍬 NVIDIA ConnectX-7 2 x 200G adapters (MCX755106AS-HEAT)
🍬 NVIDIA 800G OSFP transceivers (MMS4A20-XM800)
🍬 NVIDIA 1.6T twin-port OSFP transceivers (MMS4A00-XM)

No tricks. Just the part numbers your cluster is waiting on, in days rather than quarters.

It is allocated stock, so it moves fast. If your build is stuck on networking, send us the part numbers or the networking section of your bill of materials. We come back within 24 hours with price and lead time.

πŸ“§ contact@blade-ai.co.uk
πŸ”— https://lnkd.in/eMDcRN94

#AIInfrastructure #DataCentre #Networking #NVIDIA #B300 #SupplyChain #Halloween
Image 1 from the Blade AI LinkedIn postImage 2 from the Blade AI LinkedIn post
in
πŸ₯Š ROUND 2: air vs liquid.

Same eight NVIDIA B300 GPUs. Two very different boxes.

In the red corner: GIGABYTE G894. 8U, air-cooled, no plumbing.

In the blue corner: GIGABYTE G4L4. 4U, direct liquid-cooled.

The scorecards πŸ‘‡

🟰 GPUs: 8 Γ— B300 on both (draw)
πŸ† Rack space: 8U vs 4U (liquid wins)
πŸ† Site needs: no plumbing vs a liquid loop (air wins)

The judges' verdict: it comes down to your building. If the room is ready for liquid, the same GPUs take half the rack space. If it is not, air gets you live without touching the facility.

Which corner are you in?

πŸ‘‰ Comment "AIR" or "LIQUID", send us a DM, or get a quote: https://lnkd.in/eTWkJWrc

#NVIDIA #Blackwell #HGXB300 #GIGABYTE #LiquidCooling #GPUServers
Image from the Blade AI LinkedIn post of 23 October 2026
in
🚨 2 x CANCELLED ORDERS!! NVIDIA HGX B200. READY NOW!!

170 Supermicro HGX B200 systems have been released from cancelled orders. Finished, tested and available as soon as compliance is cleared.

β–ͺ️ 120 Γ— Supermicro SYS-A22GA-NBRT (10U)
β€’ 8 Γ— NVIDIA B200 180GB HBM3e (HGX B200 8-GPU)
β€’ 2 Γ— Intel Xeon 6960P (72-core)
β€’ 3TB DDR5-6400 (24 Γ— 128GB)
β€’ 8 Γ— NVIDIA ConnectX-7 (2 Γ— 200G)
β€’ 2 Γ— 960GB NVMe M.2
β€’ 3-year parts and labour, 3-year onsite next business day

β–ͺ️ 50 Γ— Supermicro SYS-822GS-NBRT (8U, air-cooled)
β€’ 8 Γ— NVIDIA B200 180GB HBM3e
β€’ 2 Γ— Intel Xeon 6740P
β€’ 2TB DDR5-6400 (16 Γ— 128GB)
β€’ 3 Γ— 3.84TB NVMe

New builds are quoting months. These can be in your racks this quarter.
New, through authorised channels, fully compliance-checked.

πŸ‘‰ DM me or get a quote: https://lnkd.in/eTWkJWrc

Subject to prior sale.

#NVIDIA #HGXB200 #Blackwell #B200 #Supermicro #GPUServers #AIInfrastructure
Image from the Blade AI LinkedIn post of 21 October 2026
in
🚨 CANCELLED ORDER. 64 Γ— ASUS HGX B300, 3TB. SHIPPING MID-DECEMBER.

Built for a US data-centre project that has been delayed by a year, these systems have now been released for sale, and we have early access before they reach the open market.

ASUS XA NB3I-E12 (9U)

β–ͺ️ 8 Γ— NVIDIA B300 SXM 288GB HBM3e with NVSwitch
β–ͺ️ 2 Γ— Intel Xeon 6767P (64-core)
β–ͺ️ 3TB DDR5-6400 (32 Γ— 96GB)
β–ͺ️ 8 Γ— NVIDIA ConnectX-8 SuperNIC (800Gb/s InfiniBand or 400GbE)
β–ͺ️ 7.68TB U.2 NVMe, 10 hot-swap NVMe bays
β–ͺ️ 5+5 3200W Titanium power supplies
β–ͺ️ 3-year onsite ASUS warranty, assembled and tested

Estimated ready to ship mid-December. While new B300 builds are quoting months, this is the fastest route to Blackwell Ultra we know of.

Subject to NVIDIA compliance approval.

πŸ‘‰ DM me or get a quote: https://lnkd.in/eTWkJWrc

Subject to prior sale.

#NVIDIA #HGXB300 #BlackwellUltra #B300 #ASUS #GPUServers #AIInfrastructure
Image from the Blade AI LinkedIn post of 21 October 2026
in
Blade Supply Watch 01: 1.6T optics.

Lead times of 441 days have been reported on 1.6T modules. That is a rack of switches sitting dark for over a year.

We have a route to stock in 2 to 4 weeks:

β–ͺ️ 500 Γ— NVIDIA MMS4A00-XM, 1.6T 2xDR4 twin-port OSFP
β–ͺ️ 1,000 Γ— InnoLight 1.6T OSFP224 2xDR4, the equivalent part

New. 1310nm single mode, up to 500m.

If your switches are landing before your optics, send us the part numbers and quantities today.

πŸ‘‰ Get a quote: https://lnkd.in/eTWkJWrc

Subject to prior sale.

#BladeSupplyWatch #NVIDIA #Optics #AINetworking #DataCenter
Image from the Blade AI LinkedIn post of 12 October 2026
in
πŸ₯Š ROUND 1: 2Γ— RTX PRO 4500 vs 1Γ— RTX PRO 6000

In the red corner: the heavyweight RTX PRO 6000. 96GB of memory, 600W of attitude, and a price tag to match.

In the gold corner: a tag team of two RTX PRO 4500 Blackwell Server Edition cards.

Let's go to the scorecards πŸ‘‡

🟰 Memory bandwidth: 1,600 vs 1,597 GB/s (draw)
🟰 Slot space: 2 single-slot vs 1 double-slot (draw)
πŸ† Power: 330W vs 600W (tag team wins)
πŸ† CUDA cores: 20,992 vs 24,064 (6000 wins)
πŸ† VRAM: 64GB (2Γ—32GB) vs 96GB (6000 wins)
πŸ† PRICE is the clear Knockout Blow!!! πŸ₯Š (you get 3x RTX 4500 for the price of 1x RTX 6000 on todays market)

The judges' verdict: the tag team lands close to 6000-class compute on about half the power and a far better price on todays market. And it's in stock today!!

Fair-fight note: it's two 32GB pools, not one 96GB pool. Perfect for inference and workloads that scale across GPUs. If your model needs one huge memory space, the 6000 keeps its belt.

πŸ“¦ In stock now
NVIDIA RTX PRO 4500 Blackwell Server Edition (passive PCIe, 32GB)
P/N 900-2G147-0000-000
1,000+ units, ready to ship
$5,500 per unit

Ready to rumble?
πŸ‘‰ Comment "FIGHT", send us a DM, or email contact@blade-ai.co.uk for a quote.

#NVIDIA #RTXPRO #Blackwell #GPU #AIInfrastructure #DataCenter #AIInference #HPC #RTXPRO6000 #AI #hardware
Image from the Blade AI LinkedIn post of 10 October 2026
in
AI networking is running short.

Most AI build plans still treat the network as the easy part. This month it is not.

What we are hearing from the supply chain:

- Network cards: ConnectX-6 and ConnectX-7 at 3 to 6 months, with a ConnectX-7 200G and 400G shortage expected in early 2027.
- Switches: discount pricing is being withdrawn, and buyers who lose it are paying around 10% more.
- Optics: NVIDIA transceivers at 38 to 52+ weeks, with 441 days reported on 1.6T.

Swipe through for the nine part numbers flagged very high risk, and three moves to make this month.

Planning a cluster? Send us your part numbers or the networking section of your bill of materials. We come back with price and lead time within 24 hours.

πŸ‘‰ contact@blade-ai.co.uk
πŸ‘‰ https://lnkd.in/eMDcRN94

#AIInfrastructure #DataCentre #Networking #NVIDIA #GPU #SupplyChain
Page 1 of the document in the Blade AI LinkedIn postPage 2 of the document in the Blade AI LinkedIn postPage 3 of the document in the Blade AI LinkedIn post See every page on LinkedIn
in
🚨 COMING SOON: used NVIDIA HGX H100 servers.

A lot of 70+ complete 8-GPU H100 systems is coming out of a data centre and heading to US warehouse stock shortly. We are opening the list before they land.

β–ͺ️ 8Γ— NVIDIA H100 (HGX), 8U air-cooled
β–ͺ️ 2Γ— Intel Xeon Platinum 8558, 2TB DDR5-5600
β–ͺ️ 8Γ— 400GbE ConnectX-7, plus dual-port 200GbE
β–ͺ️ Complete systems with redundant 3000W Titanium PSUs and rail kits

Proven H100 capacity, without the allocation queue.

First come, first served. These will move quickly, so get your name down now.

πŸ‘‰ Register interest: contact@blade-ai.co.uk
πŸ‘‰ Visit us www.blade-ai.co.uk

Subject to prior sale, export compliance and end-user checks.

#NVIDIA #H100 #GPUServers #AIInfrastructure #DataCentre
Image from the Blade AI LinkedIn post of 7 October 2026
in
Building out a GPU fabric? Don't let the NICs hold you up.

160 Γ— NVIDIA ConnectX-7 400G OSFP adapters (MCX75210AAS-NEAT). New pulls, tested, in the USA and ready to ship.

β–ͺ️ 400Gb/s InfiniBand NDR or 400GbE
β–ͺ️ Single-port OSFP, PCIe 5.0 x16
β–ͺ️ The standard NIC for HGX H100, H200 and B200 nodes

πŸ‘‰ Enquire now: https://lnkd.in/eTWkJWrc

Subject to prior sale.

#NVIDIA #ConnectX7 #InfiniBand #AINetworking #DataCenter
Image from the Blade AI LinkedIn post of 6 October 2026
in
🚨 CANCELLED ORDER. IN STOCK. READY TO SHIP.

NVIDIA RTX PRO 6000 Blackwell GPU servers, available immediately.

A customer order was cancelled. Their loss is your head start:

βœ… No allocation queue
βœ… No lead time
βœ… Intel Xeon 6 platform, air-cooled

Stock is limited and it is first come, first served. Once these are gone, it is back to the queue.

Need AI capacity this quarter, not next year? Message us today or email contact@blade-ai.co.uk

All sales subject to export compliance and end-user checks.

#NVIDIA #RTXPRO6000 #GPUServers #AIInfrastructure #DataCentre
Image from the Blade AI LinkedIn post of 5 October 2026
in
Running an A100 fleet? One lot just came in.

20 Γ— NVIDIA A100 80GB PCIe, new. USA stock, one lot, first come first served.

β–ͺ️ 80GB HBM2e, 2 TB/s
β–ͺ️ 300W, dual-slot passive, PCIe Gen4
β–ͺ️ Matches your existing A100 nodes, no requalification

Once these 20 are gone, there's no more.

πŸ‘‰ Claim the lot: https://lnkd.in/eTWkJWrc

Subject to prior sale and NVIDIA end-user compliance.

#NVIDIA #A100 #GPU #MachineLearning #DataCenter
Image from the Blade AI LinkedIn post of 5 October 2026
in
BLADE AI have just gained exclusive access to an upcoming deployment of 32 x B300 nodes with one of the leading AI / HPC operators in the industry:

The RFS date for the cluster is December, and the location is Canada.

Server specifications are as follows:

2x Intel Xeon 6 Performance 6767P 2.4G, 64C/128T, 24GT/s, 336M Cache, Turbo, (350W)
2TB RAM DDR5-6400
2x M.2 960GB for the OS in raid1
8x 7.68TB NVMe in the RAID configuration of your choice
8x Nvidia B300 288GB, total VRAM per node 2,304GB
InfiniBand XDR800 to GPU cluster
Ethernet 2x 400Gb to VAST Storage

36 months $5.38 per GPU / H - 25% pre-paid
60 months $4.38 per GPU / H - 25% pre-paid

Room for expansion of up to 256 nodes.

If you are looking to start 2027 with significant AI compute and are ready to commit immediately feel free to get in touch.

#nvidia #b300 #HPC #AI #baremetal
Image from the Blade AI LinkedIn post of 3 October 2026
in
Inference capacity without an HGX budget.

1,000+ NVIDIA RTX PRO 6000 Blackwell Server Edition cards, new with a 3-year manufacturer warranty. USA stock, with part shipments within weeks.

β–ͺ️ 96GB GDDR7 with ECC
β–ͺ️ Blackwell, passive dual-slot, PCIe Gen5
β–ͺ️ 8 cards per 4U/5U server
β–ͺ️ Inference, rendering and virtual workstations

πŸ‘‰ Get a quote: https://lnkd.in/eTWkJWrc

Subject to prior sale and NVIDIA end-user compliance.

#NVIDIA #RTXPRO6000 #Blackwell #AIInference #GPU
Image from the Blade AI LinkedIn post of 2 October 2026
in
Need to repopulate or expand an H100 cluster?

We have 290 Γ— NVIDIA H100 SXM5 80GB modules, new and factory sealed, in California and shipping this week.

β–ͺ️ 80GB HBM3, 3.35 TB/s
β–ͺ️ 700W SXM5 module
β–ͺ️ Fits HGX H100 8-GPU baseboards

Ideal for spares, failed-module swaps and cluster expansion without waiting on new builds.

πŸ‘‰ Enquire now: https://lnkd.in/eTWkJWrc

Subject to prior sale and NVIDIA end-user compliance.

#NVIDIA #H100 #GPU #AIInfrastructure #HPC
Image from the Blade AI LinkedIn post of 1 October 2026
in
DDR5 RDIMM is on allocation across the channel this year. Ours is on the shelf.

15,790 units of new server memory and enterprise SSD in the USA, shipping in 3–5 working days. The highlights:

β–ͺ️ 64GB DDR5-7200 RDIMM, SK hynix, Dell approved: 1,560
β–ͺ️ 128GB and 96GB DDR5-5600 RDIMM, Samsung: 1,000 each
β–ͺ️ 96GB DDR5-6400, Supermicro-qualified for HGX B300 builds
β–ͺ️ Samsung PM9A3 Gen4 and PM9D3a Gen5 NVMe: 4,400+
β–ͺ️ Samsung PM1653 15.36TB and 30.72TB SAS

Plus Kioxia, Micron, Solidigm and Phison: 134 lines in total. The parts that finish your B200 and B300 builds.

πŸ‘‰ Ask for the full priced list: https://lnkd.in/eTWkJWrc

Subject to prior sale.

#DDR5 #EnterpriseSSD #Samsung #DataCenter #AIInfrastructure
Image from the Blade AI LinkedIn post of 30 September 2026
in
In stock: 500 Γ— NVIDIA H200 NVL 141GB PCIe.

New, with a 3-year manufacturer warranty, shipping from the USA in batches now. No six-month OEM queue.

β–ͺ️ 141GB HBM3e, 4.8 TB/s
β–ͺ️ Dual-slot passive, PCIe Gen5
β–ͺ️ NVLink bridge across 2 or 4 cards
β–ͺ️ Drops into existing H100 NVL and A100 hosts

Built for LLM inference and fine-tuning in standard PCIe servers.

πŸ‘‰ Reserve your allocation: https://lnkd.in/eTWkJWrc

Subject to prior sale and NVIDIA end-user compliance.

#NVIDIA #H200 #LLM #AIInfrastructure #DataCenter
Image from the Blade AI LinkedIn post of 30 September 2026
in
New-build GPU lead times are still counted in quarters. Stock in hand is counted in days.

We pulled together what's really happening in AI hardware supply going into Q4:

β–ͺ️ Data-centre GPUs: 30–52 weeks outside priority allocation
β–ͺ️ Server memory: 23% of DRAM wafers now go to HBM, so DDR5 is on allocation
β–ͺ️ HBM: 2026 capacity is booked at the major suppliers
β–ͺ️ Packaging: CoWoS is easing, but real relief looks like 2027

Swipe through for what it means for your next build, and what's shipping now. πŸ‘‰

What's holding up your projects most right now: GPUs, memory, networking or power? Tell us in the comments.

Comment "LIST" or send us a DM and we'll share this week's full stock list.

#AIInfrastructure #NVIDIA #DataCenter #GPU #SupplyChain
Page 1 of the document in the Blade AI LinkedIn postPage 2 of the document in the Blade AI LinkedIn postPage 3 of the document in the Blade AI LinkedIn post See every page on LinkedIn
in
Limited stock: NVIDIA HGX B300 8-GPU servers, available now.

A cancelled order has freed up a small number of new, factory-sealed Supermicro 8U B300 systems. They are in Europe and can ship in about a week. Most B300 orders are quoting months.

β–ͺ️ 8Γ— NVIDIA B300, 288GB HBM3e per GPU
β–ͺ️ 2Γ— Intel Xeon 6768P, 2TB DDR5-6400
β–ͺ️ 800G ConnectX-8 networking
β–ͺ️ 3-year warranty registered to the end user

When they're gone, they're gone.

πŸ‘‰ Enquire now: https://lnkd.in/eTWkJWrc

Subject to prior sale and NVIDIA end-user compliance.

#NVIDIA #B300 #Blackwell #AIInfrastructure #DataCenter
Image from the Blade AI LinkedIn post of 28 September 2026
Email Get a quote