The common sense that you have to connect to the cloud to run AI is changing. Satya Nadella and Jensen Huang took the stage together at Microsoft's Windows and Surface event in San Francisco on October 7, 2026.
At that event, pre-orders, prices, and release dates were announced for a Surface laptop built around Nvidia's new chip, RTX Spark.
Today let's go through, step by step, how much this laptop costs, what is different about it, and what story is attached to the stocks.
Starting With the Price
✦ Surface Prices and Dates
US Microsoft Store
Surface Laptop Ultra starts at $2,599.99 at the Microsoft Store in the US. This base configuration has 24GB of unified memory and a 512GB SSD. Pre-orders began on October 7, and you can get it from October 16.
The number in the title is based on the top configuration. There are reports that the top configuration, with 128GB of memory, is $5,899.99, and assuming $1 is 1,400 won, that's about 8.26 million won before tax.
Taxes and the Korean release price have not been confirmed yet.
The Surface RTX Spark Dev Box, whose price and release schedule came out together, is $5,999 and will be sold in November through Microsoft's website, only in the US. Surface isn't the only brand with this chip in a laptop.
According to Nvidia's announcement, Acer, ASUS, Dell, HP, Lenovo, MSI, Gigabyte, and Microsoft plan to release RTX Spark systems.
What Makes RTX Spark Different?
✦ RTX Spark at a Glance
Per Nvidia's blog
RTX Spark is a structure that ties Nvidia's Blackwell GPU (up to 6,144 cores) and Grace CPU (up to 20 cores) into one chip and connects them at 600GB per second. On top of that, the CPU and GPU share up to 128GB of unified memory.
It is said to deliver up to 1 petaflop of AI performance on a theoretical FP4 basis using sparsity, and Microsoft explained that it can run AI models with more than 120 billion parameters locally.
Put these numbers side by side and it means running big models inside your own laptop without sending anything to the cloud. You don't pay token costs every time, and you don't have to send your data out.
That's why developers who handle company code and security teams are named as the first customers. From a consumer's point of view too, the most you could use before was about the 32GB of VRAM on an RTX 5090, and that limit has gone up by a lot.
✦ The Performance Claims
Versus a 16-inch MacBook Pro with M5 Pro, 64GB
Performance comparisons came out too. Microsoft compared Windows PCs with RTX Spark, in pre-production form, against a 16-inch MacBook Pro with an M5 Pro and 64GB of memory. The time until the first token was up to 2.1 times faster, they say.
It also put out figures of up to 4.3 times faster image generation and up to 6.2 times faster video generation. All three figures are results measured on pre-production devices in September 2026.
The 2.1x for the first token comes from a test commissioned by Microsoft, using the Qwen3.5 27B model with a fixed 8,192-token prompt, and the 4.3x for images and 6.2x for video are values tested by Nvidia.
The number that matters most in local AI, decode tokens per second, was not disclosed by either company.
MXC Takes Charge of Security
✦ MXC in Brief
Generally available on Windows 11, October 7, 2026
It's not only the hardware that changed. On the same day, Microsoft officially released MXC (Microsoft Execution Containers) on Windows 11.
It's a mechanism that isolates the code AI agents execute, sets by policy which files and networks an agent can access, and enforces that policy at runtime. It is designed to apply policy independently of the agent and limit abuse of permissions.
However, limitations tied to policy settings and isolation methods are stated in the published implementation. (Honestly, when I think of Microsoft's security tech... I worry it'll end up bothering users by popping up warnings every time they use AI.
They'll probably handle it well, though.)
Isolation methods run from process and session isolation to WSLc, and MicroVM, which provides a hardware virtualization boundary, is an experimental feature. It is published on GitHub under the MIT license.
OpenAI's Codex, GitHub Copilot, OpenClaw, Replit, and LM Studio already support it, and Anthropic's Claude Code is planned to support it too.
What Happens to the AI PC Market?
It's not all rosy. Sanchit Vir Gogia of Greyhound Research said that for work like meeting summaries or email triage, Spark is unnecessary and a regular AI PC will do.
Gogia also compared handing Spark to every employee to "sending a Formula One car to fetch milk." The interpretation is that Spark will be a new premium tier between workstations and AI servers rather than replacing regular laptops.
Even so, until now, to use more than 32GB of VRAM, unless a consumer invested a fortune in dedicated equipment like DGX Spark or professional-grade equipment like RTX PRO, the RTX 5090 or a Mac was about the realistic limit, but going forward Spark looks set to become another alternative.
Market conditions are tough too. IDC cut its forecast for 2026 global PC shipments to an 11.3% decline in March, because memory shortages and supply chain disruptions continue. As memory prices rise, average PC selling prices are rising too.
Gartner sees the point at which AI PCs spread to 50% of the market being pushed back to 2028 if AI PC prices rise.
Where Do the Competitors Stand?
Let's look at the competition too. Apple's M5 Pro has a unified memory bandwidth of 307GB per second and supports up to 64GB of unified memory.
Qualcomm's Snapdragon X2 Elite has an NPU in the 80-to-85 TOPS range, and Intel's Panther Lake has a 50 TOPS NPU on the 18A process. Since memory bandwidth governs decode speed in local AI, bandwidth numbers will be the item compared most from here on.
Microsoft (Microsoft Corporation, MSFT, Nasdaq)
✦ Microsoft (MSFT) Price Trend
Daily closing prices, September 9 to October 9, 2026 (Nasdaq, 23 trading days)
High Oct 9 $535.07, Low Sep 16 $490.30
Microsoft closed at $535.07 on October 9, 2026. That's up 2.38% from the previous day and about 8.8% above the September 9 close of $491.65. On October 8 it fell 1.35% on the report about OpenAI's annualized revenue, but it rebounded a day later.
On October 9, the major indexes rose together, in a rebound after the previous day's selloff in AI-related stocks. News of Windows features that run AI agents locally and the Nvidia-based Surface laptop drew investors' attention.
✦ Microsoft Price Targets
Against the October 9 close of $535.07; the dark vertical line marks the close
The Street's average target price is about $588 per Yahoo Finance, about 9.9% above the close, with 42 buy, 4 hold, and 2 sell opinions.
The highest target confirmed by institution is Arete Research's $870 (issued before May 6), 62.6% above the close, and the lowest is Rothschild & Co Redburn's $440 (September 21, raised from $400), 17.8% below.
Among recent targets, Wells Fargo set $725 on October 1 (overweight), Melius Research upgraded from 'hold' to 'buy' on October 5 and raised its target from $465 to $665, and Morgan Stanley kept a buy rating with a $600 target.
Fiscal 2027 first-quarter results are scheduled after the close on October 28.
NVIDIA (NVIDIA Corporation, NVDA, Nasdaq)
✦ NVIDIA (NVDA) Price Trend
Daily closing prices, September 9 to October 9, 2026 (Nasdaq, 23 trading days)
High Oct 6 $239.24, Low Sep 14 $210.96
NVIDIA closed at $229.28 on October 9, down 0.52% from the previous day and about 2.5% above the September 9 close of $223.67. On October 6 it hit a record close of $239.24 and reached $243.37 intraday.
On October 8, amid a drop in AI-related stocks over reports that OpenAI's annualized revenue is close to $50 billion and may be lower than expected, NVIDIA fell 2.94% too.
On September 28 the board raised the buyback authorization by $150 billion, leaving $235 billion of authorization.
✦ NVIDIA Price Targets
Against the October 9 close of $229.28; the dark vertical line marks the close
The Street's average target price is between about $321.89 and $328.72, which is 40.4% to 43.4% above the close. The highest target confirmed by institution is Raymond James's $515 (August 27, strong buy), 124.6% above the close.
The lowest confirmed by institution is Wells Fargo's $210 (October 2, newly issued), 8.4% below the close. The lowest in the whole aggregate is $180, 21.5% below the close, and the highest is $515.
Morgan Stanley cut its target from $300 to $230 on October 2, and Citigroup raised its target from $300 to $315 on August 27. It stands out that there are institutions with targets below the close.
To sum up, it's a declaration that Nvidia on the chip and Microsoft on Windows and the security layer will bring AI from the cloud to the top of your desk.
However, judging by the price and market conditions, it will be hard for this to become a laptop everyone uses right away, and it's likely to become a premium tier that developers and security-conscious companies use first.
From here, real performance reviews after the October 16 release and Microsoft's October 28 earnings are the most important points to check.
Anyway, if you're a consumer who's into local AI, you'll surely be interested in this laptop. The frustration of not being able to run the model you want because there isn't enough VRAM is something only people who've been through it know.
This product is going to solve that.
This is not investment advice. You know that, right? Let's protect our precious money ourselves.
Well then, shall we begin today's adventure?
✦ Related Posts
No comments:
Post a Comment